Departamento de Análisis Económico:
Economía Cuantitativa
Three Essays on Commodity Prices
Doctoral Thesis
Lya Paola Sierra Suárez
2014
1
Departamento de Análisis Económico: Economía Cuantitativa
TESIS DOCTORAL
Three Essays on Commodity Prices
DOCTORANDO: LYA PAOLA SIERRA SUÁREZ
DIRECTORES: DRA. DÑA. PILAR PONCELA BLANCO.
DRA. DÑA. EVA SENRA DIAZ
2
Dedicatoria
Hace cuatro años emprendí un camino de aprendizaje en todos los sentidos de mi vida,
este documento sólo representa el resultado de una parte de ellos. Es por esto que
quiero dedicar esta tesis a Fernando, mi marido, por emprender este camino conmigo,
por su paciencia, y por cuidar bien a nuestros bebés durante mi ausencia. No hubiera
podido culminar este ciclo de aprendizaje sin su continuo apoyo y amor. A mi madre,
María Eugenia, y a mi suegra, Graciela, que apartaron meses de sus vidas para
acompañar a mi familia en momentos difíciles, gracias.
También quiero dedicar esta tesis a mis padres, que son mi modelo a seguir y mis
primeros y más importantes maestros. He vivido cuatro años de continuo aprendizaje, a
nivel personal, profesional, y académico. No hubiera podido culminar esta fase de mi
vida sin la humildad necesaria, sin la apertura mental, y sin perseverancia, valores que
atribuyo a mis ellos.
En cuatro años se hicieron presentes la vida y la muerte. Por lo tanto, por último, quiero
dedicar esta tesis a la memoria de Angélica Rugeles, mi abuela; a Sofía, mi hija, que
iluminó mi vida con su nacimiento; y a Miguel Santiago, mi hijo, quien es mi ilusión y
que soportó nuestras mudanzas.
3
Agradecimientos
En primer lugar, quiero expresar mi sincero agradecimiento a mi directora de Tesis,
Pilar Poncela, por su continuo apoyo durante los últimos tres años. Sus enseñanzas y
consejos han sido mi guía desde que era su alumna en el Máster de Economía
Internacional. Durante la elaboración de mi tesis doctoral, la profesora Pilar Poncela
siempre estuvo presente y dispuesta a orientar mi trabajo. No puedo imaginar mejor
directora de tesis.
También tengo sentimientos de gratitud hacia mi directora de tesis, la profesora Eva
Senra, que fue un soporte invaluable para el desarrollo de la tesis doctoral. Gracias por
sus consejos y correcciones que siempre logran mejorar mis trabajos. A las dos, gracias
por su cariño y paciencia, les estaré eternamente agradecida.
Mi más sincero agradecimiento al profesor Roger Simpson, quien me ha ayudado a
mejorar mi escritura en inglés durante el último año.
Por otro lado, esta tesis no podría haberse llevado a cabo sin el apoyo financiero del
gobierno de Colombia, mediante el programa Francisco José de Caldas de
COLCIENCIAS, y de la Pontificia Universidad Javeriana de Cali, a quienes agradezco
inmensamente.
Finalmente, gracias de nuevo a mi familia por su apoyo incondicional. Son el motor de
mi vida.
4
5
Content
Dedicatoria........................................................................................................................ 3
Agradecimientos ............................................................................................................... 4
Introducción ...................................................................................................................... 9
Trabajos de investigación de la tesis........................................................................... 11
1. Introduction ................................................................................................................ 21
1.1.
Overview .......................................................................................................... 21
1.2. Essays of the Thesis ............................................................................................. 23
References ...................................................................................................................... 30
2. Common dynamics of non-energy commodity prices and its relation to uncertainty. 31
2.1.
Introduction ...................................................................................................... 31
2.2.
Related literature .............................................................................................. 35
2.3.
Methodology .................................................................................................... 39
2.3.1.
The FAVAR model .................................................................................. 39
2.3.2.
Data definition .......................................................................................... 41
2.4.
Empirical Results ............................................................................................. 43
2.4.1.
Dynamic factor model .............................................................................. 43
2.4.2.
The FAVAR model results ....................................................................... 47
2.4.3.
Robustness checks .................................................................................... 50
2.4.4.
Variance decompositions.......................................................................... 54
2.5.
Conclusions ...................................................................................................... 56
References ...................................................................................................................... 58
Appendix A: Non fuel commodities.............................................................................. 62
Appendix B: Macroeconomic and Financial Variables................................................. 63
6
Appendix C: Graphs of the macroeconomic and financial variables. ........................... 65
Appendix D: Residual Autocorrelation analysis of the FAVAR model. Lagrange
Multiplier (LM) Test. ..................................................................................................... 66
Appendix E: Robustness checks: different proxies for uncertainty. .............................. 67
Appendix F: Robustness checks: FAVAR model with two non-fuel commodity factors.
........................................................................................................................................ 69
3. The predictive content of co-movement in non-energy commodity price changes ... 70
3.1.
Introduction ...................................................................................................... 70
3.2.
Model specifications ........................................................................................ 72
3.3.
Data description and empirical strategy. .......................................................... 76
3.4.
Empirical results .............................................................................................. 80
3.5.
Robustness checks ........................................................................................... 83
3.6.
Conclusions ...................................................................................................... 86
References ...................................................................................................................... 88
Appendix A. ................................................................................................................... 90
Appendix B. .................................................................................................................... 91
4. Long-term links between raw materials prices, real exchange rate and relative deindustrialization in a commodity dependent economy. Empirical evidence of “Dutch
disease” in Colombia ...................................................................................................... 93
4.1.
Introduction ...................................................................................................... 93
4.2.
Dutch disease and its symptoms in Colombia ................................................. 97
4.3.
Econometric modeling and empirical evidence of Dutch disease in Colombia
107
4.3.1.
Order of integration ................................................................................ 111
4.3.2.
Johansen analysis .................................................................................... 112
4.3.3.
Results analysis ...................................................................................... 116
4.4.
Conclusions and policy recommendations .................................................... 121
References .................................................................................................................... 124
Appendix A: Summary of Dutch disease literature of Colombia ................................ 129
Appendix B: Commodities production and crude oil discoveries in Colombia. ......... 130
Appendix C: Foreign Direct Investment in the petroleum sector. Colombia ............... 132
7
Appendix D: Data sources ............................................................................................ 133
Appendix E: Graphs ..................................................................................................... 134
Appendix F: VECM system short-run coefficients of the rest of the variables of the
model. ........................................................................................................................... 135
Appendix G: Estimation results when RER is replaced by COMMO –equation (4.3) 136
5. Conclusions .............................................................................................................. 139
5.1.
Overall Conclusion ........................................................................................ 139
5.2.
Future development ....................................................................................... 141
5.3.
Dissemination of results................................................................................. 143
Conclusiones................................................................................................................. 146
Conclusiones Generales ............................................................................................ 146
Desarrollos futuros .................................................................................................... 149
Difusión de Resultados ............................................................................................. 151
8
Introducción
La presente tesis está compuesta por el desarrollo de tres trabajos de investigación
cuyos resultados pueden ser leídos de manera independiente. Los tres incluyen temas
relacionados con la evolución de los precios de las materias primas: sus características y
determinantes a nivel internacional, el reciente papel de la incertidumbre en el mercado,
el poder predictivo del movimiento conjunto de los precios de las materias primas en el
tiempo y las posibles vías por las que dicha evolución afecta la economía de un país
dependiente de los productos básicos como Colombia.
La evolución de los precios de las materias primas afecta de una u otra manera la
economía de un país debido a que los bienes de uso cotidiano están relacionados con las
materias primas, desde la gasolina que da energía a los barcos y camiones que
transportan productos, hasta el cobre en el cableado eléctrico y los alimentos
consumidos. Las fluctuaciones en los precios de los productos son de especial
relevancia para las economías dependientes de materias primas, no sólo porque
representan una gran proporción de las exportaciones totales sino también porque una
gran cantidad de ingresos fiscales provienen a menudo de este sector. En este sentido
choques en los precios de las materias primas exportadas pueden generar una mayor
vulnerabilidad de estos países en términos comerciales y también pueden producir una
política fiscal pro-cíclica si el país no se encuentra suficientemente diversificado a nivel
de exportaciones y a nivel tributario.
Adicionalmente la evolución de los precios de las materias primas está directamente
relacionada con la inflación, por lo tanto, países importadores de materias primas
también pueden verse afectados por fuertes movimientos en los precios. Así, periodos
9
de auge como el acontecido entre los años 2007 y 2008, puede por un lado dificultar la
política monetaria en los países desarrollados y por otro llevar a niveles más altos de
pobreza en países pobres, debido a que gran parte de los ingresos de los hogares en
estos países se utiliza para la compra de alimentos.
En vista de la importancia de los precios de los productos básicos para muchos países en
el mundo, es relevante conocer las fuerzas que impulsan el movimiento internacional de
sus precios; la forma de mejorar la capacidad de predicción en los precios de las
materias primas; y los mecanismos por los cuales las fluctuaciones en los precios de las
materias primas puede afectar a las economías dependientes, como Colombia.
En esta introducción se presenta un breve resumen de los capítulos presentados en la
tesis, poniendo especial énfasis en la motivación y la contribución de cada uno de ellos.
El resto de la tesis se encuentra organizada de la siguiente manera: Los capítulos 2, 3 y
4 contienen los trabajos de investigación de la tesis. El capítulo 5, el cual contiene las
conclusiones, se divide en tres apartados: El primero, denominado Conclusiones
Generales redondea los principales resultados de la tesis en su conjunto; En el segundo
apartado, denominado Desarrollos Futuros se evalúan nuevas líneas de investigación
derivadas de la tesis; finalmente, en el apartado Difusión de Resultados se enumeran los
congresos y seminarios en los que han sido presentados los resultados de la
investigación, así como las publicaciones resultantes de ellos y los premios recibidos.
10
Trabajos de investigación de la tesis
En este apartado se ofrece una síntesis de los próximos capítulos de la tesis, en
particular se presenta en detalle los objetivos de la investigación y los principales
hallazgos y aportaciones en cada uno de ellos.
Capítulo 2. Dinámica conjunta en los precios de las materias primas no energéticas
y su relación con la incertidumbre.
El propósito de este capítulo consiste en mejorar la evidencia empírica sobre los precios
de los productos básicos en varios aspectos. En primer lugar, se trata de identificar la
importancia de los movimientos sincronizados de 44 series de precios de productos
básicos con frecuencia mensual. El objetivo es determinar si la evolución conjunta de
los precios de diferentes materias primas no energéticas ha incrementado desde finales
del año 2003 como consecuencia de la llamada financialización de los mercados de
materias primas. La financialización en el mercado de materias primas hace referencia a
dos fenómenos relacionados: primero, el aumento desmesurado de inversiones
destinadas a los índices de materias primas y, segundo, al incremento en el número de
operadores financieros especulativos en este mercado a partir de finales del año 2003.
La hipótesis planteada radica en la idea que el proceso de financialización puede
transformar la determinación de los precios en el mercado de productos básicos en el
corto plazo. Por lo tanto, como resultado de este fenómeno los precios de las materias
primas en su conjunto se asemejan más, en el corto plazo, al precio de un activo
especulativo, como las acciones y divisas. De acuerdo con ello, y sin dejar de lado la
importancia de los factores macroeconómicos a nivel internacional, la financialización
11
de los mercados de productos básicos puede conducir a una creciente importancia de
variables especulativas en la determinación de precios de las materias primas no
energéticas.
El segundo objetivo del artículo consiste en evaluar tanto la importancia relativa de
variables macroeconómicas fundamentales como la de variables financieras
en la
determinación del movimiento conjunto entre los precios de materias primas no
energéticas en el corto plazo. Es importante destacar que se evalúa el papel de la
incertidumbre de mercado como un nuevo posible determinante de la evolución común
de los precios. Se utilizan datos mensuales desde febrero de 1992 hasta diciembre de
2012 y se divide la muestra en dos períodos (previo y posterior a diciembre de 2003)
con el fin de evaluar si la importancia relativa de variables financieras como el precio de
las acciones y la incertidumbre del mercado cambia en el período de financialización
(posterior a Diciembre de 2003).
En relación a la metodología, se diagnostica el movimiento conjunto en los precios
utilizando un modelo factorial dinámico estimado por Componentes Principales.
Adicionalmente, se estima un modelo que combina los resultados actuales del análisis
factorial dinámico con los modelos de vectores autoregresivos (FAVAR, por sus siglas
en inglés) propuesto por Bernanke, Boivin y Eliasz (2005). La estimación del modelo
FAVAR tiene como objetivo medir las relaciones dinámicas entre las variables
macroeconómicas a nivel internacional, o fundamentos, y las variables financieras con
el movimiento conjunto en los precios de las materias primas. Se presentaron los
resultados de las funciones de impulso respuesta y la descomposición de la varianza del
factor no energético del modelo FAVAR. Los resultados del primero permiten rastrear
los efectos en el tiempo en los precios de las materias primas no energéticas ante
choques exógenos de las otras variables. Los resultados de la descomposición de
12
varianza muestran la importancia relativa de cada una de las variables en la
determinación del precio conjunto de las materias primas no energéticas.
Las conclusiones de este segundo capítulo se resumen a continuación. En primer lugar,
se encontró una mayor sincronización entre los precios de las materias primas a partir de
diciembre de 2003, como se sugiere en la hipótesis de financialización. Más
precisamente, la varianza en los precios de las materias primas explicada por el
comportamiento común de los precios de los productos básicos se incrementó de 9 %
entre febrero de 1992 y noviembre de 2003, a 23 % entre diciembre de 2003 y
diciembre de 2012. Esto significa que después de 2004 el comportamiento común de los
precios de las materias primas tiene un peso mayor en las fluctuaciones de los precios
en comparación con el período anterior a la financialización, es decir, antes de 2004.
En segundo lugar, el análisis impulso-respuesta en el modelo FAVAR indica que los
choques en el índice bursátil así como en la incertidumbre de mercado afectan
significativamente los precios de los productos básicos no energéticos sólo después de
finales de 2003. Un choque positivo en el mercado de valores es seguido por un
incremento en los precios de las materias primas no energéticas, lo cual muestra un
mayor vínculo entre los mercados para el segundo sub-período. En relación con los
efectos de la incertidumbre, se encontró que un aumento en esta variable conduce a una
disminución de los precios de las materias primas no energéticas, lo que significa que
los inversores reaccionan vendiendo activos en materias primas ante aumentos en la
incertidumbre.
En tercer lugar, el análisis de descomposición de la varianza en el modelo FAVAR
muestra que la incertidumbre juega un papel relevante en la explicación de la
fluctuación no energética sólo en el segundo periodo. En particular, el porcentaje de la
13
varianza del error de predicción del factor no energético a un horizonte de 12 meses que
es atribuible a un choque en la incertidumbre, pasó de menos del 5% en el primer
periodo, a 17,4% en el segundo. De hecho, a partir de esta fecha los choques de
incertidumbre representan una mayor proporción en la descomposición de varianza de
los precios de las materias primas no energéticas que otras variables macroeconómicas
que previamente tuvieron un mayor peso, tales como la tasa de cambio real y la
demanda mundial. Lo anterior pone de relieve la creciente importancia de la
incertidumbre como un determinante de las comunalidades en las materias primas no
energéticas.
La contribución más importante de este capítulo a la literatura radica en la
determinación del papel de la incertidumbre en la formación de los precios de las
materias primas no energéticas en un contexto de corto plazo. Existen pocos estudios
que analizan este tema en general, y ninguno que evalúe el papel de la incertidumbre en
un contexto de corto plazo y especulativo, por lo tanto, el artículo llena este vacío. Por
otra parte, teniendo en cuenta que se divide el período muestral para tener en cuenta la
fecha de inicio de la financialización, el estudio comprueba que el movimiento conjunto
en los precios de las materias primas no energéticas ha sido recientemente más afectado
por variables especulativas. Otras aportaciones de la tesis a la literatura de movimiento
conjunto en los precios de las materias primas no energéticas son: el uso de datos de
más alta frecuencia que los utilizados anteriormente en la literatura en el modelo
FAVAR y el uso de una variable que no ha sido utilizada previamente en el estudio de
la incertidumbre en los precios de los productos básicos: el Chicago Board Options
Exchange Volatility Index (VIX), la cual tiene en cuenta la volatilidad esperada en el
mercado futuro. Se prueba la robustez de los resultados utilizando otras medidas de
incertidumbre como el índice de incertidumbre de Política Económica y el Índice
14
bursátil de Incertidumbre construido por Baker, Bloom y Davis (2013) que tampoco han
sido utilizados en el pasado, según refiere la literatura del tema.
Capítulo 3. Análisis del contenido predictivo del movimiento conjunto en la
inflación de los precios de las materias primas no energéticas.
Este capítulo se deriva de los hallazgos encontrados en el capítulo anterior.
Específicamente existen dos resultados que motivaron esta investigación. En primer
lugar, se encontró que el movimiento conjunto entre diferentes precios no energéticos
aumenta a partir de finales del año 2003. Adicionalmente, en este periodo existe una
gran fracción del movimiento de los precios de los productos básicos que no es
atribuible a variables macroeconómicas fundamentales. Por lo tanto, un objetivo de este
artículo es explorar si el movimiento conjunto entre los precios de materias primas no
energéticas, tiene algún contenido predictivo sobre la inflación de los precios de las
materias primas.
Por otra parte, el artículo también tiene como objetivo evaluar qué grado de
comunalidad tiene un mejor desempeño a nivel de predicción: un movimiento conjunto
de gran escala, que comprende la evolución común de una amplia gama de precios de
las materias primas no energéticas, o un movimiento conjunto de pequeña escala, el
cual contiene el patrón común de los precios de las materias primas dentro de la misma
categoría. Adicionalmente, el artículo explora si la predictabilidad de los precios de los
productos varía a través de las diferentes categorías.
Se utiliza la base de datos del Fondo Monetario Internacional que comprende los
precios mensuales de 44 materias primas no energéticas divididas en las categorías:
cereales, carnes y mariscos, bebidas, aceite vegetal y harinas proteicas, materias primas
15
agrícolas y metales. Metodológicamente se utiliza un Modelo Factorial Dinámico
(DFM, por sus siglas en inglés) para extraer el movimiento conjunto de la inflación en
los precios de las materias primas no energéticas. Se utilizan dos métodos de estimación
dentro del enfoque DFM: componentes principales para extraer la evolución común de
todo el conjunto de datos de precios de productos básicos, y el filtro de Kalman para
extraer los factores latentes dentro de las categorías. La idea básica consiste en tomar el
factor extraído, estimado ya sea por componentes principales o por el filtro de Kalman,
y utilizarlo para predecir la inflación de las materias primas no energéticas.
Adicionalmente, se estima un modelo autoregresivo (AR, por sus siglas en inglés), y un
modelo de caminata aleatoria (RW, por sus siglas en inglés), este último utilizado como
referencia. Dentro del procedimiento econométrico se contempla la comparación de los
diferentes modelos mediante la evaluación de los pronósticos fuera de muestra. Esta
comparación se lleva a cabo mediante el ratio de la raíz cuadrada del error cuadrático
medio (RMSE, por sus siglas en inglés) de predicción entre los pronósticos de los
modelos estimados un periodo hacia adelante y aquellos generados por una caminata
aleatoria. Se toman datos desde enero de 2004 hasta diciembre de 2013, siendo 2011:01
- 2013:12 el período de evaluación.
Las principales conclusiones del artículo se presentan a continuación. En términos
generales se encontraron resultados prometedores en los modelos factoriales dinámicos
estimados por medio del filtro de Kalman, los cuales tienen en cuenta el movimiento
conjunto por categoría de materias primas. Específicamente, los resultados muestran
una capacidad predictiva superior frente a los pronósticos generados por la caminata
aleatoria en las categorías metales, aceites vegetales y harinas proteicas. Este resultado,
a su vez, contrasta con la baja capacidad predictiva encontrada en el modelo factorial
dinámico, estimado por medio de componentes principales, el cual toma en cuenta la
16
evolución conjunta en la totalidad de las series de precios de las materias primas.
Además, en el capítulo se sugiere que el modelo autoregresivo es más preciso que la
caminata aleatoria en la mayoría de los precios de materias primas no energéticas en un
horizonte mensual.
La principal contribución de este capítulo a la literatura tiene que ver con el aporte de
evidencia empírica del grado de comunalidad que es útil para fines de predicción. Se
demuestra que la evolución conjunta de los precios de las materias primas con
características similares (como el existente dentro de cada categoría) tiene poder de
predicción sobre los precios de las materias primas no energéticas. Por el contrario, el
contenido predictivo del movimiento conjunto
existente en largas cantidades de
materias primas resulta muy bajo. La contribución es importante no sólo porque se
reporta el éxito relativo del uso de modelos factoriales en la predicción de la inflación
de las materias primas no energética en horizontes cortos, sino también porque se ha
añadido al debate académico el tema de la capacidad de pronóstico en los cambios de
los precios de las materias primas
Capítulo 4. Relaciones de largo plazo entre los precios de las materias primas, el
tipo de cambio real y la desindustrialización relativa en una economía dependiente
de materias primas. Evidencia empírica de “Enfermedad Holandesa” en
Colombia.
Este capítulo refleja un cambio en la dirección del análisis de la evolución de los precios
de las materias primas en comparación con los estudios previos, los cuales evalúan
principalmente la sincronización de los precios de las materias primas no energéticas,
los factores que los afectan y la capacidad predictiva de dicho movimiento conjunto . El
capítulo, por el contrario, se mueve hacia el análisis del impacto de los precios de los
17
productos básicos en las variables macroeconómicas de una economía pequeña y
dependiente de estos, como la de Colombia. Acorde a lo comentado al inicio de la
introducción, las fluctuaciones en los precios de las materias primas son cruciales en
varios aspectos de la economía. Por lo tanto, una perspectiva global de la evolución de
los precios de las materias primas debería tener en cuenta no sólo los factores que
determinan los precios de las materias primas a nivel internacional, sino también el
impacto que éstos pueden tener en una economía exportadora de materias primas. De
esta manera, el artículo se introduce en la literatura que evalúa los posibles efectos
negativos de los periodos de auge en los precios de las materias primas sobre las
economías exportadoras de estos bienes. En particular, el artículo se centra en evaluar
posibles síntomas del fenómeno conocido como la Enfermedad Holandesa en Colombia.
El término Enfermedad Holandesa data del año 1970 y hace referencia a los efectos
nocivos que trajo consigo el descubrimiento de grandes depósitos de gas en el Mar del
Norte sobre el sector industrial en Holanda. El repentino incremento en la riqueza de un
país exportador de materias primas puede crear una entrada de capitales sin precedentes
generando la apreciación real de la moneda holandesa y, por lo tanto, la pérdida de
competitividad a nivel internacional de los productos industriales exportados por el país.
La dependencia de Colombia en las exportaciones de materias primas y el posible efecto
sobre la desindustrialización del país causado por la apreciación del tipo de cambio
motiva esta investigación.
Colombia es un caso especial dentro de las economías dependientes de materias primas
dado que en el tiempo sus exportaciones han estado concentradas en materias primas
diferentes,
café primero
y más
recientemente el
petróleo. Por lo
tanto,
metodológicamente el estudio se realiza mediante análisis de largo plazo y evalúa a
través de un Modelo de Corrección de Error (VECM, por sus siglas en inglés) si los
18
precios de los productos básicos están relacionados con el tipo de cambio real y la
producción manufacturera relativa. También se tienen en cuenta variables como la
productividad, el gasto del gobierno, el grado de apertura comercial y la inversión
extranjera para el período comprendido entre 1972 y 2012.
Las estimaciones muestran que los precios de las materias primas se relacionan
positivamente con la tasa de cambio real en el largo plazo, sin embargo la producción
manufacturera relativa no se ve afectada por apreciaciones reales de la moneda local.
Por lo tanto, los aumentos en los precios de las materias primas tienen un efecto
negativo en la competitividad del país sin causar desindustrialización relativa. Los
resultados también muestran que el gasto público es una fuente importante de presión
sobre el tipo de cambio real en Colombia.
La contribución más importante de este artículo se encuentra en el análisis del
fenómeno de la Enfermedad Holandesa en un contexto de largo plazo. El auge en los
precios de las materias primas en los países productores crea no sólo posibles
desventajas, como la pérdida de competitividad y un mayor nivel de especialización en
la producción del recurso exportado y en los sectores no comercializables, sino también
ventajas como el aumento de la riqueza del país, mayores ingresos fiscales para
inversión social y una mejora en la balanza de pagos. Por lo tanto, los efectos netos
derivados de períodos de expansión en las materias primas exportadas por un país no se
pueden observar en un contexto de corto plazo. Hasta ahora, los estudios se han
centrado especialmente en el último auge del café (por ejemplo Puyana, 2000, Meisel
1998, Edwards, 1984 y Kamas, 1986) sin cubrir los dos episodios históricos de auge en
el sector de las materias primas exportadas: el aumento de los precios del café (1975 1977) y el aumento en los precios del petróleo (2003-2008). Por lo tanto, dado que el
capítulo tiene en cuenta una perspectiva de largo plazo, ofrece la capacidad de evaluar si
19
los efectos adversos asociados a la enfermedad holandesa compensan los efectos
beneficiosos de la expansión.
Bibliografía
Baker, S., Bloom, N. and Davis S. J. (2013) Measuring Economic Policy Uncertainty,
Chicago Booth Research Paper doi: 10.2139/ssrn.2198490, Chicago.
Bernanke, B., Boiving, J. and Eliasz, P. (2005) Measuring the effects of monetary
policy: a factor-augmented vector autoregressive (FAVAR) approach, The
Quarterly Journal of Economics, 120, 387-422.
Edwards S. (1984) Coffee Money and inflation in Colombia, World Development, 12,
1107-1117.
Meisel, A. (1998) Dutch Disease and Banana Exports in the Colombian Caribbean,
1910-1950, Cuadernos de Historia Económica y Empresarial, 26, April.
Puyana, A. (2000) Dutch Disease, Macroeconomic Policies, and Rural Poverty in
Colombia, International Journal of Politics, Culture, and Society, 14(1), 205-233.
20
Chapter 1
Introduction
1.1. Overview
This thesis relates to raw material prices, and concentrates on issues such as the global
factors that drive the developments in a large set of non-energy commodity prices; the
impact of market uncertainty in determining common evolution among commodity
prices; the predictive power of co-movement in raw material prices; and the possible
channels whereby the evolution of raw material prices impacts the economy of a
commodity-dependent country such as Colombia. These topics are presented in three
chapters, which can be read as independent essays.
The evolution of commodity prices is a very important issue since it can affect in one
way or another the economy of a country. Many of the things which are eaten and used
everyday are related to raw materials, from the gasoline that powers the ships and trucks
that transport products, to the copper in electrical wiring and the food on our plates.
Fluctuations in commodity prices are especially relevant to commodity-dependent
21
economies, not only because raw materials account for a large share of total exports,
which can lead to a greater vulnerability of these countries in terms of trade shocks, but
also from a fiscal standpoint, because a great amount of fiscal income often comes from
this sector. Hence, commodity price booms and busts can result in procyclical
government spending if the country is not sufficiently diversified in matters of taxation,
or lacks tax regulations.
In the case of commodity importers, swings of raw material
prices are directly related with inflation, which may make monetary policy more
difficult to manage during periods of boom and bust in commodity prices. Finally, for
many low development countries that import commodities, the increase in the prices of
these goods can lead to higher levels of poverty, since a larger share of their income is
used to buy food.
In view of the importance of commodity prices for most of the countries around the
world, it is relevant to know which forces are behind the international movement of raw
material prices, how to improve forecasting power in commodity prices, and how the
evolution of raw material prices can affect commodity-dependent economies such as
Colombia.
This introduction continues with a brief summary of the essays presented for this thesis,
with particular attention to the motivation and contribution of each one. The remainder
of the thesis is organized as follows: chapter 2-4 contains the essays of the thesis. The
conclusion in Chapter 5 is then divided into three parts: an overall conclusion, which
rounds off the conclusions of the thesis as a whole; future developments, which assess
further research topics derived from the thesis; and dissemination of results, which lists
the congresses and seminars where the essays have been presented, as well as the state
of publications and the awards that have received.
22
1.2. Essays of the Thesis
This chapter section offers an outline of the forthcoming individual chapters of the
thesis: in particular, it offers an in-depth summary of the research objectives and the
major findings and contributions.
Chapter 2. Common dynamics of non-energy commodity prices and its relation to
uncertainty.
The purpose of this essay is to improve the empirical evidence on commodity prices in
various dimensions. First, we attempt to identify the extent of synchronized movements
in 44 monthly non-energy commodity price series. The aim is to ascertain whether the
links among commodity prices, or co-movement, have increased since the end of 2003,
as result of the large increase of assets allocated to commodity indices by speculative
financial traders, or so called financialization of commodity markets. Our hypothesis
relies on the idea that the financialization process may transform price determination in
the commodity market in the short term. Hence, commodity prices as a whole perform
more like a speculative price, such as stock prices and currencies. Accordingly, without
disregarding the importance of classical macroeconomic factors, financialization in the
commodity markets may lead to a growing relevance of speculative variables in nonenergy price determination.
Second, we attempt to evaluate the relative importance of macroeconomic fundamentals
as well as financial variables in determining co-movement between non-energy prices in
23
the short term. Importantly, we determine the role of market uncertainty as a potential
driver of overall price movements.
We use monthly data from February 1992-
December 2012 and break the sample down into, say, pre-2004 and post-2004 periods,
in order to evaluate whether the relative importance of financial variables such as the
stock market and market uncertainty changed in the financialization period.
With regard to the methodology, we diagnose the overall co-movement using a
Dynamic Factor Model estimated by principal components. In order to assess the
relationship of fundamentals, financial and uncertainty variables with the co-movement
in commodity prices, we use a Factor-Augmented Vector Autoregressive (FAVAR)
approach, proposed by Bernanke, Boivin and Eliasz (2005). From the FAVAR model
we present the impulse response functions, since they can trace the effects over time on
a variable of an exogenous shock from another variable; and the variance decomposition
of the non-energy factor, as it shows the relative relevance of each of the variables in
determining the non-energy price co-movement.
The conclusions drawn in this first essay are summarized as follows: first and foremost,
we found a greater synchronization among raw materials since December 2003, as
suggested in the financialization hypothesis.
More precisely, the variance of
commodity prices explained by the common behavior of these prices, increased from
9% between February 1992 and November 2003, to 23% between December 2003 and
December 2012. This means that after 2004 the common behavior of commodity prices
accounts for a larger share of fluctuations than in the pre-financialization period, namely
pre-2004. Secondly, impulse response estimates in the FAVAR model indicate that
shocks in the stock market index as well as in the market uncertainty variable
significantly affect non-fuel commodity prices only after late 2003. A stock market
shock is followed by increases in price co-movement, which show the strengthening of
24
cross market linkage for the second sub-period. In relation to the effects of uncertainty,
we find that an increase in this variable leads to a decrease in non-energy commodity
prices, which means that uncertainty makes commodity investors more risk averse and
willing to sell. Thirdly, the variance decomposition analysis in the FAVAR model
shows that the uncertainty element plays a larger role in explaining non-energy
fluctuation for the second sub-period. In particular, the percentage of variance of the
forecasting error of the non-energy factor, at a 12 month horizon, which is attributable
to a shock in uncertainty, increased from less than 5% in the first sub-period, to 17.4%
in the second subperiod. In fact, after late 2003, uncertainty shocks account for a larger
share of variance decomposition of non-energy commodity prices than fundamentals,
such as real exchange rate and world demand, which emphasizes the growing
importance of uncertainty as an important determinant of communalities in non-energy
commodities.
The most important contribution of this essay to the literature has to do with the
understanding of the role of uncertainty, as well as speculative variables, such as the
stock market prices in the non-energy price determination within a short term context.
Few studies exist that analyze this topic in general, and none in a short and speculative
context. This study aims to fill this void. Moreover, by considering a sample breakdown
taking into consideration the starting date of the financialization period, we check
whether co-movement has recently been more affected by speculative variables. Other
contributions of the thesis to the literature of co-movement in non-energy commodity
prices are: the use of higher frequency data in the FAVAR model; and the use of a
previously unused variable in the study of uncertainty in commodity prices: the Chicago
Board Options Exchange Volatility Index (VIX), which takes into account the expected
volatility in the future market. We prove the robustness of our results against other
25
measures of uncertainty such as the Economic Policy Uncertainty Measure and the
Equity Uncertainty Measure constructed by Baker, Bloom and Davis (2013), neither of
which have ever been used in this type of study.
Chapter 3. The predictive content of co-movement in non-energy commodity price
changes.
This essay is a result of the findings in the previous chapter “Common dynamics of nonenergy commodity prices and its relation to uncertainty” and of the idea that commodity
prices may recently behave more as speculative prices. In the first essay we found that
the co-movement among different non-energy prices largely increased after late 2003
when compared with the previous period. Additionally, for the second sub-period, there
is a great fraction of movement in commodity prices that does not seem to be
attributable to fundamentals. The aim of this essay, therefore, is to explore whether comovement in prices of non-energy commodities has any predictive content over spot
price changes.
Moreover, we aim to assess which degree of communalities or co-movement performs
better for forecasting purposes: a large scale co-movement, which comprises the
common evolution of a wide range of non-energy commodity prices, or a small-scale
co-movement, which contains the common pattern of commodity prices within the same
category. It is also of interest to explore whether the predictability of commodity prices
varies across different types of categories.
We use the International Monetary Fund commodities database, which comprises 44
monthly commodity price series split in the categories: cereals, meat and seafood,
26
beverages, vegetable oil and protein meals, agricultural raw materials and metals.
Methodologically, we use a Dynamic Factor Model (DFM) framework to extract a
latent factor that drives the co-movement on non-energy commodity price inflation.
We use two estimation methods within the DFM approach: principal component, to
extract the common evolution of the whole commodity price data set, which we call the
large-scale factor model; and the Kalman filter to extract factors within categories,
which we call the small-scale factor model. The basic idea is to take the extracted
factor, estimated either by principal components, or by Kalman filter, and used it to
forecast commodity inflation. We also estimated a univariate autoregressive (AR)
model, and a random walk model, used as a benchmark. Our measure of forecasting
performance is the out-of-sample root mean square error of prediction (RMSE) for onestep-ahead forecasts.
The sample began in January 2004 and finished in December
2013; the forecasting evaluation period corresponds to 2011:01– 2013:12, changing the
estimation sample as needed to generate true ex-ante one step ahead forecasts.
The main conclusions of this essay may be summarized as follows.
Overall, we
obtained promising results with the small-scale factor model.
Specifically,
predictability is strongest for metals, vegetable oils and protein meals, where RMSE by
far outperforms both the AR and the random walk specification. In contrast, the largescale factor model do not show improvements, compared to the small-scale factor and
the AR model, which implies that they cannot be exploited for forecasting purposes.
The main contribution of this essay is to provide an empirical assessment of the degree
of communalities that could be useful for forecasting purposes. This essay demonstrates
that the common evolution of commodity prices with similar characteristics has
forecasting power over non-energy commodity price changes.
In contrast, the
predictive content of large-scale DFM´s forecasts on the price inflation of non-energy
27
commodities proved disappointing. The contribution is important not only because
relative success in using factor models in forecasting the nominal price of non-energy
commodity changes at short horizons is reported, but also because we contributed to the
academic debate about the forecastability on commodity price changes.
Chapter 4. Long-term links between raw material prices, real exchange rate and
relative de-industrialization in a commodity-dependent economy. Empirical
evidence of “Dutch Disease” in Colombia.
This essay reflects a change in the direction of previous studies, which evaluate the
synchronization of non-energy commodity prices, factors that impact them and the
capacity of co-movement to forecast.
In this essay I move toward the impact of
commodity prices on macroeconomic variables in a small commodity-dependent
economy such as Colombia will be. As I previously commented, for commoditydependent economies the evolution of these prices is crucial in several aspects of the
economy. Hence, a global perspective of the evolution of commodity prices has to take
into account not only that which drives commodity prices at international level, but also
what the impact of these price improvements is in a commodity-exporting economy
such as Colombia. In this way, this essay deals with the literature that evaluates
possible negative effects of commodity booms on commodity-dependent economies. In
particular, the essay centers on evaluating possible symptoms of the phenomenon
known as Dutch Disease in the long run.
The term Dutch Disease appears in relation to the discovery in the Netherlands of large
gas deposits in the North Sea and its harmful effects on the country´s industrial sector.
The sudden increase in the country’s wealth created an inflow of capital never
28
previously seen, which led to an appreciation of its currency and, therefore, a loss of
competiveness in the non-energy exporting sector. Colombian commodity export
dependence, and its possible effect on the de-industrialization of the country due to the
appreciation of the exchange rate, motivates this research.
Colombia is a special case among commodity-dependent economies since it has been
dependent on two different commodities, coffee first, and more recently oil. Hence,
methodologically the study focuses on long-term analysis and testing by means of a
Vector Error Correction Model about the relation among commodity prices and the real
exchange rate and the relative manufacturing output.
We also take into account
variables such as productivity, government expenditure, the degree of trade openness
and international inflows for the period between 1972 and 2012.
Estimations show that commodity prices are positively related to the real exchange rate
in the long term; however, relative manufacturing output is not negatively affected by
the real exchange rate. Thus, increases in commodity prices have a negative effect on
the competitiveness of the country without causing relative de-industrialization as yet.
Our result also shows that public spending is a major source of pressure on the
Colombian real exchange rate.
The most important contribution of this essay has to do with the analysis of the Dutch
Disease phenomenon in a long term context. Commodity booms in producing countries
create not only possible disadvantages such as the loss of competitiveness and higher
levels of specialization in the production of the resource and in non-tradable sectors, but
also advantages such as an increase in the country’s wealth, greater fiscal income for
social investment and an improvement in the balance of payments. Therefore, net
effects may not be observed within a short term context. Until now, studies have
29
centered specially on the last coffee boom (e.g. Puyana, 2000, Meisel 1998, Edwards,
1984 and Kamas, 1986) without covering the two historical episodes of boom in the
commodity exporting sector: the increase in coffee prices (1975-1977) and the increase
in oil prices (2003-2008). Thus, since this essay takes into account a long term
perspective, it is able to evaluate whether the adverse effects associated to Dutch
disease offset the beneficial effects of commodity upsurge.
References
Baker, S., Bloom, N. and Davis S. J. (2013) Measuring Economic Policy Uncertainty,
Chicago Booth Research Paper doi: 10.2139/ssrn.2198490, Chicago.
Bernanke, B., Boiving, J. and Eliasz, P. (2005) Measuring the effects of monetary
policy: a factor-augmented vector autoregressive (FAVAR) approach, The
Quarterly Journal of Economics, 120, 387-422.
Edwards S. (1984) Coffee Money and inflation in Colombia, World Development, 12,
1107-1117.
Meisel, A. (1998) Dutch Disease and Banana Exports in the Colombian Caribbean,
1910-1950, Cuadernos de Historia Económica y Empresarial, 26, April.
Puyana, A. (2000) Dutch Disease, Macroeconomic Policies, and Rural Poverty in
Colombia, International Journal of Politics, Culture, and Society, 14(1), 205-233.
30
Chapter 2
Common dynamics of non-energy
commodity prices and its relation to
uncertainty.
2.1. Introduction
One interesting characteristic of raw material prices is their tendency to co-move in
response to the same shock. Thus, prices which should apparently not be correlated
have a common evolution in time. For Pindyck and Rotemberg (1990), there are
macroeconomic variables such as the industrial production of developed countries,
interest rates and the exchange rate, which influence a large group of raw materials,
since they affect the current market as well as the short-term expectation of demand and
future supply. Pindyck and Rotemberg (1990) found that commodity prices co-move in
“excess” of what can be explained by these variables. This evidence casts a shadow of
doubt over the competitive formation of prices in the commodity market, and offers the
possibility of studying the role of speculation in this market.
31
Recently, and in view of the significant increase in raw material prices in the period
between 2002 and 2008 (see Figure 2.1), researchers have once again taken up the
debate on the importance of the basic macroeconomic variables which may affect comovement with regard to the role of speculation by index investors or so-called
financialization of commodity markets (see, e.g., Tang and Xiong, 2012, and Irwin,
Sanders and Merrin, 2009). An intense debate among scholars started with a Hedge
Fund manager´s testimony, that of Mr Michael W. Masters, at the U.S Senate. Masters
(2008) drew attention to the fact that the increase in speculative activity in commodity
markets, due to the arrivals of new financial traders like Corporate and Government
Pension Funds, modified the price determination in the commodity market. According
to Masters (2008), assets allocated to commodity indices by institutional investors had
risen from 13 billion at the end of 2003 to 260 billion as of March 2008, driving
commodity prices higher.
The aim of this paper is to improve the empirical evidence on commodity prices comovement in several dimensions. First, we attempt to identify the extent of comovements in a larger set of commodities, in particular, we analyze 44 monthly nonfuel commodity price series. The aim is to ascertain whether the links among
commodity prices, or co-movement, have increased since the end of 2003, as suggested
by the promoters of the financialization hypothesis (Tang and Xiong, 2012 and Masters,
2008, among others). Second, we attempt to determine the role of uncertainty as a
potential driver of non-energy co-movement in the short run. The extent to which this
co-movement is driven by macroeconomic fundamentals and financial variables is also
examined, in order to analyze their relative importance. While we evaluate the effect of
uncertainty on commodity co-movement, we add to the recent business cycle literature
on the macroeconomic effects of “uncertainty” shocks, Bloom (2009), Bernanke (2012)
32
and Baker, Bloom and Davis (2013). Finally, as suggested by Frankel and Rose (2010),
there is a lack of empirical evidence with higher frequency data, since most analyses are
performed using annual data.
The use of monthly data allows us to focus on
fluctuations due not only to fundamentals but also to speculative or financial causes.
Figure 2.1: Indices of Market Prices for Non-fuel (Beverage Food and Industrial
Inputs) and Fuel (Energy) Commodities (2005=100, in terms of U.S. dollars).
280
240
200
160
120
80
40
0
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
BEVERAGES_INDEX
ENERGY_INDEX
INDUSTRIAL_INPUT_INDEX
FOOD_INDEX
Source: International Monetary Fund, IMF.
We use monthly non fuel commodity price data from January 1992 until December
2012 and proceed in two steps: First, the use of a large number of commodities and data
from different sectors allows us to use a dynamic factor model approach, and to
estimate it through principal components. We diagnosed the overall co-movement and
confirmed the presence of one latent factor driving commodity prices with the
information criteria suggested by Bai and Ng (2002). While we were evaluating the
importance of this common factor for the variance of non-fuel commodity prices before
and after the end of 2003, we found a very interesting fact: the variance of commodity
prices explained by the common factor jumped from 9% in the first sub-period,
33
February 1992- November 2003, to 23% in the second sub-period, December 2003December 2012. In terms of the scope of the analysis, we present a `before and after´
perspective by looking at two periods; before and after December 2003. Second, we use
a Factor-Augmented Vector Autoregressive (FAVAR) approach, proposed by
Bernanke, Boivin and Eliasz (2005), to study if the co-movement of commodity prices
is affected in the same way by macroeconomic and speculative variables, as well as the
uncertainty variable in the two sub-periods.
Our study provides an understanding of the role of uncertainty and fundamental
variables that drive co-movement in commodity prices in these two periods. Therefore,
our contributions to the literature on co-movement in commodity prices are twofold.
First, by considering two different periods, we check whether co-movement is affected
in the same way by fundamental and speculative variables, irrespective of time. It is our
aim to ascertain which variables; fundamental, financial, or uncertainty, have a greater
influence on the common behavior of non-fuel commodity prices.
Second, we examine the importance of market uncertainty in the co-movement of a
range of non-fuel commodity prices in the short run. The closest paper to ours is Byrne,
Fazio and Fiess (2013), although the two approaches differ in several aspects. First, we
focus on the short-term relationship between co-movement in non-fuel commodity
prices and macroeconomic and financial variables on a monthly instead of annual basis.
The use of higher frequency data allows us to concentrate on speculative pricemovements. We believe that the effects of uncertainty over commodity prices are better
analyzed in a short-term context. Second, we apply a previously unused variable in the
study of uncertainty in commodity prices: The Chicago Board Options Exchange
Volatility Index (VIX). This variable reflects a market estimate of future volatility based
on the weighted average of the implied volatilities for a wide range of options. Since
34
uncertainty is forward looking, the variable VIX should be a better measure of
uncertainty given that it takes into account the expected volatility in the future market.
Third, we break the sample down in order to analyze the differences along the comovement before and after the end of 2003 more efficiently. We believe that the sharp
increase in the co-movement after that date deserves different treatment in the FAVAR
model.
Fourth, the estimation properties regarding consistency of the factor co-
movement are better accomplished as we do not concentrate on a few commodities, as
Byrne et al. (2013) do, but try to include as many as possible. Finally, we add two more
variables to the FAVAR, a stock market index variable and the real U.S exchange rate,
as they affect co-movement in the short term according to the literature (e.g.
Vansteenkiste, 2009, Coleman, 2012).
The rest of the paper is organized as follows. In the next sub section, we briefly review
the relevant literature. Section 2 reviews the methodology used in the study. Section 3
shows the main results and robustness checks. The conclusions are presented in section
4.
2.2. Related literature
We build mainly on the strand of literature that studies co-movements of commodity
prices. However, our work is indirectly related to the growing literature on uncertainty
and its effects on the economy.
The literature of co-movements in commodity prices is vast and begins with the seminal
work of Pindyck and Rotemberg (1990) who state the excess of co-movement
35
hypothesis. The idea put forward by Pindyck and Rotemberg (1990) suggests that
prices of a wide range of commodities, at first sight uncorrelated and with cross
elasticity of the demand close to zero, follow a common evolution over time. This
conclusion is accepted by the authors, after taking into account a series of
macroeconomic variables which may affect the set of commodity prices. For Pindyck
and Rotemberg (1990), the common behavior of raw material prices should only occur
as a reaction to common macroeconomic shocks. They found that the prices of
commodities co-move in excess of what can be explained by macroeconomic
fundamentals. The evidence of this "excess" of co-movement between different raw
material prices implies not only that the price formation is not fully competitive, but
also that speculation in commodity markets may play an important role.
In their rejection of the excess of co-movement hypothesis, some authors suggest that
macroeconomic variables largely affect the development of raw material prices, e.g
Dornbusch (1985), Chu and Morisson (1984), Borenzstein and Reinhart (1994) and
Vansteenkiste (2009). Macroeconomic variables in empirical work include those related
to demand, such as the production of developed countries, see Borenzstein and Reinhart
(1994); variables related to the cost of supplies such as gasoline and fertilizer prices, see
Baffes (2007); and variables such as the international situation, which can affect future
expectations, and future returns, such as the effective exchange rate, and the US real
interest rate, see Calvo (2008)1. The relative importance of each one of these
determinants, as well as the factor attributed to speculation and uncertainty on the
evolution of commodity prices, is currently the subject of debate.
1
The link to the state of the business cycle is explored in Camacho and Pérez-Quirós (2013) in a different
context using Markov switching models.
36
Financialization of commodity markets is cited as one of the sources of speculation, as
well as being one of the main reasons behind the surge of raw material prices after
2003. This process has been described as the substantial increase in the investment in
commodities as a form of asset. Tang and Xiong (2012) describe how the low
correlation between commodity and market indices could lead to the investors' belief
that raw material future investments could reduce portfolio risk. Therefore, after the
collapse of the dot-com bubble around 2000 and 2001, various instruments based on
commodity indices attracted investment. According to these authors, the financialization
of commodity markets not only pushed prices up after 2003, but also permitted a greater
correlation
among
different
commodities.
For
Natanelov,
Mackenzie
and
Huylendbroeck (2011) and Irwin et al. (2009) the proliferation of a variety of
instruments such as exchange-traded funds (ETFs) and structured notes (ETNs), which
bring together a number of commodities, also influence the co-movement of prices2.
While one strand of academic research investigations focused on whether
financialization is the cause of the upsurge in commodity prices, other authors have
concentrated on the correlation between different commodities or between commodities
and other financial indicators such as equities. Surveys by Irwin and Sanders (2011) and
Fattouh, Kilian and Mahadeva (2013) cast doubts on the idea that the increased
speculation in oil future markets in the post-financialization period was a key factor of
the upsurge of oil prices. On the other side, Büyüksahin and Robe (2012) and
Henderson, Pearson and Wang (2012) present evidence in favor of increasing
correlation between equity indexes and commodities due to the presence of hedge
funds. In the same line, Hamilton and Wu (2013) document that the risk premium in
crude oil futures on average decreased and became more volatile since 2005. Since we
2
In their article, Tang and Xiong (2012) use the end of 2003 and beginning of 2004 as the starting date of
the financialization of the commodities.
37
divide our sample in December 2003, when index investment started to flow into
commodity markets, we seek to add to the literature on the financialization of
commodity markets by identifying the impact of energy and stock markets in
determining the short run evolution on non-energy price co-movements3.
Recently, the growing empirical literature on uncertainty has focused on its potential
effects on the business cycle (e.g. Bloom 2009, Bernanke 2012, and Baker et al. (2013).
According to Bloom (2009), higher policy uncertainty generates a slowdown in the
economy, since businesses tend to postpone investment when there is an uncertain
environment. Thus, an uncertainty shock has real effects on output, employment and
productivity growth. Uncertainty can also affect risk-premium, which consequently
increases the financing cost for entrepreneurs and countries. It may also increase
precautionary savings in households, so generating a reduction in aggregate demand.
Despite the apparent importance of uncertainty for the real economy, the literature about
its effects on commodity prices is scarce. Beck (1993, 2001) took commodity price
volatility in the future market as a measure of risk. He found evidence that expected
price risks have a significant effect on price behavior only for storable commodities. For
Dixit and Pindyck (1994), uncertainty lowers production because the benefit of waiting
grows in line with opportunity costs. Uncertainty, therefore, generates an increase in
prices. On the contrary, Byrne et al. (2013) empirically found the opposite relationship
between risk and commodity prices in a low frequency scenario. This result is in the
line with literature that relates uncertainty with equity prices in the short run, suggesting
that prices fall precipitously on negative news, and hence induce higher risk premiums
(Pástor and Veronesi, 2011, Bekaert, Engstrom and Xing, 2009).
3
Total net assets of commodity exchange trade funds grew 10.000% between 2004 and 2010 according to
the Investment Company Institute.
38
2.3. Methodology
2.3.1.
The FAVAR model
In the first part of this section, we provide a summary of the factor-augmented vector
autoregressive (FAVAR) model that we use in the empirical section. For additional
details, see Bernanke et al. (2005).
Dynamic factor models suggest that the information from a large number of time series
can be summarized by a relatively small number of common factors plus idiosyncratic
noises. In this context, we extract a factor that represents co-movements in commodity
prices. We would like to assess for the responses of this factor to common
macroeconomic and speculative shocks.
For this purpose, we propose a Factor
Augmented VAR (FAVAR) approach along the lines of Bernanke et al. (2005). The
model is summarized in the following two equations:


[  ] = () [ −1 ] + 

−1
(2.1)
 = Λ + 
(2.2)
where  = (1 , … ,  )′ is the  × 1 vector of observed variables (non-energy
commodity inflations in our case); Λ = ( Λ´1 , … , Λ´ ),  × , is the factor loading
matrix;  is the  × 1 vector of common factors;  is the  × 1 vector of
idiosyncratic noises;  is the  × 1 vector of macroeconomic and financial variables;
() is the ( + ) × ( + ) matrix of lag polynomials and  ~(0, ∑) is the ( +
) × 1 vector of error terms in the FAVAR specification.
39
Note that in equation (2.1) the latent variable ( ) summarizes the developments of nonenergy commodity prices. Therefore, the factor,  , represents the common pattern of
commodities. In this context, equation (2.1) represents the joint dynamics of ( ,  )
and we call it FAVAR.
We cannot directly estimate equation (2.1) because the factor (Ft) is unobservable;
therefore, we need to construct the factor beforehand. In our case, the dynamic factor
model given by equation (2.2) gives the non-energy commodity price inflations (labeled
as  ) driven by a latent component,  , that is common to all series and an idiosyncratic
autoregressive component,  . The matrix Λ represents the loading of the common
factors onto series. Each element of the error or idiosyncratic term,  , contains the
dynamics specific to each commodity price inflation, although it is assumed to be
weakly correlated4. The factor may also follow an AR process.
With regard to the estimation, we consider the two-step method favored by Bernanke et
al. (2005), in which the factor is extracted prior to estimation of the FAVAR. The use
of a large number of commodities allows us to estimate equation (2.2) through Principal
Components. With this we assume that the weighted averages of the idiosyncratic
disturbances will converge to zero by the weak law of large numbers, so linear
combinations of the observed series are consistent estimators of the common factors.
Consistency of the static Principal Components estimator has been demonstrated by
Stock and Watson (2002) when both, the number of series N, and the time dimension T
converge to infinity. Due to this consistency result, we can treat  as observed for
inference purposes.
4
For a discussion of dynamic factor models, see for instance, Bai and Ng (2008) and Stock and Watson
(2011).
40
2.3.2.
Data definition
In this section we describe the data used in the factor model, related to commodity
prices and the macroeconomic and financial variables used in the FAVAR model.
For the dynamic factor analysis we used 44 monthly non-fuel commodity price series
from February 1992 to December 2012. Monthly series of commodity prices were
obtained from the International Monetary Fund database (IMF IFS). We include in our
study the raw materials available in the following categories: food, beverages,
agricultural raw material and metals. A summary of the commodities used in this study
is shown in Appendix A.
With regard to the estimation, commodity prices are log differentiated and standardized,
prior to the factor extraction by principal components. We follow Stock and Watson
(2011) in this respect.
With the FAVAR model we attempt to determine the extent to which this common
factor is driven by macroeconomic fundamentals, or whether uncertainty plays a major
role. For this purpose we select the most important macroeconomic variables used in the
literature of co-movements, and combine them with our proxy for uncertainty. Here we
add the role of uncertainty as a potential determinant of commonalities in commodity
prices. Specifically, we use the Chicago Board Options Exchange Volatility Index
(VIX) which reflects a market estimate of future volatility. To the best of the authors’
knowledge, this variable has not been used before in the analysis of commodity prices.
The rest of the variables are:
41

The United States exchange rate. We use the U.S real effective exchange rate.
We expect that the decline in the real effective exchange rate may have added
momentum to the upward commodities price movement.

United States real interest rate, proxied in our analysis by the U.S. 3-Month
Certificate of Deposit. We expect that lower rates of return on bonds will
increase the speculative demand for commodities and hence further raise their
price.

World demand, proxied in our analysis by the World Industrial Production.

Stock Market index. We take for our analysis the MSCI world index, which
includes a large collection of world stocks in the developed markets. We use this
variable as a proxy of the financial market condition.

Supply shocks are proxied in our study by the Energy Index of the IMF. For
some authors such as Krugman (2008), the increase in oil prices may explain the
contemporaneous increase in other commodities, such as food products, through
two different channels. Firstly, higher energy prices cause upsurges in the
production cost which impacts on the rest of commodities final price. Secondly,
the increased biofuel demand may lead to a reduction in food supply devoted to
final consumption. The last two channels are more associated to long-run
effects, since raw material supply is highly inelastic. In the short run, if the
financialization hypothesis prevails, we would expect that energy and nonenergy prices move in the same direction, given the speculative trade on
commodities. The energy index is composed of natural gas and coal, besides oil.
Appendix B presents further details regarding the variables we include in the model.
Graphical representations of the variables are shown in Appendix C.
42
2.4. Empirical Results
In this section we focus first on the specification of the Dynamic Factor Model and the
analysis of the factor and loadings estimated by Principal Components. The estimation
of the unobserved factors is the first steps, since we estimate the FAVAR using the twostep approach of Bernanke et al. (2005). Afterwards, we move on to the estimation of
the FAVAR for both periods and compare their results.
2.4.1.
Dynamic factor model
We identify the factor structure using the information criteria proposed by Bai and Ng
(2002). Bai and Ng (2002) solve the optimization problem that comes from minimizing
the sum of squared residuals (divided by NT), defined by:
  2

() = ()−1 ∑
=1 ∑=1( − λ  ) ,
(2.3)
where the super index k in λ  denotes that k factors are considered. To determine the
true number of factors, r (with  ≤ ), Bai and Ng (2002) formulate the following
information criteria:
̂ ) + (, )
() = (, 
(2.4)
̂ )) + (, ),
() = ln ((, 
(2.5)
43
̂ ) is the average residual variance when k factors are assumed and
where (, 
(, ) is a penalty function. Bai and Ng (2002) suggested three penalty functions to be
taken into account for principal component estimation:
1 (, ) =
+
2 (, ) =
+
3 (, ) =


ln(+)
(2.6)
2
ln(
)
(2.7)

2
ln(
)
(2.8)
2

where  = {√, √}.
In our case, the three criteria suggest that there is at least one common factor in the data,
as can be seen in Table 2.1. Accordingly, we estimate one common factor, which we
name co-movement, for the entire sample and also for the 2 periods we want to analyze:
Pre Dec-2003 and Post Dic-2003.
Table 2.1: Number of factors estimated using Bai and Ng(2002) Criteria
Sample
Dates
No. Obs
IC1
IC2
IC3
Full
1992:2-2012:12
251
1
1
1
Notes: All estimates use N=44 series
Figure 2.2, plots the first estimated common factor, which we call co-movement in nonfuel commodity prices. According to this factor, commodity price booms and busts tend
to be relatively short-lived and the factor is more volatile after the end of 2003.
Importantly, the variance explained by the factor increases significantly between
44
subsamples. When estimating the first principal component for the period between
February 1992 and November 2003, the proportion of the variance of commodity prices
explained by the factor is only 9%. However, when the same procedure is performed for
the period between December 2003 and December 2012, the variance explained
increases to 23%. This would suggest that after the end of 2003, non-fuel commodity
prices have become more synchronized or that fluctuations are higher.
Figure 2.2: Common Factor for Non-Energy Commodity Prices.
Non-Energy Factor
8
4
0
-4
-8
-12
-16
-20
92
94
96
98
00
02
04
06
08
10
12
Table 2.2 shows the commodities with factor loadings above 0.10 in both periods.
Although loadings in the first sub-period are larger for four commodities (greater than
0.30), fewer raw material prices reach 0.10 in their factor loading. For the second subperiod, in contrast, there are more commodities with loadings above 0.10.
The
correlation of the factor with each of the prices is greater in most of the commodities for
the second sub-period. It is worth noting that some commodities greatly increased their
relationship with the factor in the second period. Sugar, Rubber and Zinc for example,
had a factor loading of only 0.123, 0.105 and 0.128, respectively, in the first period.
After December 2003, however, their loadings went up to 0.21, for all three
45
commodities. The relationship between commodity prices and the factor increased for
edibles in the IMF category of "Food" such as Fish and Beef, and in "Beverages" such
as Tea and Coffee. In contrast, the factor loading decreases for "Raw Materials" related
to the group Vegetable Oils and Protein Meal.
Table 2.2: Variance shares explained by the common factor and correlation for the
two sub-periods. The table shows only commodities with a loading greater than
0.10.
Soybeans
Soybean oil
Maize
Soybean meal
Sunflower/Safflower Oil
Barley
Wheat
Nickel
Rapeseed oil
Tin
Copper
Cotton
Lead
Sugar
Zinc
Rubber
1992:2-2003:11
Average
correlation/
Loadings
Global
factor
0.393
0.803
0.353
0.721
0.319
0.650
0.311
0.634
0.300
0.612
0.288
0.588
0.229
0.469
0.221
0.451
0.195
0.398
0.186
0.379
0.175
0.358
0.168
0.343
0.168
0.343
0.123
0.252
0.128
0.261
0.105
0.214
Soybean oil
Copper
Soybeans
Tin
Coarse
Wool Fine
Rubber
Sugar
Zinc
Rapeseed oil
Maize
Barley
Nickel
Lead
Soybean meal
Cotton
Wheat
Other Milds of Coffee
Coffee Robusta
Cocoa beans
Sawnwood
Fish
Olive Oil
Sugar Free Market
Groundnuts
Aluminium
Sugar US market
Uranium
2003:12-2012:12
Average
correlation/
Loadings
Global
factor
0.266
0.803
0.238
0.720
0.231
0.698
0.225
0.678
0.219
0.661
0.217
0.654
0.213
0.645
0.211
0.639
0.210
0.633
0.209
0.631
0.207
0.625
0.207
0.624
0.200
0.605
0.191
0.577
0.183
0.553
0.180
0.544
0.168
0.509
0.165
0.498
0.158
0.478
0.135
0.409
0.127
0.384
0.123
0.371
0.121
0.364
0.120
0.362
0.117
0.353
0.110
0.332
0.110
0.331
0.103
0.312
Our analysis has highlighted, hitherto, that the increase in commodity communalities
started before the financial crisis. In this sense, our work is consistent with the vision of
46
Tang and Xiong (2012), in which the financialization of commodities, starting at the
end of 2003, has increased the synchronization among raw material prices.
2.4.2.
The FAVAR model results
In this section, we present the estimation results of the FAVAR model for the two
periods analyzed. Our purpose is to examine the short-run linkages among the variables
considered in the study. In particular, we are interested in analyzing the impact of
uncertainty and macroeconomic and speculative variable shocks on non-fuel commodity
prices. We use impulse response functions, as they can trace over time the effects on a
variable of an exogenous shock from another variable. Before starting with the results,
some final specifications of the model are discussed briefly.
As suggested by Bernanke et al. (2005), we use all variable data in first differences in
order to induce stationarity5. Moreover, all price series have been transformed into real
prices by dividing them by the US Consumer Price Index. We use Generalized Impulse
responses for the impulse response function as described by Pesaran and Shin (1998),
whose approach does not depend on the VAR ordering.
Our main results in both sub-periods are shown in Figure 2.3 and 2.4 below. Each figure
shows the impulse responses of the non-fuel commodity factor to a one-standard
deviation shock in all of the macroeconomic and financial variables selected. The
FAVAR models, within the first period (pre December 2003) and the second period
(post December-2003), were estimated with five and nine lags, respectively, in order to
5
The variable VIX is included in first differences; however, our results are robust to the inclusion of this
variable in the levels.
47
account for residual autocorrelation. The properties of the residuals of the estimated
models have been analyzed. For details see Appendix D 6.
Figure 2.3: Accumulated responses of the non-fuel commodity prices to shocks in
macroeconomic and financial variables. Period 1992:2/2003:11
Response of the non-energy f actor to
World Demand
Real Interest Rate
4
4
4
3
3
3
2
2
2
1
1
1
0
0
0
-1
-1
-1
-2
-2
-3
-2
-3
1
2
3
4
5
6
7
8
9
10
11
12
-3
1
World Stock Index
4
2
3
4
5
6
7
8
9
10
11
1
12
3
2
2
2
1
1
1
0
0
0
-1
-1
-1
-2
2
3
4
5
6
7
8
9
10
11
5
6
7
8
9
10
11
12
9
10
11
12
-3
1
12
4
-2
-3
1
3
4
3
-3
2
Energy Prices
Real Ef f ectiv e Exchange Rate
4
3
-2
Uncertainty
2
3
4
5
6
7
8
9
10
11
1
12
2
3
4
5
6
7
8
Figure 2.4: Accumulated responses of the non-fuel commodity prices to shocks in
macroeconomic and financial variables. Period 2003:12/2012:12
Uncertainty
World Demand
Real Interest Rate
6
6
6
4
4
4
2
2
2
0
0
0
-2
-2
-2
-4
-4
-4
-6
-6
1
2
3
4
5
6
7
8
-6
1
2
3
4
5
6
7
8
1
2
3
4
5
6
7
8
6
7
8
Energy Prices
Real Ef f ectiv e Exchange Rate
World Stock Index
6
6
6
4
4
4
2
2
2
0
0
0
-2
-2
-2
-4
-4
-4
-6
-6
1
2
3
4
5
6
7
8
-6
1
2
3
4
5
6
7
8
1
2
3
4
5
6
A dummy on December 2008 has been included in the FAVAR model for the second period to take into
account the world economic crisis. We did not detect heteroscedasticity in the residuals of the FAVAR
model.
48
For the period prior to December 2003, the impulse response functions of Figure 2.3
show that the co-movement in non-energy commodities is not significantly affected by
most of the macroeconomic or speculative variables. Non-energy co-movement is only
affected by a world demand shock.
Instead, the shapes of the estimated impulse
response functions for the second sub-period are in line with co-movement literature,
see Figure 2.4. We obtain the following results for the second period:
Impulse response estimates from the FAVAR model indicated that a world demand
shock seems to show a significant positive rise in non-energy commodity prices. Other
macroeconomic variables, such as real interest rate and US real effective exchange rate,
have a significant negative impact on non-fuel commodity prices. It seems that in the
short run a positive shock in the US interest rate can make investors switch their
portfolio investments from risky assets, such as commodity futures, to more
conservative ones, such as US treasury bills. The negative effect of a shock on real
interest rate on the non-energy factor lasts for 5 months, one of the more lasting effects
regarding the impulse response results of the FAVAR model. On the other hand, a real
appreciation of the US dollar makes commodity prices fall, as most of the raw material
prices are internationally traded and quoted in U.S dollars. The negative effect takes
four months to vanish. These results agree with the view that the increase in global
demand, the real devaluation of the US dollar and the easy monetary policy of the
United States may have added momentum to the upward price movement after 2003.
A supply shock, proxied in this study by energy prices, has an immediate effect on the
non-fuel factor. However, this effect tends to disappear after the second month. In this
49
context, we agree with the idea that there is a spillover effect from energy to non-energy
commodity prices, enhanced perhaps by the financialization of commodity markets.
This effect, however, is less lasting than the shocks in the fundamentals.
With regard to financial variables, we find significant responses of non-fuel commodity
prices. Impulse response estimates indicate that a shock in the stock market index is
followed by a two month rise in non-fuel commodity prices. These results show the
strengthening of cross market linkage after the late 20037.
In relation with the effects of uncertainty, that it is proxied by VIX index in the study,
we find that an increase in this variable leads to a decrease in the non-energy
commodity prices for two months. It seems that uncertainty or volatility in the financial
market makes investors more risk averse and bearing. This result is consistent with the
ideas of Dixit and Pindyck (1994), in which uncertainty is associated with movements
in commodity prices. The negative relationship between our uncertainty variable and
commodity prices are in line with the results of Byrne et al. (2013) who tested it for the
long run.
2.4.3.
Robustness checks
We propose to check the robustness of our results against other measures of uncertainty,
the number of factors within the FAVAR model, and the inclusion of inventories into
the model.
7
Our findings are in line with the fast-growing literature on co-movements between commodity and
equity markets over time –see, e.g., Silvennoinen and Thorp (2013), Stoll and Whaley (2010),
Büyüksahin, Haigh and Robe (2010) and Büyüksahin and Robe (2012).
50
For other measures of uncertainty, we use both the “Economic Policy Uncertainty
Measure” and the “Equity Uncertainty Measure” constructed by Baker et al. (2013).
The former is constructed taking into account policy-related economic uncertainty with
three types of components: frequency of newspaper coverage of policy-related
economic uncertainty; federal tax code provisions which were about to expire; and
disagreement among economic forecasters. For the second sub-period we took the
United States as well as the European Monthly Index of Economic Policy Uncertainty
of Baker et al. (2013). Since the European Monthly Index is constructed from January
1997, we only performed this measure for the second sub-period. The latter, the
“Equity Uncertainty Measure” is based on an analysis of news articles in the United
States containing terms related to market uncertainty. Robustness checks appear in
Appendix E.
The results for the first sub-period show that all our non-energy commodity price
reactions are not significant to any alternative uncertainty measure as is shown in our
baseline model.
The results for the second sub-period show that a shock in the
European Economic Policy Uncertainty as well as in the Equity Uncertainty Measure
has a negative impact on the non-energy commodity factor. The United States Policy
Uncertainty, however, has no impact on our non-energy factor. From a quantitative
standpoint, differences are found for the non-energy responses predicted by the VAR,
embedding the European Policy Uncertainty indicator, which predicts much larger
responses. However, all non-energy commodity responses are significant and take a
sign in line with that suggested by the FAVAR with the VIX indicator.
Although we found only one factor in the whole sample, the determination of the
number of factors on the basis of subsamples, say pre-2003 and post-2003 periods, by
the information criteria of Bai and Ng (2002) were not so clear, since the third IC
51
pointed the possibility of 2 common factors. To tackle this issue, we evaluate whether
our results are robust to two factors. We extract the second component, called a shape
component, and re-estimate the FAVAR model, taking into account two factors for the
second sub-period. The loadings of the shape component show a separation between
metals commodities, with positive loadings, and edibles, with negative loadings. On the
other hand, impulse response functions show the same results as our baseline model
with regard to the responses of our first factor, called co-movement, to impulses on
uncertainty and the rest of the variables. With regard to the shape component, we found
significant responses to impulses of energy prices and the real effective exchange rate.
Appendix F reports the robustness estimations of the FAVAR model with two factors
for the second sub-period.
As a final robustness check we include an inventory variable in our FAVAR model8.
For the second period we found two results: First, our results are robust to the inclusion
of inventories. In fact, the FAVAR model with inventories implies a more important
role for uncertainty shocks in explaining fluctuations in the co-movement of non-fuel
commodity prices than previous estimates. As shown in Figure 2.5, an increase in
uncertainty leads to a decrease in co-movement for three months. Second, inventories
have a positive effect on non-fuel commodity prices. Either through fear of future
production shortage or speculation (greater expected future prices), inventory
accumulation leads to the reduction in the commodities available for current use,
causing the non-fuel spot prices to rise.
8
We used inventory data of only 12 non fuel commodities due to the lack of information available for
the rest of the commodities at a monthly frequency. We use world inventories as far as possible, but
substitute with US inventories when these are missing. We describe the inventory data used in Appendix
B. The inventory variable is taken from estimating the first principal component of the 12 non-fuel
inventories.
52
With respect to the first sub-period, we estimated the FAVAR with inventories of
metals due to the lack of a longer data span for the rest of stocks of commodities.
Results show little evidence that a positive shock on metal inventories affects the nonfuel commodity factor. This result contrasts with the positive relationship between
inventories and non-fuel commodity prices for the second sub-period9.
Figure 2.5: Accumulated responses of the non-fuel commodity prices to shocks in
macroeconomic, financial and inventory variables. Period 2003:12/2012:12
Real Interest Rate
World Demand
Uncertainty
8
8
8
4
4
4
0
0
0
-4
-4
-4
-8
-8
-8
-12
-12
-12
1
2
3
4
5
6
7
8
9
10
11
1
12
2
3
4
5
6
7
8
9
10
11
1
12
World Stock Index
8
8
4
4
4
0
0
0
-4
-4
-4
-8
-8
-8
-12
-12
-12
2
3
4
5
6
7
8
3
4
9
10
11
12
1
2
3
4
5
6
7
8
9
10
5
6
7
8
9
10
11
12
9
10
11
12
Inv entories
8
1
2
Real Ef f ectiv e Exchange Rate
11
12
1
2
3
4
5
6
7
8
Energy Prices
8
4
0
-4
-8
-12
1
2
3
4
5
6
7
8
9
10
11
12
9
We estimated the FAVAR model for both sub-periods, taking into account the inventories of metals. For
the second sub-period we obtained similar results as the model with total inventories. Estimation results
are available upon request.
53
2.4.4.
Variance decompositions
Now we present the variance decomposition of the non-energy factor for both subperiods in order to assess the relative importance of each of the variables in determining
the non-energy price co-movement. Yet, variance decomposition requires orthogonality
of the inputs if we want the sum of variance due to each component to total 100%. We
rely on a given Cholesky decomposition and, therefore, we propose the following
ordering of the variables in the VAR: in both of the sub-periods we place our nonenergy factor last based on the assumption that non-energy commodity market shocks
have no contemporaneous effects on macroeconomic variables. World demand, proxied
in our study by global industrial production, is placed first, followed by the interest rate,
the volatility variable (VIX), the Stock Market index (MSCI), real effective exchange
rate and the IMF energy index. This implies that demand shocks instantaneously affect
all equations of the system, which seems logical if we consider that the real exchange
rates and interest rates respond greatly to macroeconomic conditions10.
Figure 2.6 illustrates the percentage of variance of the forecasting error of the nonenergy factor, at a 12 month horizon, that is attributable to a shock in each of the
macroeconomic and financial variables of the model.
10
As a robustness check for this part, we evaluated a different Cholesky ordering of the variables in the
VAR with similar results.
54
Figure 2.6: Variance decomposition of the non-energy factor
80
70
60
50
40
30
20
10
0
World
Demand
Real Interest
Rate
Uncertainty
World Stock Real Effective
Market
Exchange Rate
1992:2-2003:11
Energy
Non-energy
factor
2003:12-2012:12
Note: Variance decomposition at horizon 12.
In the first sub-period the non-energy factor is the most important driver of its own
prices, accounting for more than 70% of price fluctuation. Therefore, inertia has a
dominant role in explaining the factor fluctuations. Considering the other variables
rather than the lags themselves of the common factor, world demand shocks are the
most important driver, explaining up to 7% of non-energy price changes. The rest of the
variables shocks account for less than 5% of the non-energy factor.
In the second sub-period the relative importance of the shocks is greatly altered. Real
interest rate is now the most important driver of co-movements in non-energy
commodity prices, accounting for up to 26% of price fluctuations. The role of a shock in
uncertainty is notable, explaining 17.4% of the variation in non-energy prices at all
horizons, the second most important driver of the non-energy factor for this period.
Shocks in the fundamentals, such as real effective exchange rate and world demand,
account for up to 17% and 8%, respectively. Finally, the stock market index variable
and the energy variable explain 8% and 7%, respectively, of the variation in non-energy
price fluctuations.
55
The most important result from this comparison is the increase in the role of uncertainty
to explain non-energy fluctuations. The fact that the uncertainty shock accounts for a
larger share of variance decomposition of non-energy commodity prices than
fundamentals, such as real exchange rate and world demand, emphasizes its importance
in determining non-energy spot price formation. Financialization of commodities may
make investors much more aware of short term economic developments and may cause
passive investors to take collective decisions such as selling risky assets, like
commodities, when uncertainty increases, given their correlation with other risky assets.
In conclusion, in the short term, the speculative variables used in our study, which
reflect the financial market condition (world stock index) and uncertainty (VIX),
significantly affect the real prices of non-fuel commodities in the second period, after
the end of 2003. The role of the uncertainty element in determining the joint movements
of non-energy prices is noteworthy.
2.5. Conclusions
This article improves the understanding of the co-movement among non-fuel
commodity prices in the short run adding to the literature on financialization of
commodity markets. Firstly, we evaluate the magnitude in the co-movement of 44
monthly non-fuel commodity prices from February 1992 to December 2012. Secondly,
we break our sample in December 2003, as it is the starting date of unprecedented
upsurge of investment into commodity trade funds, in an attempt to ascertain whether
co-movement is affected in the same way by fundamentals, financial and speculative
variables. We evaluate the responses of non-energy commodities to macroeconomic
and financial shocks for these two sub-periods. As we focus in the short run, special
56
attention is paid on the role of the stock market index variable as well as uncertainty in
the financial market as a possible determinant of commonalities in commodity prices.
With regard to the methodology, we use the FAVAR approach of Bernanke et al. (2005)
in order to check what drives co-movement in non-energy commodity prices.
Our results highlight the importance of co-movement between non-fuel commodities
that started after the financialization of commodities in late 2003. In this period we
found not only an increase in the variance explained by the factor attributable to the comovement between raw materials, but also the impulse response functions show that
variables such as uncertainty and the stock market index significantly impact the factor
that relates co-movements in non-energy commodity prices. Moreover, the variance
decomposition analysis shows that the uncertainty element plays a larger role in
explaining non-energy fluctuation than fundamentals such as the real exchange rate and
the real interest rate.
In short, although classical macroeconomic factors of supply and demand are able to
explain much of the sharp movements in the short term in non-fuel commodity prices,
the growing importance of uncertainty as an important determinant of communalities in
non-energy commodities is an element that cannot be ignored.
57
References
Baffes, J. (2007) Oil spills on other commodities, Resources Policy, 32, 126-134.
Bai, J. and Ng, S. (2002) Determining the Number of Factors in Approximate Factor
Models, Econometrica, 70, 191-221.
Bai, J. and Ng, S. (2008) Large Dimensional Factor Analysis, Foundations and Trends
in Econometrics, 3, 89-163.
Baker, S., Bloom, N. and Davis S. J. (2013) Measuring Economic Policy Uncertainty,
Chicago Booth Research Paper doi: 10.2139/ssrn.2198490, Chicago.
Beck, S. (1993) A rational expectations model of time varying risk premia in
commodities futures markets: Theory and Evidence, International Economic
Review, 34, 149-168
Beck, S. (2001) Autoregressive conditional heteroscedasticity in commodity spot prices,
Journal of Applied Econometrics, 16, 115–132.
Bekaert, G., Engstrom, E. and Xing, Y. (2009) Risk, uncertainty, and asset prices,
Journal of Financial Economics, 91, 59-82.
Bernanke, B., Boiving, J. and Eliasz, P. (2005) Measuring the effects of monetary
policy: a factor-augmented vector autoregressive (FAVAR) approach, The
Quarterly Journal of Economics, 120, 387-422.
Bernanke, B. (2012) The Economic Recovery and Economic Policy, Speech held at the
New York Economic Club, New York, November 20.
Bloom, N. (2009) The Impact of Uncertainty Shocks, Econometrica, 77, 623-685.
Byrne, J., Fazio, G. and Fiess N. (2013), Primary commodity prices: Co-movements,
common factors and fundamentals, Journal of Development Economics, 101, 1626.
58
Borensztein, E. and Reinhart, C. (1994) The macroeconomic determinants of
commodity prices, IMF Staff papers, 41, 236.
Büyüksahin, B., Haigh, M. and Robe, M.A. (2010) Commodities and Equities: Ever a
‘Market of One’? , Journal of Alternative Investments, 12, 75–95.
Büyüksahin, B. and Robe, M.A. (2012) Speculators, commodities and cross-market
linkages doi: 10.2139/ssrn.1707103
Calvo, G. (2008) Exploding commodity prices, lax monetary policy, and sovereign
wealth funds, VoxEU. 20 June.
Camacho, M. and Pérez-Quirós, G. (2013) Commodity prices and the business cycle in
Latin America: Living and dying by commodities?, forthcoming: Emerging
Markets Finance and Trade.
Chu, K. and Morrison, T. (1984) The 1981-82 recession and non-oil primary
commodity prices, Staff papers - International Monetary Fund, 31, 93-140.
Coleman, L. (2012) Explaining crude oil prices using fundamental measures, Energy
Policy, 40, 318-324.
Dixit, A.K. and Pindyck, R.S. (1994) Investment under uncertainty, Princeton
University Press, Princeton, N.J.
Dornbusch, R. (1985) Policy and Performance Links between LDC Debtors and
Industrial Nations, Economic Studies Program, The Brookings Institution, 16, 303368.
Fattouh, B., Kilian, L. and Mahadeva, L. (2013) The role of speculation in oil markets:
What have we learned so far? , Energy Journal, forthcoming.
Frankel, J. and Rose, A. (2010) Determinants of Agricultural and Mineral Commodity
Prices, Working Paper Series rwp10-038, Harvard University, John F. Kennedy
School of Government, Cambridge.
59
Hamilton, J.D. and Wu, J.C. (2013) Risk premia in crude oil futures prices, NBER
Working Papers 19056, National Bureau of Economic Research, Inc.
Henderson, B., Pearson, N., and Wang, L. (2012) New Evidence on the Financialization
of Commodity Markets, Working Paper, University of Illinois, Urbana-Champaign.
Irwin, S.H., Sanders, D.R. and Merrin, R. (2009) Devil or angel? The role of
speculation in the recent commodity price boom (and bust), Journal of Agricultural
and Applied Economics, 41, 2.
Irwin, S.H. and Sanders, D.R. (2011) Index funds, financialization, and commodity
futures markets, Applied Economic Perspectives and Policy, 33, 1–31.
Krugman, P. (2008) Running out of planet to exploit, The New York times, 21.
Masters, M. (2008) Testimony before the Committee on Homeland Security and
Governmental Affairs. United States Senate. May 20.
Natanelov, V., Alam, M.J., McKenzie, A.M. and Van Huylenbroeck, G. (2011) Is there
co-movement of agricultural commodities futures prices and crude oil?, Energy
Policy, 39, 4971-4984.
Pástor, L.
and Veronesi, P. (2011) Political uncertainty and risk premia, CEPR
Discussion Paper 8601, London.
Pesaran, H. and Shin, Y. (1998) Generalized impulse response analysis in linear
multivariate models, Economics Letters, 58, 17-29.
Pindyck, R. and Rotemberg, J. (1990) The Excess co-movement of commodity prices,
Economic Journal, 100, 1173-1189.
Silvennoinen, A. and Thorp S. (2013) Financialization, crisis and commodity
correlation dynamics, Journal of International Financial Markets, Institutions, and
Money, 24, 42-65.
60
Stock, J. and Watson, M. (2002) Forecasting using Principal components from a large
number of predictors, Journal of the American Statistical Association 97, 11671179.
Stock, J. and Watson, M. (2011) Dynamic Factor Models, in Clements MJ, Hendry DF
ed., Oxford Handbook on Economic Forecasting, Oxford University Press, Oxford.
Stoll, H. and Whaley, R. (2010) Commodity index investing and commodity futures
prices, Journal of Applied Finance, 1, 7–46.
Tang, K., and Xiong, W. (2012) Index investment and the financialization of
commodities, Financial Analysts Journal, 68, 54-74.
Vansteenkiste, I. (2009) How Important are Common Factors in Driving Non-Fuel
Commodity Prices? A Dynamic Factor Analysis, Working Paper Series 1072,
European Central Bank.
61
Appendix A: Non fuel commodities
IMF Category
Edibles
Commodity
Rice
Maiz
Barley
Wheat
Beef
Poultry (chicken)
Lamb
Fish (salmon)
Swine (pork)
Shrimp
Cocoa beans
Coffee, Robusta
Coffee, Other Mild Arabicas,
Tea
Vegetable oils
Soybeans
Sunflower Oil
and protein meals
Soybean Meal
Olive Oil
Soybean Oil
Groundnuts (peanuts)
Palm oil
Rapeseed oil, crude.
Cereals
Meet and seafood
Beverages
Fishmeal
Other Edibles
Sugar EU.
Oranges
Sugar US.
Bananas
Soft Logs
Industrial
Agricultural
Hard Logs
Wool, fine
Inputs
raw materials
Hard Sawnwood
Rubber
Soft Sawnwood
Hides
Cotton
Wool, coarse
Aluminum
Zinc
Copper
Nickel
Iron Ore
Lead
Tin
Uranium
Metals
62
Appendix B: Macroeconomic and Financial Variables
U.S Real Effective Exchange rate. Reer Based on Rel.cp /Index Number /averages
/seas. adjusted /Cnt: United States /Source: IMF, Wash
U.S Real interest rate. 3-Month Certificate of Deposit: Secondary Market Rate/Unit:
Percent /Cnt: United States /Source: FRED
Energy Index. The Commodity Fuel (energy) Index includes Crude oil (petroleum),
Natural Gas, and Coal Price Indices. Base year: 2005/source: IMF.
World Industrial Production. Data from CPB Netherlands Bureau for Economic Policy
Analysis
MSCI (Morgan Stanley Capital International) World Index. Data from Bloomberg
VIX index. The Chicago Board Options Exchange Volatility Index reflects a market
estimate of future volatility, based on the weighted average of the implied volatilities
for a wide range of strikes. 1st & 2nd month expirations are used until 8 days from
expiration, then the 2nd and 3rd are used. Source: Bloomberg [BBGID
BBG000JW9B77]
We used the Consumer Price Index to deflate the series that was not in real terms.
Especifically: Cpi all Items City Average /Index Number /Base year: 2005 /averages
/Cnt: United States /Source: IMF, Wash
Name and sources of the inventory data used in in the robustness check section
Name
Wheat Ending
Stocks
Million Bushels
Sugar Stocks to UseMonths
Aluminium
Warehouse Stocks
Metric Tonnes
Copper
Warehouse Stocks
Metric Tonnes
Tin
Warehouse Stocks
Metric Tonnes
Zinc
Warehouse Stocks
Metric Tonnes
Nickel
Ubication
United States
Frequency
Monthly
Period
1990:1-2012:12
Source
United States
Department of
Agriculture, USDA.
F.O. Licht
Global
Monthly
19977:1-2012:12
Global
Monthly
1987:7-2012:12
London Metal
Exchange
Global
Monthly
1988:10-2012:12
London Metal
Exchange
Global
Monthly
1989:05-2012:12
London Metal
Exchange
Global
Monthly
1988:09-2012:12
London Metal
Exchange
Global
Monthly
1987:7-2012:12
London Metal
63
metric tonnes
Warehouse Stocks
Metric Tonnes
Lead
London Metal
Exchange
Warehouse Stocks
Lead Index
Soybean Oil
1000 Metric Tons
Ending Stocks
Soybean Meal
Ending Stocks
1000 Metric Tons
Exchange
Global
Monthly
1987:7-2012:12
London Metal
Exchange
United States
Monthly
1994:12-2012:12
USDA
United States
Monthly
1994:12-2012:12
USDA
Cocoa
Metric Tonne
Global
Monthly
2002:1-2012:12
Corn Ending Stock
1000 Metric Tons
Global Green Coffee
Stocks to Use-Days
United States
Monthly
1994:12-2012:12
NYSE Liffe (London
International
Financial Futures and
Options Exchange)
USDA
Global
Monthly
1997:1-2012:12
F.O. Licht
64
Appendix C: Graphs of the macroeconomic and financial
variables.
Uncertainty (VIX)
Energ y Index
World demand (World industrial production)
70
130
60
120
2.5
2.0
110
50
100
1.5
90
1.0
40
30
80
0.5
20
70
10
60
92
94
96
98
00
02
04
06
08
10
12
0.0
92
Real World Stock Index (MSCI)
94
96
98
00
02
04
06
08
10
12
92
94
96
98
Non-energy factor for the first sub-period
18
00
02
04
06
08
10
12
06
08
10
12
Real interest rate
8
.10
16
.08
4
14
.06
12
0
.04
10
-4
.02
8
6
-8
92
94
96
98
00
02
04
06
08
10
12
.00
92
Real effective exchang e rate
94
96
98
00
02
04
06
08
10
12
92
94
96
98
00
02
04
Non-energ y factor for the second sub-period
120
10
5
110
0
100
-5
-10
90
-15
80
-20
92
94
96
98
00
02
04
06
08
10
12
92
94
96
98
00
02
04
06
08
10
12
65
Appendix D: Residual Autocorrelation analysis of the
FAVAR model. Lagrange Multiplier (LM) Test.
1992:2-2003:11
2003:12-2012:11
Test
Chi-sq
p-value
LM(1)
^2 (49)= 44.49
0.6563
LM(2)
^2 (49)= 43.09
0.7101
LM(3)
^2 (49)= 51.92
0.3605
LM(4)
^2 (49)= 44.26
0.6651
LM(5)
^2 (49)= 47.06
0.5517
LM(6)
^2 (49)= 39.49
0.8318
LM(7)
^2 (49)= 48.78
0.4820
LM(8)
^2 (49)= 38.42
0.8616
LM(9)
^2 (49)= 50.33
0.4099
LM(10)
^2 (49)= 55.48
0.2436
LM(11)
^2 (49)= 50.33
0.4202
LM(12)
^2 (49)= 35.30
0.9292
Test
Chi-sq
p-value
LM(1)
^2 (49)= 55.26
0.2501
LM(2)
^2 (49)= 47.01
0.5539
LM(3)
^2 (49)= 46.43
0.5778
LM(4)
^2 (49)= 50.03
0.4319
LM(5)
^2 (49)= 60.54
0.1247
LM(6)
^2 (49)= 49.22
0.4641
LM(7)
^2 (49)= 39.56
0.8297
LM(8)
^2 (49)= 56.75
0.2085
LM(9)
^2 (49)= 52.42
0.3425
LM(10)
^2 (49)= 43.87
0.6824
LM(11)
^2 (49)= 55.29
0.2490
LM(12)
^2 (49)= 46.67
0.5694
66
Appendix E: Robustness checks: different proxies for
uncertainty.
Accumulated responses of the non-fuel commodity prices to shocks in macroeconomic
and financial variables. Uncertainty is proxy for the Equity Uncertainty Measure
constructed by Baker, Bloom, and Davis (2012). Period 2003:12/2012:12
World Demand
Real Interest Rate
Eq uity Uncertainty
8
8
8
4
4
4
0
0
0
-4
-4
-4
-8
-8
1
2
3
4
5
6
7
8
9
10
11
-8
12
1
2
World Stock Index
3
4
5
6
7
8
9
10
11
12
1
8
4
4
4
0
0
0
-4
-4
-4
-8
-8
3
4
5
6
7
8
4
5
9
10
11
6
7
8
9
10
11
12
9
10
11
12
Energy Prices
8
2
3
Real Effective Exchang e Rate
8
1
2
-8
12
1
2
3
4
5
6
7
8
9
10
11
12
1
2
3
4
5
6
7
8
Accumulated responses of the non-fuel commodity prices to shocks in macroeconomic
and financial variables. Uncertainty is proxy for the European Policy Uncertainty
Measure constructed by Baker, Bloom, and Davis (2012). Period 2003:12/2012:12
World Demand
Real Interest Rate
EU Policy Uncertainty
8
8
8
4
4
4
0
0
0
-4
-4
-4
-8
-8
1
2
3
4
5
6
7
8
9
-8
1
10
2
3
World Stock Index
4
5
6
7
8
9
1
10
8
4
4
4
0
0
0
-4
-4
-4
-8
-8
3
4
5
6
7
8
9
10
4
5
6
7
8
9
10
8
9
10
Energy Prices
8
2
3
Real Exchange Rate
8
1
2
-8
1
2
3
4
5
6
7
8
9
10
1
2
3
4
5
6
7
67
Accumulated responses of the non-fuel commodity prices to shocks in macroeconomic
and financial variables. Uncertainty is proxy for the Equity Uncertainty Measure
constructed by Baker, Bloom, and Davis (2012). Period 1992:2/2003:11
World Demand
Real Interest Rate
Equity Uncertainty
4
4
4
3
3
3
2
2
2
1
1
1
0
0
0
-1
-1
-1
-2
-2
-2
-3
-3
1
2
3
4
5
6
7
8
9
10
11
12
-3
1
2
3
World Stock Index
4
5
6
7
8
9
10
11
12
1
4
4
3
3
3
2
2
2
1
1
1
0
0
0
-1
-1
-1
-2
-2
-2
-3
-3
2
3
4
5
6
7
8
3
4
9
10
11
12
5
6
7
8
9
10
11
12
9
10
11
12
Energy Prices
4
1
2
Real Exchange Rate
-3
1
2
3
4
5
6
7
8
9
10
11
12
1
2
3
4
5
6
7
8
68
Appendix F: Robustness checks: FAVAR model with two
non-fuel commodity factors.
Accumulated responses of the co-movement factor (first principal component) to shocks
in macroeconomic and financial variables.
Accumulated responses of the Shape Component (second principal component) to
shocks in macroeconomic and financial variables.
69
Chapter 3.
The predictive content of co-movement in
non-energy commodity price changes
3.1. Introduction
In a recent study Poncela, Senra and Sierra (2013) found that there has been an increase
in co-movement in a large range of non-energy commodity prices since 2004, perhaps
enhanced by the financialization in the commodity markets . Thus, prices which should
apparently not be correlated, increased their common evolution in time. According to
this study, the variance of commodity prices explained by the common behavior of 44
non-energy commodity prices, jumped from 9% between February 1992 and November
2003, to 23% between December 2003 and December 2012. This means that after 2004
the common behavior of non-energy commodity prices accounts for a larger share of
those fluctuations. It is therefore of interest to explore whether co-movement in prices
of raw materials has some predictive power over each non energy commodity price.
To analyze the consequences of this stylized fact in forecasting, we compare several
models against a baseline random walk alternative. We aim to explore the predictability
70
of 44 non-fuel commodity spot prices measured on a monthly basis. For this purpose,
we use a Dynamic Factor Model (DFM) to extract a latent factor that drives the comovement on non-energy commodity prices. We evaluate two variants: a large-scale
DFM that uses the whole commodity price data and estimate their co-movement
through Principal Components and a small-scale DFM that takes into account the
communalities into commodities of the same category and estimate factors by means of
the Kalman filter. Our measure of forecasting performance is the out-of-sample root
mean square error of prediction (RMSE) for one-step-ahead forecasts.
Although the literature on commodity price forecasts is extensive, it provides only scant
empirical evidence of the role of co-movement in commodity prices as a possible source
of predictability in non-energy spot prices.
The recent literature has focused on
evaluating whether macroeconomic and financial variables have some predictive power
over commodity price spot indices, with mixed results. Chen, Rogoff and Rossi (2010)
found that exchange rate fluctuations in a group of commodity-dependent countries
have robust power in forecasting commodity price indices . Groen and Pesenti (2011)
used a large set of macroeconomic variables, apart from exchange rates, to evaluate
their predictive power over commodity indices. They did not find a robust validation of
Chen et. al (2010)´s previous conclusions.
Moreover, although the inclusion of
multivariate macroeconomic variables improves the forecasts, it does not produce an
overwhelming advantage of spot price predictability when compared with the random
walk model. Gargano and Timmermann (2014) found that the predictability power of
macroeconomic and financial variables depends on the state of the economy.
Another branch of the literature has focused on whether futures prices are good
predictors of future spot prices. Chinn y Coibion (2013) evaluate the forecasts of a
range of commodity prices finding that futures prices for precious and base metals
71
display very limited predictive content for future price changes. In contrast, futures
prices for energy and agricultural commodities do relatively better in terms of predicting
subsequent price changes. In regard to oil prices, Alquist and Kilian (2010) use two
models: one that considers the current level of futures prices as the predictor and the
second which is based on the futures spread, to conclude that oil futures prices fail to
improve on the accuracy of simple no-change forecasts.
Our paper considers the following research questions: First, does co-movement in nonenergy commodity prices has predictive power over non-energy commodity prices?
Second, has co-movement in commodity prices by category added power to the
prediction in comparison with the large scale co-movement in commodity prices? Third,
does the predictability of commodity prices vary across different types of categories,
such as agricultural versus raw industrial commodities? We aim to answer these
questions using dynamic factor models.
The paper is organized as follows. In section 2 we present the different models we
estimate. In section 3 we describe the data and the methodological procedure we
propose. In section 4 we report the estimation and forecasting results. Finally, in section
5 we conclude.
3.2. Model specifications
The first two models are limited to the information embedded in each commodity price
time series itself: the first is a random walk model, used as benchmark, and the second
is a univariate autoregressive (AR) model.
72
Let , be the spot price of the i-th commodity at time t, i=1,…n and  = 1, … . . , .
Then , = ln(, )-ln(,−1 ) denotes its related non-energy commodity price inflation.
Then, the unconditional mean benchmark model is:
, =  + , ,
(3.1)
which implies that the best forecast of the spot price of commodities is simply the
current spot price plus the drift  if it were different from zero.
The AR( ) model for the i-th commodity, i=1,….,n follows the specification:
, =  + ,1 ,−1 + ⋯ + , ,− + , ,
 = 1, … , .
(3.2)
The subsequent models include a latent variable, or factor, that represents the common
pattern of commodity prices. The general DFM specification assumes that the i-th
commodity price inflation, labelled as  , is driven by a latent component,  ,which is
common to all series plus an idiosyncratic component, , 11. For instance, specific to
each  we obtain:
 =   + , ,
∀
 = 1, … , 
(3.3)
where  is the loading of the common factor into the -th commodity. Accordingly, the
factor base regressions, related to the large-scale DFM follow the specification:
,+1 =   + ,+1
(3.4)
We also evaluate whether the inclusion of the forecast of the idiosyncratic component of
the DFM, , , improves the forecasting performance, or it is only the forecast of the
11
Although the DFM may have multiple factors, we have only identified the factor structure using the information
criteria proposed by Bai and Ng (2002) which confirm that there is one factor in the commodity price data.
73
common part what is valuable for forecasting12. Then, the factor base regression related
to the large-scale DFM that takes into account the idiosyncratic component follows the
specification:
,+1 = 1  + 2 , + ,+1
(3.5)
Besides estimating a large-scale DFM, which takes into account a single common factor
to all the commodity price series (equations 3.4-3.5), in this paper we also estimate a set
of small-scale DFM models by introducing dynamic factors which are common only to
the series within each set. More precisely, let us consider L commodity categories, and
for each category (category l=1,2,…,L)  commodity price series. Then, if we are
interested in one-step-ahead predictions, the baseline model for each commodity price
in the  ℎ category can be decomposed into the following components:

, =  , + ,,
 = 1, …  ,
∀ ∀
(3.6)
where within each category l, , is the factor or co-movement variable common to all
series in the category,  represents the factor loading, and , named idiosyncratic
component, collects the dynamics specific to each commodity price inflation. Both the
common factor and the idiosyncratic component may follow AR processes of order 
and  , respectively.



, = 0,1
,−1 + ⋯ + 0,
,− + 0,
(3.7)





,
= ,1
,−1
+ ⋯ + ,

+  ,
,
 ,−
(3.8)
12
Currently, small-scale factor models also include the forecasting of the idiosyncratic component (see, for instance,
Camacho and Perez-Quirós, 2010) while forecasting through large scale factor models only use the common factors
embedded in a forecasting equation with own lags of the target variable to reproduce specific dynamics (see, for
instance, Stock and Watson, 2011). The advantage of including the forecast of the idiosyncratic component instead of
own lags of the target variable could be due to the fact that the idiosyncratic component is uncorrelated with the
common factors. We aim to check the usefulness of the idiosyncratic component in factor forecasting with and
without this component.
74
where  is the standard deviation of the idiosyncratic component, and , ∼ (0,1)
 = , …  ,  = 1, … , , are the innovations to the law motions for equations (3.7) and
(3.8), respectively. We also evaluate whether the inclusion of the forecast of the
idiosyncratic component of the DFM improves the forecasting performance in the
small-scale DFMs.
With regard to the estimation method for the large-scale DFM, we use Principal
Components. The large number of commodities used to evaluate the factor in the largescale DFM, allows us to assume consistency of the Principal Components estimator13.
As regards to the small-scale DFMs, we estimate them in the state-space using the
Kalman filter. The smaller number of variables involved in the factor models by
category impedes us from using the estimator of principal components in this latter
case. The Kalman filter also produces filtered inferences of the common factor that can
be used in the prediction equation (3.9 and 3.10) to compute OLS forecasts of the

variable ,+1
.
To sum up, the different models, and its variations, that we estimate and compare in
terms of forecasting with the baseline random walk in this study can be summarized as:
1. Autoregressive (AR) model.
2.
Large-scale DFM
2.1. Large-scale DFM with idiosyncratic component.
2.2. Large-scale DFM without idiosyncratic component.
3. Small-Scale DFM
3.1. Small-Scale DFM with idiosyncratic component.
3.2. Small-Scale DFM without idiosyncratic component.
13
For a discussion of dynamic factor models and its estimation methods, see for instance Stock and Watson (2011).
75
3.3. Data description and empirical strategy.
We use 44 monthly non-fuel commodity price series from the International Monetary
Fund database (IMF IFS). In accordance with the increase in the co-movement in nonenergy commodity prices found in Poncela, et al. (2013), we began our sample in
January 2004 and finished in December 2013.
We include in our study the raw
materials available in the following categories: cereals, meat and seafood, beverages,
vegetable oil and protein meals, agricultural raw materials and metals. A summary of
the commodities and their categories is shown in appendix A.
Figure 3.1 presents the non-energy commodity prices per category from January 1980 to
December 2013. The starting date of our sample, January 2004, is marked with a
vertical line in all plots. Our sample is characterized by a great upsurge in several of the
non-energy commodity prices until mid-2008, and a drastic decline during the global
financial crisis. After mid-2009, prices began to recover the upswing in several of the
categories, being remarkable: agricultural raw material, cereals and metals. Notably, if
we compare both the pre-2004 and post-2004 samples, there is an increase in the scale
of the boom and bust cycles for industrial inputs such as agricultural raw material and
metals, and edibles such as cereals, vegetable oils and protein meals.
76
Figure 3.1: Non-energy commodity prices per category (2005=100, in terms of U.S.
dollars).
400
500
Metals
Cereals
350
400
300
250
300
200
200
150
100
100
50
0
0
80
82
84
86
88
90
92
94
96
98
Rice
Maiz
500
00
02
04
06
08
10
12
80
82
84
86
88
Barley
Wheat
90
92
94
96
Aluminum
Zinc
Uranium
360
Agricultural raw materials
98
00
02
04
Copper
Nickel
06
08
10
12
Tin
Lead
Vegetable oils and protein meals.
320
400
280
240
300
200
160
200
120
80
100
40
0
0
80
82
84
86
88
90
92
94
96
Hard Logs
Soft Sawnwood
Wool, fine
98
00
02
Soft Logs
Cotton
Rubber
04
06
08
10
12
80
82
Hard Sawnwood
Wool, coarse
Hides
84
86
88
90
Soybeans
Palm oil
Olive Oil
300
350
Meet and seaf ood
92
94
96
98
00
02
Soybean Meal
Fishmeal
Groundnuts (peanuts)
04
06
08
10
12
Soybean Oil
Sunflower Oil
Rapeseed oil, crude.
Bev erages
300
250
250
200
200
150
150
100
100
50
50
0
80
82
84
86
88
90
92
94
96
Beef
Swine (pork)
Fish (salmon)
98
00
02
04
06
08
Lamb
Poultry (chicken)
10
12
0
80
82
84
86
88
90
92
94
Cocoa beans
Coffee, Robusta
96
98
00
02
04
06
08
10
12
Coffee, Other Mild Arabicas,
Tea
Source: International Monetary Fund, IMF
Table 3.1 shows the descriptive statistics of the non-energy commodity price inflation
for the period 2004:1-2013:12. Average inflations of non-energy commodities over the
considered period are mostly positive, only three commodities have negative nominal
average inflation (nickel, olive oil and lamb). The largest mean inflations correspond to
77
metals such as copper and tin, with 0.98% and 1.10% per month, respectively. The
biggest values of volatility also coincide with the metal category: nickel, copper and
lead reports the greatest volatilities. Other commodities that exhibit large volatilities
are: rubber, sunflower oil and swine (pork). The rest of the descriptive statistics show
greater kurtosis than the normal distribution in all commodity price inflation while
skewness can be either positive or negative. Finally, as it can be seen in bottom of table
3.1, the serial correlation term suggests first order autocorrelation is present in most of
commodity prices, which justifies the first lagged term in equation (3.2).
Table 3.1: Summary statistics for non-energy commodity price inflation
Agricultural
raw
materials
Hard Logs
Soft
Logs
Mean (%)
Std. Dev. (%)
Skewness
Kurtosis
AR(1)
0,317
3,370
0,032
3,927
0,332
0,146
6,352
0,304
3,803
0,362
0,421
2,247
-0,404
4,781
0,133
Veg. Oil and
Prot. Meal
Soybeans
Soybean
Meal
Mean (%)
Std. Dev. (%)
Skewness
Kurtosis
AR(1)
0,455
6,982
-0,710
5,135
0,334
Cotton
Wool,
coarse
Wool,
fine
Rubber
Hides
0,065
5,823
0,364
5,936
0,330
0,144
6,898
-0,623
6,606
0,401
0,556
6,002
-0,168
5,817
0,345
0,495
5,954
0,244
4,397
0,333
0,588
8,970
-1,061
6,159
0,267
0,325
7,498
-2,689
27,844
0,199
Soybean
Oil
Palm oil
Fishmeal
Sunflower
Oil
Olive
Oil
0,551
7,725
-0,663
5,030
0,316
0,286
6,235
-0,561
4,615
0,361
0,423
7,699
-0,788
5,810
0,433
0,702
5,181
1,297
8,056
0,297
0,414
10,398
2,384
21,403
0,417
-0,166
4,251
1,175
6,786
0,261
Aluminum
Copper
Tin
Zinc
Nickel
Lead
Uranium
0,989
8,112
-0,892
6,941
0,441
Coffee,
Arabicas
1,103
7,508
-0,317
3,185
0,279
Coffee,
Robusta
0,587
7,847
-0,483
4,096
0,327
-0,016
9,864
-0,340
3,968
0,299
0,942
9,204
-0,762
4,048
0,251
0,821
7,233
-0,430
5,732
0,454
Beverages
0,061
5,823
-0,608
4,222
0,308
Cocoa
beans
Mean (%)
Std. Dev. (%)
Skewness
Kurtosis
AR(1)
Cereals
0,450
5,784
-0,153
3,328
0,204
Rice
0,558
6,143
0,315
2,971
0,113
Barley
0,777
5,817
-0,051
3,216
0,217
Maiz
0,150
7,641
0,071
3,593
0,123
Wheat
Mean (%)
Std. Dev. (%)
Skewness
Kurtosis
0,684
6,899
2,470
16,131
0,366
7,165
-0,364
5,316
0,473
7,086
-0,210
4,559
0,472
7,268
0,476
5,027
Metals
Mean (%)
Std. Dev. (%)
Skewness
Kurtosis
AR(1)
Hard
Soft
Sawnwood Sawnwood
Ground- Rapeseed
nuts
oil
0,739
5,307
0,050
6,061
0,288
0,377
6,047
-0,265
4,743
0,235
Tea
78
AR(1)
0,521
0,295
Lamb
0,226
Swine
(pork)
0,225
Poultry
(chicken)
Meat seafood
Beef
Fish
(salmon)
Shrimp
Mean (%)
Std. Dev. (%)
Skewness
Kurtosis
AR(1)
0,473
4,137
-0,113
8,249
0,176
-0,268
3,292
-0,285
4,009
0,473
0,399
8,309
0,036
3,223
0,069
0,344
1,370
-0,040
2,978
0,749
0,775
7,492
-0,241
3,477
0,252
0,371
3,651
0,707
9,059
0,373
Once the different model alternatives have been exposed in the previous section, our
procedure in analyzing the data is as follows:
1. Following Stock and Watson (2011), we log differentiated and standardized
commodity price data, prior to the factor extraction either by Principal
Components or Kalman filter.
2. We determine the number of factors of the large-scale DFM using information
criteria proposed by Bai and Ng (2002). For small factor models we analyze the
eigenstructure of their variance-covariance matrix to determine the number of
factors.
3. We estimate the DFM through both principal components and the Kalman
filter.
4. We generate one-step-ahead forecasts. We start our out-of-sample forecasts
in 2010:12, re-estimate the models adding one data point at the time. In other
words, we use an expanding window. The evaluation period is 2011:01–
2013:12.
5. We compute the RMSE for each model to assess its forecasting performance.
6. We compare the RMSE of every model with that of the random walk.
79
3.4. Empirical results
In this section we evaluate the predictive content of co-movement for non-energy
commodity inflations. In particular, we examine whether joint movements of
commodity prices can be used as predictors of commodity price inflations.
Before estimating the above mentioned different factor approaches, we confirm the
presence of only one factor, which we call co-movement, in both large-scale DFM and
small-scale DFMs14.
Results for the AR model, large-scale factor models as well as the small-scale models
are presented in table 3.2. The table compares the forecasting results in terms of the
ratio of the RMSE of every model over the RMSE of the random walk forecast. Hence,
a ratio less than one means that the model improves the benchmark forecast, while
values above one suggest the opposite. We evaluate the statistical significance of the
out-of-sample predictability results using the test statistics proposed by Diebold and
Mariano (1995).
The first column of table 3.2 shows the RMSE ratios of the AR model, while the second
and third columns show the RMSE ratios of the large-scale factor model without and
with idiosyncratic component respectively. The forecasting results for the AR model
report that for 43 commodities the model improves the random walk; 24 of these
improvements were significant at the 10 percent level according to the test of Diebold
and Mariano (1995). In addition, for 38 commodities the large-scale DFM beat the
random walk forecasts, although only for 16 of these the differences between both
models are significant. We did not find any differences, in terms of number of
14
In the Kalman filter specification, we allow one autoregressive lag in the state-space transition
equation.
80
commodities that outperform the random walk predictions, between large-scale factor
model forecasts that take into account the idiosyncratic component of the factor analysis
and models that do not.
With regard to the small-scale factor models the results are more encouraging. Columns
4 and 5 in table 3.2 report forecasting performance of these models relative to the naïve
random walk model. First and foremost, the small-scale factor models provide better
predictions than the large-scale DFM approach and the AR model in most of the
commodities within the categories beverages, vegetable oils and protein meals, as well
as agricultural raw materials and metals, which means that co-movements by
commodity category added power to predictions. In this categories, RMSE outperform
by far both the AR and the random walk specification in most of commodities.
Predictability results of the small-scale factor models for cereals and meat and seafood
are mixed.
Table 3.2: Ratios of the RMSE of the univariate autoregressive (AR) model, large-scale
and small-scale factor models over the RMSE of the random walk model for the period
of analysis 2004:1-2013:12.
(AR)
model
Large-Scale
DFM
(without
idiosyncratic
component)
Small-Scale
Large-Scale
DFM
DFM (with
(without
idiosyncratic
idiosyncratic
component)
component)
Small-Scale
DFM (with
idiosyncratic
component)
Cereals
Rice
0.872*
1.002
1.004
1.634*
1.641*
Barley
0.956
1.094
1.103
1.565
1.565
Maiz
0.785*
0.799
0.802
0.723*
0.723*
Wheat
0.754*
0.756
0.757
0.755*
0.754*
Meet and seafood
Beef
0.785
0.819
0.819
1.020
1.025
Lamb
1.042**
1.356**
1.357*
1.410*
1.406*
Swine (pork)
0.716*
0.709
0.709
0.298*
0.301*
Poultry (chicken)
0.934
1.267
1.267
6.666*
6.669*
RMSPE Model/RMSPE random
walk.
81
Fish (salmon)
0.800*
0.804
0.805
0.303***
0.302***
Shrimp
0.971
1.155
1.155
1.154
1.147
Beverages
Cocoa beans
0.825
0.856
0.858
0.674*
0.674*
Coffee, Other Mild Arabicas,
0.820*** 0.832**
0.835**
0.630*
0.631*
Coffee, Robusta
0.788**
0.787**
0.792**
0.746*
0.749*
Tea
0.891*
0.927*
0.928*
0.928
0.926
Vegetable oils and protein meals
Soybeans
0.810
0.846
0.849
0.561**
0.562**
Soybean Meal
0.808
0.837
0.838
0.437***
0.440***
Soybean Oil
0.800*
0.834
0.843
0.724*
0.724*
Palm oil
0.807**
0.814*
0.820*
0.465***
0.463***
Fishmeal
0.842**
0.899
0.899
0.520**
0.517**
Sunflower Oil
0.806**
0.812***
0.818**
0.609*
0.606*
Olive Oil
0.797
0.796
0.797
0.480**
0.482**
Groundnuts (peanuts)
0.782
0.681
0.681
0.338***
0.341***
Rapeseed oil, crude.
0.762*
0.787*
0.795*
0.792*
0.791*
Sugar, bananas and orange
Sugar, European import Price
0.761*
0.723
0.725
5.770***
5.771***
Sugar, Free Market
0.812**
0.841***
0.841**
1.488
1.488
Sugar, U.S. import price
0.873
0.966
0.966
2.601**
2.601**
Bananas
0.795**
0.784**
0.788**
2.526**
2.526**
Oranges
0.807**
0.814*
0.820*
0.732
0.732
Agricultural raw materials
Hard Logs
0.932
1.046
1.046
1.470
1.470
Soft Logs
0.515*** 0.578***
0.578***
0.431***
0.432***
Hard Sawnwood
0.779*
0.788
0.790
2.207
2.207
Soft Sawnwood
0.711**
0.680**
0.680**
0.928
0.925
Cotton
0.833
0.854
0.857
0.550**
0.549**
Wool, coarse
0.813
0.843
0.845
0.696*
0.697*
Wool, fine
0.840
0.902
0.906
0.784
0.785
Rubber
0.788
0.795
0.803
0.521***
0.522***
Hides
0.710
0.666
0.667
0.666
0.669
Metals
Aluminum
0.761*** 0.736**
0.740**
0.640
0.641**
Copper
0.820*
0.856
0.865
0.749
0.750*
Tin
0.787
0.795
0.800
0.457
0.457***
Zinc
0.752*** 0.697***
0.699***
0.527
0.531***
Nickel
0.789
0.818
0.514
0.513***
Lead
0.749*** 0.732**
0.737**
0.509
0.513***
Uranium
0.899
1.086
1.083
1.067
0.806
1.081
Notes: This table reports the ratio of the root mean square error of prediction of the models, to the root mean square
error of prediction of the random walk model, RMSPEModel/RMSPErandom walk. Values smaller than one indicate that
the model perform better than the random walk. We compute the Diebold-Mariano (1995) test statistic for the null
hypothesis that the corresponding MSE differential is zero.
*** Indicates statistical significance at the 1% level.
** Indicates statistical significance at the 5% level.
* Indicates statistical significance at the 10% level.
82
3.5. Robustness checks
As a robustness check we estimate all models for a prior period (1992:2-2003:12) and
compare their forecasting performance, in terms of RMSE ratios, with the second period
in order to assess whether commodity prices were more or less predictable in different
subsamples. Results for the AR model, large-scale factor models as well as the smallscale models for the first period are presented in appendix B.
The methodology for estimating each model is the same as described in previous
sections. Regarding the first period, we begin the out-of-sample forecasts in 2000:12,
therefore, the evaluation period is 2001:1-2003:12. We follow the same procedure
explained in section 3, and use an expanding window with a size of 36 months. We
compare the predictive content for every model i in both of the periods (pre-2004 and
post -2004) by means of the following difference:
(


)
−2004
− (


)
−2004
(3.9)
When there is a value over one in the above difference, the predictive content of the
model is enhanced in the second period, given that the RMSE against the random walk
model is lower in the post-2004 period compared with the pre-2004 period.
In general, small-scale models both with and without idiosyncratic component,
performed better in the second period. In the post-2004 period, compared to pre-2004,
the small-scale models with and without specific component improved their prediction
83
versus random walk in 28 commodities. Importantly, the ratio between the RMSE of
the small-scale DFM to the RMSE of the random walk reduced more for commodities
in the categories of vegetables and protein meal, agricultural raw materials, and metals.
These results suggest that the increase in the overall movement of non-energy
commodity prices since 2004 has causes the small-scale dynamic factor models to
improve their predictive content.
Regarding the AR model, as well, as the large-scale DFM we found inconclusive results
since approximately half of commodities´ prediction improved with these models and
half worsened for the second period.
In addition to the comparison between periods of each model, we perform an analysis of
the different models in each period. That is, we compare the forecasting performance
among models, in terms of their RMSE ratios, in order to assess which model has the
best behavior in each period. Specifically, we compare the predictive capability between
models i and j for each period by means of the following difference:

(
)

,
−2004
(3.10)

(
)

−2004
Values above one mean that the model j outperforms the model prediction of i, since j
has a RMSE lower than RMSE of the model i, and vice versa. The results can be
summarized as:
84
a. In both the first and the second periods, the AR model outperforms the large
scale factor models in its ability to predict changes in the prices of non-energy
commodities. The ratios of the RMSE of the large scale models to the RMSE of
the AR model are lower than one in approximately 33 to 37 commodities.
b. Results show that the large-scale DFM model without idiosyncratic component
performs better, in terms of predictability, than the large-scale DFM with
idiosyncratic component for both periods.
For the first period, for 32
commodities, the large-scale DFM without idiosyncratic component beats the
model with this element, while for the second period, it does so for 38
commodities.
c. Regarding the small-scale models, in both periods the model without
idiosyncratic component reports lower ratios to the small-scale DFM with
idiosyncratic component. For the first period, for 31 commodities, the smallscale DFM without idiosyncratic component outperforms, in terms of
forecasting performance, the small-scale DFM with idiosyncratic component.
For the second period, it does so for 25 commodities.
d. While comparing the AR model with the small-scale DFM, we found interesting
results. For the first period, say pre-2004, the ratios of the RMSE of the smallscale models over the RMSE of the AR model forecast were lower than one for
18 commodities (for both models with and without idiosyncratic component). In
contrast, for the second period, say post-2004, the small-scale models increased
their predictive content to overcome the AR model in 27 commodities for both
models with and without idiosyncratic component).
e. As in the previous point, we found an increase in the predictability of smallscale DFM in the second period in comparison to the large-scale DFM. That is,
85
pre-2004, the large-scale model beats the small-scale DFMs in 23 to 24
commodities.
In contrast, for the post-2004 period, the small-scale DFM
outperform the large-scale DFM in 30 commodities. Therefore it is only the
common part what is informative for forecasting commodity returns.
These results reaffirm the increase in the predictive content of the small-scale factor
models compared with both the autoregressive model and the large-scale DFM for the
second period. In addition, the results in both forecasting sample highlight that for
DFM, the inclusion of the idiosyncratic component does not improve the forecasting
performance of these models in relation to the naïve random walk model. Therefore, it
is only the common part what is informative for forecasting commodity returns.
3.6. Conclusions
To understand and predict changes in commodity prices it is important not only for
commodity dependent countries, due to the fact that commodity price swings directly
affect their term of trade and cycle, but also for commodity importing countries,
because commodity prices impact inflation and may interfere with monetary policy
goals.
We examine the predictability of non-energy commodity price changes when we take
into account the co-movement of either a large range of commodities, or the comovement within a specific category of raw material prices. We use a dynamic factor
model approach and estimate the communalities of non-energy commodity price
inflations either by Principal Components, for the case of large-scale factor, or Kalman
Filter, in the case of the small-scale (category) factor.
86
We found that co-movement in extensive data of commodity prices has poorer
predictive power over non-energy commodity prices since 2004 when comparing to the
small-scale factor models and univariate AR model. Conversely, communalities into
categories such as oils and protein meals, as well as metals seem to substantially
improve the forecasting performance of the random walk model. For these categories
we found reductions in the RMSE up to 50%. In the robustness checks, we found that
small-scale DFM has gained predictive power since 2004. In fact, in the previous
period, say 1992:2-2003:12, the predictability of small and large-scale factor models
were similar.
Finally, adding the forecast of the idiosyncratic component did not
improve the results and, therefore, it is only the common part what is valuable for
forecasting.
Before 2004 non-energy commodity prices were quite stable, especially for industrial
inputs such as agricultural raw material and metals, and edibles such as cereals,
vegetables oils and protein meals. On the contrary, after 2004, assets allocated to
commodity indices increased, leading to the so-called financialization in the commodity
markets. This new feature generates not only greater synchronization among
commodities (co-movement), but also introduces higher levels of uncertainty to the
market. In this paper we have studied the predictive power of co-movement in nonenergy commodity prices, further work should include the recent role of uncertainty in
commodity markets as a possible source of predictability.
87
References
Alquist, R. and Kilian, L. (2010) What do we learn from the price of crude oil futures?,
Journal of Applied Econometrics, 25, 539–573.
Bai, J. and Ng, S. (2002) Determining the Number of Factors in Approximate Factor
Models, Econometrica, 70, 191-221.
Büyüksahin, B. and Robe, M.A. (2012) Speculators, commodities and cross-market
linkages ideas, doi: 10.2139/ssrn.1707103.
Camacho, M and Perez-Quiros, G. (2010) Introducing the euro-sting: Short-term
indicator of euro area growth, Journal of Applied Econometrics, 25(4), 663-694.
Chen, Y., Rogoff, K., & Rossi, B. (2010) Can exchange rates forecast commodity
prices? Quarterly Journal of Economics, 125, 1145–1194.
Chinn M.D. and Coibion, O. (2013) The Predictive Content of Commodity Futures, The
Journal of Futures Markets, doi: 10.1002/fut.21615.
Diebold, F. and Mariano, R.S. (1995) Comparing Predictive Accuracy, Journal of
Business & Economic Statistics, 13(3), 253-263.
Gargano, A. and Timmermann, A. (2014) Forecasting commodity price indexes using
macroeconomic and financial predictors, International Journal of Forecasting, doi:
10.1016/j.ijforecast.2013.09.003
Groen, J. J. J., & Pesenti, P. A. (2011). Commodity prices, commodity currencies, and
global economic developments. In T. Ito, & A. Rose (Eds.), Commodity prices and
markets, East Asia Seminar on Economics, Vol. 20 (pp. 15–42). Chicago:
University of Chicago Press.
Henderson, B., Pearson, N., and Wang, L. (2012) New Evidence on the Financialization
of Commodity Markets, Working Paper, University of Illinois, Urbana-Champaign.
88
Irwin, S.H., Sanders, D.R. and Merrin, R. (2009) Devil or angel? The role of
speculation in the recent commodity price boom (and bust), Journal of Agricultural
and Applied Economics, 41, 2.
Poncela, P., Senra, E. and Sierra, L. (2013) The dynamics of co-movement and the role
of uncertainty as a driver of non-energy commodity prices, Documento de Trabajo
No. 733, Fundación de las Cajas de Ahorros, FUNCAS, Madrid.
Stock, J. and Watson, M. (2011) Dynamic Factor Models, in Clements MJ, Hendry DF
ed., Oxford Handbook on Economic Forecasting, Oxford University Press, Oxford.
Tang, K., and Xiong, W. (2012) Index investment and the financialization of
commodities, Financial Analysts Journal, 68, 54-74.
89
Appendix A.
Non-energy commodity prices
IMF Category
Edibles
Cereals
Meet and seafood
Commodity
Rice
Maiz
Barley
Wheat
Beef
Poultry (chicken)
Lamb
Fish (salmon)
Swine (pork)
Shrimp
Beverages
Cocoa beans
Coffee, Other Mild
Arabicas,
Coffee, Robusta
Vegetable oils
Soybeans
Sunflower Oil
and protein meals
Soybean Meal
Olive Oil
Groundnuts
(peanuts)
Rapeseed oil,
crude.
Soybean Oil
Palm oil
Tea
Fishmeal
Other Edibles
Sugar EU.
Oranges
Sugar US.
Industrial
Agricultural
Bananas
Hard Logs
Soft Logs
Inputs
raw materials
Hard Sawnwood
Rubber
Soft Sawnwood
Hides
Cotton
Wool, coarse
Aluminum
Zinc
Tin
Nickel
Uranium
Lead
Metals
Wool, fine
Tin
90
Appendix B.
Ratios of the RMSE of the the univariate autoregressive (AR) model, large-scale and
small-scale factor models over the RMSE of the random walk model for the first period
of analysis (1992:2-2013:12).
(AR)
model
Large-Scale
DFM
(without
idiosyncratic
component)
Small-Scale
Large-Scale
DFM
DFM (with
(without
idiosyncratic
idiosyncratic
component)
component)
Small-Scale
DFM (with
idiosyncratic
component)
Cereals
Rice
0.787*
0.8055
0.8057
1.515*
1.518*
Barley
0.8837
1.0290
1.0321
1.2524
1.2531
Maiz
0.808*
0.8382
0.8423
1.2262
1.2268
Wheat
0.8178
0.8438
0.8450
0.8387
0.8383
Meet and seafood
Beef
0.8508
0.9003
0.9003
0.8226
0.8251
Lamb
0.744*** 0.703**
0.703**
0.7629
0.7644
Swine (pork)
0.752*
0.7559
0.7558
0.356***
0.357***
Poultry (chicken)
0.9027
1.1232
1.1235
4.347***
4.350***
Fish (salmon)
0.826*
0.8704
0.8701
0.6469
0.6500
Shrimp
0.8298
0.8614
0.8615
0.8626
0.8627
Beverages
Cocoa beans
0.8039
0.8181
0.8181
0.537**
0.537**
Coffee, Other Mild Arabicas,
0.749*** 0.750**
0.750**
0.662**
0.665**
Coffee, Robusta
0.766**
0.769**
0.769**
0.565***
0.564***
Tea
0.750*
0.749*
0.749*
0.750*
0.749*
Vegetable oils and protein meals
Soybeans
0.8535
0.8975
0.8997
0.9837
0.9849
Soybean Meal
0.8638
0.9037
0.9052
0.9742
0.9773
Soybean Oil
0.801*
0.8196
0.8213
0.7842
0.7836
Palm oil
0.790**
0.800*
0.802*
0.493***
0.490***
Fishmeal
0.810*
0.8386
0.8392
1.789**
1.791**
Sunflower Oil
0.798**
0.822*
0.823*
0.610**
0.608**
Olive Oil
0.8750
0.9302
0.9307
1.920***
1.918***
Groundnuts (peanuts)
0.9148
1.0494
1.0492
1.773**
1.772**
Rapeseed oil, crude.
0.790*
0.782*
0.785*
0.770*
0.769*
Sugar, bananas and orange
Sugar, European import price
0.788*
0.8095
0.8100
1.380***
1.382***
Sugar, Free Market
0.730**
0.695**
0.695**
1.370*
1.373*
Sugar, U.S. import price
0.8751
0.9555
0.9556
1.894***
1.895***
Bananas
0.699**
0.694**
0.695**
0.613**
0.619**
Oranges
0.790**
0.800*
0.802*
0.8346
0.8251
Agricultural raw materials
Hard Logs
0.9144
1.0898
1.0898
1.9059
1.9092
Soft Logs
0.476*** 0.567***
0.566***
0.5902
0.5861
RMSPE Model/RMSPE random
walk.
91
Hard Sawnwood
0.783*
0.8024
0.8021
2.1566
2.1591
Soft Sawnwood
0.494**
0.572**
0.572**
0.3779
0.3787
Cotton
0.8929
1.0721
1.0737
1.0255
1.0304
Wool, coarse
0.8874
0.9286
0.9286
0.8678
0.8680
Wool, fine
0.8314
0.8583
0.8588
0.6635
0.6645
Rubber
0.8112
0.8331
0.8333
0.7197
0.7201
Hides
0.9247
1.0384
1.0381
1.0388
1.0324
Metals
Aluminum
0.741*** 0.745**
0.745**
0.8209
0.8210
Copper
0.837*
0.8904
0.8913
0.8865
0.8868
Tin
0.8649
0.8942
0.8939
0.629*
0.629*
Zinc
0.742*** 0.732***
0.733***
0.704**
0.704**
Nickel
0.8526
0.9194
0.458***
0.459***
Lead
0.752*** 0.743**
0.744**
0.539***
0.536***
Uranium
0.8780
0.9279
0.9363
0.9304
0.9184
0.9278
92
Chapter 4.
Long-term links between raw materials
prices, real exchange rate and relative deindustrialization in a commodity
dependent economy. Empirical evidence
of “Dutch disease” in Colombia
4.1. Introduction
The rise of the raw material sector can generate an increase in the national income and
improvements in the balance of payments. However, it may also affect in a negative
way the production of the export manufacturing sector. Dutch disease is frequently
understood as the de-industrialization process of an economy, which is associated to the
93
real exchange rate appreciation, produced as a consequence of an export windfall due to
a resource discovery or a raw material export boom.
Lately there is an open debate in Colombia about the possible symptoms of Dutch
disease due to the rise in the energy sector and the deep appreciation of the exchange
rate. Colombia has been enjoying and at the same time suffering both phenomena over
the last decade. High oil prices and flexible petroleum legislation in relation to the
exploration of oil wells in 2003 have made Colombia a main exporter of petroleum
today15. The post-2003 period coincides also with a drastic real exchange rate
appreciation, one of the highest in the world16.
Concern about the existence of
symptoms of the disease in Colombia is, therefore, a valid issue.
The Colombian commodity export dependence, and its possible effect on deindustrialization of the country due to the appreciation of the exchange rate, motivates
this research. Even though crude oil has not always been well ranked in exports,
Colombia has historically been dependent on raw materials. The country experienced a
bonanza in the prices of coffee, the former main export product in the mid-seventies.
The coffee boom generated great resources from the exports, but it also affected the real
exchange rate, which motivated similar debates about de-industrialization (see, e.g.,
Kamas, 1986 and Meisel, 1998).
This paper seeks to explore the evidence of symptoms of Dutch disease in Colombia in
the long term. The Colombian economy presents specific features regarding this issue
that makes this analysis distinctive. First, it has experienced two booms associated to
two different commodities in two different periods of time: coffee and oil. Because of
15
In 2012, petroleum and its derivatives represented 53% of the total exports of Colombia, (National
Administrative Department of Statistics-DANE).
16
Between the years 2003 and 2011, the real exchange rate appreciated 52%, the seventh highest
appreciation of the 95 countries the World Bank reports on.
94
that, it seems that the threat of Dutch disease in Colombia has been present along the
years and makes relevant the analysis of the long term. As far the authors’ knowledge,
there is no econometric approach on Dutch disease in this country with a long enough
span of data that covers the two historical export booms that Colombia has experienced.
Notably, most papers center in the coffee boom (e.g. Puyana, 2000, Meisel 1998,
Edwards, 1984 and Kamas, 1986) and do not examine the long term effects of a
commodity windfall. More specifically, they do not analyze whether the adverse effects
associated to Dutch disease offset the beneficial effects of commodity upsurge. Looking
at the most recent export boom, associated with petroleum, the long term study of Dutch
disease in Colombia is also crucial when analyzing the post-2003 appreciation trend in
the country, jointly with an increase of 215% in real oil prices, Colombian main export
product in that period. Monetary policy might be different if we expect that the upward
trend observed since 2003 will not change in the short or medium term. Our paper
focuses on the impact that the real prices of the two main commodities (coffee,
petroleum) exported by Colombia in the period between 1972 and 2011, have had on
relative manufacturing output, through the real exchange rate.
Second, there is no methodological consensus on how to research this phenomenon.
Usually, articles are divided into two types. On the one hand, those that work with a
large collection of cross-section data or short panels and use static or dynamic panel
models; see Coudert, Couharde and Mignon (2011), Lartey (2007, 2011), Acosta and
Mandelman (2009), and Kang and Lee (2011), and on the other, authors who use time
series to find the relation between real exchange rate and several macroeconomic and
financial variables; Benedictow, Fjærtoft, and Løfsnæs (2013), Beine, Bos and
Coulombe (2012), Algieri (2011) and Egert (2008), among others. It is worth
emphasizing that the country specific details related to Dutch disease are unexplored in
95
cross-section studies. Furthermore, data and types of time series techniques differ
according to the country of study.
Finally, although this paper makes no in-depth study of the policy implications, if the
symptoms of Dutch disease were confirmed, policies should be oriented toward
preventing industry dissolution and toward a better use of the resources coming from
commodities. How to spend the revenues from the oil boom is crucial, not only to avoid
Dutch disease effects, but also to take better advantage of the benefits associated with
the positive income shock.
We estimate a Vector Error Correction Model (VECM) to explore the evidence of
Dutch disease in Colombia, which enables us to find equilibrium relations among
variables in the long term, based on the cointegration evidence. The cointegration
concept, introduced by Engle and Granger (1987) and extended by Johansen (1991),
consists on determining whether two or more variables come together in the long term,
and how deviations may affect the short term.
We consider annual series between 1972 and 2012 to determine whether commodity
prices are related to the real exchange rate and the ratio between manufacturing to
services output in Colombia in the long term. We also take into account variables such
as productivity, government expenditure, degree of trade openness and international
inflows.
The rest of this article is organized as follows. Section two focuses on the description
and definition of Dutch disease symptoms and offers an initial analysis of some of the
key variables related to Dutch disease in Colombia. The econometric model is
developed in section three. The last section offers conclusions and recommendations for
economic policy.
96
4.2. Dutch disease and its symptoms in Colombia
The phenomenon known as Dutch disease refers to the negative effects on the
manufacturing sector due to a large increase in a country income. The term was first
used in an article in The Economist in 1977 to describe The Netherlands experience of
discovering large gas deposits in the North Sea and their harmful effects on the
manufacturing sector of the country.
The sudden increase of that country’s wealth created an unprecedented inflow of
capital, which produced an appreciation of its currency and, therefore, a loss of
competitiveness of the non-gas exporting sectors.
The discovery of natural resource deposits and the increase in the price of these
resources in the producing countries create great advantages, such as an increase in the
country’s wealth, greater fiscal income for social investment and an improvement in the
balance of payments. However, the de-industrialization resulting from the loss of
competitiveness may produce a high level of specialization in the production of the
resource and in non-tradable sectors, which may leave the rest of the economy more
vulnerable to external shocks with regard to international prices.
Corden and Neary’s (1982) paper was the first to analyze the de-industrialization
phenomenon produced by the boom of a sector which had been traditionally extractive
from a theoretical view. The authors divide the economy into three sectors. Two of
these produce internationally tradable goods. The first is the booming extractive sector
(BS); like gas in The Netherlands, oil in Venezuela and minerals in Australia. The
second is traditional manufacturing (MS), and the third is a non-tradable sector such as
services and construction (NTS). The Dutch disease will result in a contraction in the
traditional export sector (MS), through spending and resource movement effects.
97
In the spending effect, the competitiveness of the manufacturing sector is noted due to
the real appreciation of the national currency, through the increase of either the nominal
exchange rate or the national prices. To explain this aspect more clearly, let´s suppose
that the country is currently witnessing a growth period in the export booming sector,
which generates a huge influx of foreign currency. Supposing that the country has a
flexible exchange rate regime, the windfall will generate appreciation of the local
currency. On the other hand, the resource boom generates greater income to the
government, through taxes and royalties, as well as directly to the owners of the factors.
The subsequent increase in the demand drives up the prices of non-tradable goods17.
Both the nominal appreciation of the local currency and the increase in local prices,
generate a real appreciation of currency and a contraction in the traditional
manufacturing sector (de-industrialization).
At a theoretical level, extensions of the Corden and Neary (1982) models have been
developed and they illustrate the spending effect, including or modifying assumptions to
the core model. Van der Ploeg (2011), for instance, takes the Salter (1959) and Swan
(1960) model to illustrate, the effects of a resource boom in the overall welfare of the
country. Other extensions are Neary (1988), who widens the model considering intersectoral duality; Neary and Purvis (1982), who evaluate the effects of a BS windfall
taking into account a tradable sector which is intensive in capital; and Morshed and
Trunovsky (2004) who create a dynamic model with adjustment costs in investment
allowing free movement of capital among tradable and non-tradable sectors. In any of
these models, the final effect is an increase in the relative prices of non-tradable goods,
which leads to the expansion of the NTS and a contraction of the MS. In contrast, Buiter
17
At this point two clarifications are necessary. First: the increase in the general price level occurs due to
the rising prices of the non-tradable goods sector, as prices in the tradable sector are internationally set.
Second: the elasticity of demand for non-tradable goods with respect to income must be positive to allow
an increase in their demand.
98
and Purvis (1983) found, in a theoretical approach, that in the long run, manufacturing
output responds in a positive way to oil shocks when the country is small and a net oil
exporter. However, increases in the price of oil can have transitional negative effect.
Besides the spending effect discussed so far, there are also resource movement effects
arising from natural resource boom (Corden and Neary, 1982). The de-industrialization
occurs because of the usual appreciation of the real exchange rate (the spending effect),
and also by the movement of labor from the manufacturing and the non-tradable sectors
towards the resource sector (the resource movement effect). According to Corden
(1984) if nominal wages are rigid this might increase unemployment. The factor
reallocation among sectors has been dynamically modeled by Krugman (1987), Bruno
and Sachs (1982), and Chaterji and Price (1988). Krugman (1987) introduce economies
of scale to the core model. Bruno and Sachs (1982) simulate an infinite-horizon
economy to search for the transitional dynamics of Dutch disease shocks, and Chaterji
and Price (1988) include long-run unemployment effects.
Other authors, e.g. Egert (2008), consider that the boom in an extractive sector can also
encourage both national and international investment in that sector. The capital inflow
to the country energy sector, as a form of foreign direct investment, can accentuate the
real exchange rate appreciation. Lartey (2008) finds that the relative de-industrialization
can be produced by the capital inflow to the country, even if not necessarily related to
the extractive sector. In other words, there is empirical evidence about the relation
between real exchange rate appreciation and variables such as, foreign direct
investment, the level of financial liberalization, remittances and foreign aid; see Guha
(2013), Lartey (2011), Acosta and Mandelman (2009) Arellano et al. (2005), Prati and
99
Tressel (2005), and Nkusu (2004)18. Hence, Dutch disease may be triggered by deposits
discovered in the energy sector, as a result of variations of international prices of
extractive products or commodities, or as a result of capital inflows such as foreign
direct investment, remittances and international aid.
To summarize, Dutch disease symptoms are mainly a real appreciation of local currency
and zero increase, or even the decrease, of the manufacturing sector related to that of
non-tradable goods or services (relative de-industrialization). For the later, we focus on
the spending effect on the relative manufacturing output.
There are few articles regarding the effects of a booming resource sector in Colombia
and, as mentioned before, most of them are related to the effects of coffee boom.
Edwards (1984), Kamas (1986) and Meisel (1998) report the influence that the coffee
boom had in the production of other tradable goods. Edwards (1984), in a theoretical
model, shows that a coffee export boom will generally generate a short-run increase in
money supply, inflation and real appreciation. Kamas (1986) show that most of the
conditions associated with the Dutch disease occurred in Colombia during the years
1975-198019. The author finds an increase in the production of non-tradable goods
sectors such as construction and public works, residential income and government
services. Moreover, the growth of manufacturing production decreased along that
period. However, Kamas (1986) OLS estimations were not significant when analyzing
a negative relationship between appreciation of the real exchange rate and
manufacturing sectors. On the other hand, Meisel (1998) shows analytical evidence of
Dutch disease in the first half of the twentieth century as a result of the rise in
18
Buiter and Purvis (1983) examine the relative importance of different shocks as sources of deindustrialization. Therefore, they search for the impact of an increase in oil prices, a domestic oil
discovery, and monetary disinflation, over the real exchange rate.
19
During these years the international coffee price abruptly increased as a result of frost in Brazil, which
destroyed much of the crop and reduced the production capacity of the country.
100
international coffee sales. The author argues that appreciation of the real exchange rate
was due to high coffee prices, sluggishness banana exports and limited economic
growth of the Colombian Caribbean Coast. Suescún (1997) uses a neoclassical growth
model to examine the influence of a shock in coffee prices on resource reallocation.
The author finds temporary de-industrialization and appreciation of the real exchange
rate. The effects of the oil discoveries during the nineties are shown by Puyana (2000)
in a descriptive analysis about its possible effects on agriculture and rural poverty. As
an oil exporting country, Colombia has been included in the Dutch disease cross
country studies of Davis (1995) and Ismail (2010) on the effects in development and
manufacturing, respectively. A summary table of the literature is presented in Appendix
A.
Before examining each of the symptoms of Dutch disease in Colombia, it is important
to highlight the participation of commodities in the total exports of the country. During
the seventies and eighties, the exports in this type of goods represented on average 63%
of total exports, see Figure 4.1. Only at the beginning of 1990 and 2000 fell their share
in total exports under 50%. However, since November 2002, with the increase of
international prices, their participation has begun to increase again and regained the
70% values of 1970.
101
Figure 4.1: Commodity exports as a share of total exports in Colombia, percent,
1970-2011.
.8
.7
.6
.5
.4
.3
.2
1970
1975
1980
1985
1990
COMMODITIES
1995
2000
2005
2010
OTHER
Source: National Administrative Department of Statistics, DANE, and Central Bank of Colombia .
Among commodities, the exports in Colombia are mainly concentrated in two kinds of
goods: coffee and petroleum (see Figure 4.2). In the seventies and eighties, coffee had a
participation in the exports of commodities that was always above 80%. The peak of
coffee participation was in the year 1981, when it represented 97% of the total exports
of commodities. Since that year, the importance of coffee has fallen in a dramatic way,
and petroleum has gained a place in the export of raw materials. Likewise, coal
improves its position in the basket of exported commodities.
102
Figure 4.2: Commodity groups as a share of total commodity exports in Colombia
(percentage).
1.0
Coffee
0.8
Oil
0.6
0.4
Coal
0.2
0.0
1970
1975
1980
1985
1990
COFFEE
OIL
1995
2000
2005
2010
COAL
FERRONICKEL
Source: National Administrative Department of Statistics, DANE, and Central Bank of Colombia.
Even though one could think that the relative importance of coffee, coal and petroleum
in the composition of the exports is marked by the evolution of their prices
internationally, a simple graphic analysis does not seem to show it. As can be
appreciated in Figure 4.3, the prices of the three main raw materials exported by
Colombia increased since 2002, and they seemed to move together since then. The
reason of the fall of the participation of coffee lies in a drastic reduction of the
production of coffee since 1992. On the contrary, the production of oil and coal has an
increasing trend since the mid-eighties, see Appendix B. The rise of the oil sector has
been accentuated since 2003 when conditions to explore the country became less rigid
with act 1760. With this flexibleness of the legislation, direct foreign investment
towards the petroleum sector began to rise, see Appendix C. Therefore, the country
received, on the one hand, currency from oil prices, and on the other, foreign direct
investment.
103
Figure 4.3: Commodity prices in constant US$. 2005=100
500
400
300
200
100
0
80
82
84
86
88
90
92
94
96
98
00
02
04
06
08
10
12
W TI Crude Oil Price
Coal Prices
Arabic Coffee Prices
Source: International Monetary Fund, IMF.
After analyzing the importance of raw materials for the country exports, we focus again
on the real exchange rate, one of the variables affected according to the Dutch disease
hypothesis. Figure 4.4 shows the evolution of the real exchange rate since 1972. The
color shade represents episodes of boom in the commodity exporting sector: the
increase in coffee prices (1975-1977) and the increase in oil prices (2003-2008). The red
lines identify the crude oil discoveries that made Colombia an oil export dependent
country (Caño Limón in 1983, Cusiana in 1989 and Cuapiagua in 1993). When
evaluating the real exchange rate from a historical perspective, it seems that huge
increases in the main commodity exporting prices bring appreciation of the real
exchange rate, while resource discoveries do not necessarily do so. While there was a
pronounced real devaluation of the Colombian peso during the eighties, it was followed
by different episodes of appreciation/devaluation. Therefore, the concern of the
104
Colombian non-traditional exporting sectors is due to the real exchange rate growth
since 200320, its possible upward trend and the loss of share of the country’s industrial
exports.
Figure 4.4: Colombian Real Exchange Rate. 2005 = 100
180
160
140
120
100
80
60
1975
1980
1985
1990
1995
2000
2005
2010
Source: IMF International Financial Statistics, and authors´ calculations.
Another symptom of Dutch disease is a zero increase or even the decrease of the
manufacturing sector related to that of non-tradable goods or services (relative deindustrialization). Figure 4.5 displays the contribution of the services and industry
sectors to the Gross Domestic Product (GDP). The graph shows that the largest loss in
manufacturing value relative to services took place in the early nineties. Since 1994 the
contribution of the service sector to GDP has been greater than that of the industrial
sector, though the difference seems to have stabilized since 2000. A similar assessment
can be deduced by analyzing only the portion of the industry related to manufacturing,
see Figure 4.6. Despite the real appreciation taking place since 2003, manufactures do
not seem to lose weight against the services during the first decade of the century. The
loss of the relative manufacturing output coincides with the trade liberalization policies
introduced in Colombia in the early nineties. The Colombian government in 1990
undertook an ambitious economic liberalization program. The economic openness
20
From 2003 to 2011 the real exchange rate in Colombia has grown by 53% (IMF).
105
package reduced the average trade tariff from 34.5% to 11.4% in a year. Likewise, the
effective rate of protection decreased from 60% to 26.2% at the end of 1990.
Figure 4.5: Contribution of the Sectors to the Gross Domestic Product (share).
65
60
55
Services
50
Industry
40
1965
1967
1969
1971
1973
1975
1977
1979
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
2003
2005
2007
2009
2011
45
Source: Data Service & Information. World Bank Statistics.
Figure 4.6: Ratio Manufacturing to Services. Value added in local currency.
0,6
0,55
0,5
0,45
0,4
0,35
0,3
0,25
0,2
19721974197619781980198219841986198819901992199419961998200020022004200620082010
Source: Data Service & Information. World Bank Statistics
106
4.3. Econometric modeling and empirical evidence of
Dutch disease in Colombia
To assess the empirical relation between commodity prices and the Dutch disease
symptoms in Colombia, we formulate a Vector Error Correction Model (VECM).
Through VECM we can seek long run as well as short run effects of Dutch disease, if
they exist.
The model will seek the estimation of two equations related to the main symptoms: 1)
deterioration of international competitiveness in Colombia related to the real
appreciation of national currency, 2) the relative de-industrialization related to
commodity prices. Due to the lack of higher frequency data, we consider the larger
annual span available from 1972 to 201121.
The first equation considers the real exchange rate, RER, as a measure of
competitiveness in international markets. An increase in RER or a real appreciation of
the peso implies that the prices of local goods are more expensive than those from the
rest of the world. In this sense, the competitiveness of Colombian products, compared
with those of its competitors, would be reduced. A usual way of defining the real
exchange rate is given by the following equation:

 =  ∗ ∗,
(4.1)
where the real exchange rate (RER) is represented in the equation as the prices of
domestic goods (P) relative to those from abroad (P*), adjusted by the nominal
exchange rate (S). Based on this definition, we construct a proxy for the real exchange
21
According to Otero and Smith (1999) the power to detect long run equilibrium relationships through, for
instance, Johansen cointegration tests depends more on the total sample length than on the number of
observations.
107
rate taking into account Colombian CPI and the US CPI, which is Colombia’s main
trade partner.
From an empirical point of view, the real exchange rate is related to sectorial
productivity differentials (the Balassa-Samuelson effect), terms of trade, country
openness to international trade and foreign capital inflows among other policy variables
such as government spending (see, e.g., Edwards, 1989 and Alberola, 2003). The
relationship between the real exchange rate and its determinants is expressed in equation
(4.2).
RER=f(PRODUCT,COMM,GOV, OPEN, INFLOW).
(4.2)
The relative productivity in the model, denoted by PRODUCT, is linked to the BalassaSamuelson effect and it is proxied by the ratio between the consumer price index (CPI)
to the producer price index (PPI). This approximation stems from the idea that non
tradable goods are included in the CPI but not (or not much) in the PPI. Therefore, the
Balassa-Samuelson effect, which relates productivity growth differentials to the relative
price between non-tradable and tradable goods, should be caught through this variable.
We followed in this approach the works of DeLoach (2001), Alberola(2003), BénassyQuéré et al., (2004).
Commodity prices, COMM, are especially relevant when
analyzing the Dutch disease symptoms in Colombia and can also be considered as a
proxy variable of the terms of trade22. As shown in the previous section, the relative
importance of coffee and oil in Colombian commodity exports changes over the period
between 1972 and 2011. For this reason, we constructed a commodity price index
taking into account their relative export weight in each year. With this, we aim to catch
better the size of the commodity windfall. The increase in the international commodity
22
This variable has been included recently by Coudert, Couharde and Mignon (2011), who analyze the
long term relationship between the real effective exchange rate and the terms of trade for 52 commodity
exporters and 12 oil exporters.
108
prices produces a surge of income through exports and an improvement in the terms of
trade in commodity exporting countries like Colombia. In addition, the coefficient
which accompanies this variable is expected to have a positive sign or a direct relation
with the real exchange rate, since an increase in commodity prices would cause a real
appreciation of the peso against the dollar.
The government expenditure, represented by the variable GOV, accounts for the share
of government consumption in the GDP and it is expected to affect the country`s
competitiveness positively as the government tends to consume more in non-tradable
goods. Then an expansionary fiscal policy affects the national demand pushing prices of
non-tradable up since they are the ones that are not international fixed23.
The degree of openness, OPEN, defined as the ratio of imports plus exports to GDP,
influences the long run real exchange rate negatively. Trade-liberalizing reforms that
include, for example, reductions in tariffs, lead to a decline in the domestic price of
imported goods that triggers an excess demand for imported goods and a reduction of
domestic demand for non-traded goods. As a result, the real exchange rate depreciates
to restore the equilibrium in the non-traded market.
The last variable in the equation is that related with the foreign capital inflows, denoted
by INFLOW, which is associated with real exchange rate appreciation in the long run.
According to Corden (1994) a foreign capital surge affects the economy by raising the
domestic demand for both traded and non-traded goods. On the non-traded goods
market, this excess of demand has to match a proportional increase of the non-traded
supply in order to ensure market equilibrium. This in turn leads to a rise of the price of
non-traded goods and, furthermore, to a real appreciation. We used data of private
23
According to Lartey (2007) the extent of the effect will depend on the marginal propensity to consume
non-tradable goods. .
109
capital flow from the World Bank that account for both net foreign direct investment
and portfolio investment. Detailed data descriptions are reported in Appendix D and the
graphs of the data in Appendix E.
The second equation in our model is expected to prove the relative de-industrialization
due to a surge in commodity prices,
MAN_SERV = F (RER, PRODUCT, OPEN, INFLOW,GOV).
(4.3)
We include, as a dependent variable, the ratio between Colombia manufacturing
production to services production (MAN_SERV). Hereinafter, this relation will be
referred to as relative manufacturing output. If there is, indeed, evidence of Dutch
disease in the country, it is expected that the real exchange rate, RER, will have a
negative effect on the relative manufacturing output. Increases in the real exchange rate,
a real appreciation, will cause a loss of the country competitiveness that leads to a
contraction of manufacturing exports and a contraction of manufacturing output. In
equation (4.3), relative manufacturing output is also related to productivity, openness,
private inflows and government spending. The variable productivity reflects the
outsourcing process of the developing economies. The constant increases in
productivity in these countries will eventually result in the strengthening of the services
sector in relation to the manufacturing sector. Thus, it can be assumed that productivity
has a negative sign. One of the explanations for this relation is based on the greater
exposure of economies to international trade24. We include in the last equation the
variable openness, OPEN, with the aim of testing the common view that commercial
liberalization had brought about a de-industrialization process, or if contrary to this
24
As long as the countries grow, and trade with more competitive countries in terms of manufacturing, the
prices of these products will drop on the national market. Domestic producers will be forced to improve
their productivity or withdraw from the market. The services sector is not exposed to international trade,
and so benefits to the extent that the country grows and opens its doors to international trade; see Kang
and Lee (2011).
110
view, trade openness benefits manufacturing by increasing international demand. We
also want to see if financial liberalization, proxied in our analysis by Private Inflows,
INFLOW, has positively affected the manufacturing production, as suggested by Kang
and Lee (2011). On the other hand, as government spending, GOV, tends to allocate
more towards non-tradable goods, so we can expect that an increase in this variable
benefits more the service sector, so reducing the ratio MAN_SERV.
4.3.1.
Order of integration
Before estimating the model, we perform the augmented Dickey-Fuller (ADF) unit root
tests to check the number of unit roots in each time series. The variables related with the
commodity prices, productivity, private inflows and relative manufacturing output were
log-transformed. From now on we represent with capital letters the original variables
and with lower case the log-transformed ones. The number of lags in the Augmented
Dickey-Fuller test is based on Schwartz’s information criterion. The results are shown
in Table 4.1.
The analysis of the graphs in Appendix E, as well as the results of the tests in the
previous table, suggest a need for at least one unit root for all series. The t-statistic and
the p-values of the ADF indicate that the unit root null hypothesis cannot be rejected for
all the series. Considering the first differences, the null hypothesis is rejected for the
usual significance levels. Therefore, it is concluded that these series are integrated of
order one, I(1).
111
Table 4.1: Unit root tests
Augmented Dickey-Fuller test statistic
ADF in levels
t-Statistics
Prob.
ADF in first differences
t-Statistics
Prob.
Constant
Inflow
-1.37791
0.2821
-3.68344
0.0008
GOV
-1.40317
0.5709
-4.82064
0.0004
Man_serv
-0.32524
0.9118
-3.65052
0.0093
Commo
-2.19290
0.2120
-7.06425
0.0000
OPEN
-1.67939
0.4335
-7.43676
0.0000
RER
-1.90978
0.3242
-3.13937
0.0327
Product
Constant and linear
trend
-1.03114
0.7319
-3.09833
0.0354
Inflow
-2.29349
0.1451
-3.66816
0.0000
GOV
-1.11901
0.9127
-4.80108
0.0022
Man_serv
-2.56762
0.2963
-3.56508
0.0470
Commo
-1.93562
0.6165
-6.95643
0.0000
OPEN
-2.47366
0.3386
-7.33044
0.0000
RER
-1.45163
0.8270
-3.35052
0.0749
Product
-2.52666
0.3143
-4.18595
0.0109
Null Hypothesis: There is a unit root.
4.3.2.
Johansen analysis
The absence of stationarity in all the series implies that a co-integration analysis must
be carried out, to analyze if the series are related in the long term. In this sense, we
apply Johansen´s (1991) methodology and estimate a Vector Autoregressive model and
test for cointegration. Table 4.2 shows the results of both trace and maximum
eigenvalue cointegration test statistics for the linear trend option, as it seems to adjust
best to data. To identify the lag length in the VAR, the Akaike Information (AIC), the
Hannan-Quinn (HQ) and the Schwarz (SC) Criteria have been implemented. The chosen
lag structure is one (the smallest value) following the SC and HQ criterion, as it also
ensures that residuals are white noise. One dummy relative to 2009 has been included
in the cointegration test to take into account the world economic crisis.
112
Table 4.2: Johansen Cointegration Test. Sample period: 1972-2011
Trend assumption: Linear deterministic trend
Trace test
MaximumEigenvalue test
Null
Hypothesis
Alternative
Trace Stat.
Prob.**
Null
Hypothesis
Alternative
Trace Stat.
Prob.**
r<0*
r=1
185.1671
0.0000
r<0*
r=1
65.29065
0.0002
r<1*
r=2
119.8764
0.0004
r<1*
r=2
48.61277
0.0044
r<2
r=3
71.26365
0.0382
r<2
r=3
29.19212
0.1638
* denotes rejection of the hypothesis at the 0.01 level
**MacKinnon-Haug-Michelis (1999) p-values
In Table 4.2, the trace statistic and the Maximum Eigenvalue test show the presence of
two co-integrating relations with a significance level of 1%. The maximum eigenvalue
statistic is located in two co-integrating equations at 5% and 1% significance level.
Finally we consider two cointegration relations since this would be adopted at 1%
significance level by the two tests and allows equations (4.2) and (4.3) to be estimated.
We also impose restrictions to identify the cointegration relation in equations (4.2) and
(4.3). Therefore, the dependent variable for each co-integrating equation is imposed,
giving the value of zero to the variables that are not included in each of the equations.
Results suggest that commodity prices (Commo) are co-integrated with the real
exchange rate (RER) and with the relative manufacturing output (Man_Serv). The cointegrating relations can be formalized in the following equations, in which the standard
errors are in brackets:
RER=
-486
+
130.08 PRODUCT+
47.08 Commo
(43.051)
Man_Serv =
1.84
+ 0.007 RER
(0.002)
(4.585)
–
0.21 Product +
(0.811)
+ 0.68 GOV -
2.31 OPEN -
(0.205)
(0.3506)
0.07 Inflow (0.051)
0.025GOV +
(0.038)
14.01 Inflow
(4.4)
(2.426)
0.055 OPEN
(0.007)
113
(4.5)
Equations (4.4) and (4.5) are the estimated versions of equations (4.2) and (4.3),
respectively. In this sense, equation (4.4) can be identified as a long run competitiveness
equation and equation (4.5) as a long run relative manufacturing output equation.
As regards the model for the short run dynamics, we present our estimation results in
Tables 4.3 and 4.4. The results that refer to the speed of adjustment of the long run
equilibrium deviations or error correction term are shown in Table 4.3.
Table 4.3: Speed of adjustment
Error
∆(RER)
∆(Commo) ∆(Man_Serv) ∆(GOV) ∆(OPENESS) ∆(Product)
Correction
Long-run
0.084
0.000
0.000
-0.164
0.042*
-0.001
competiveness
[ 0.897]
[-0.004]
[ 0.015]
[-1.147]
[ 0.485]
[-5.912]
equation
Long-run
27.647*
0.198
0.006
-1.541
-1.280
-0.003
relative
manufacturing
[ 6.017]
[ 0.922]
[ 0.145]
[-0.219]
[-0.300]
[-0.172]
equation
∆=1-L is the difference operator where L is the lag operator such that  = −1 .
t-statistics are in brackets and the p-value for the LR-test on the restrictions is 0.26.
*denotes significance at 5% level
∆(Inflows)
0.009
[ 0.774]
-0.054
[-0.092]
As seen in Table 4.3, the first cointegration relation (long-run competitiveness equation)
is affecting the short-run movements of ∆(), while it is not significant for the
short-run dynamics of the remaining variables. In this sense, deviations from the longterm equilibrium of the real exchange rate impact on the short run movements of the
ratio of non-trade to trade prices. The error correction term associated to the second
cointegration relation (long-run relative manufacturing equation) is significant for the
variable ∆ (RER).
The short-run dynamics for ∆(RER) and ∆(Man_Serv) are reported in Table 4.4, The
equations for the remaining variable in the model are given in the Appendix F. The real
exchange rate is affected by commodity prices, the country’s openness to international
trade and the previous exchange rate in the short term. On the other hand, the relative
114
manufacturing output is significantly affected by the variable, RER, which relates to the
real effective exchange rate.
Table 4.4: VECM system short-run coefficients.
∆(RER)t
∆(Man_Ser)t
Coefficient
t-statistics
Coefficient
t-statistics
∆(RER)t-1
-0.437
[-2.167]*
-0.004
[-2.164]*
∆(Commo)t-1
10.550
[ 2.175]*
0.059
[ 1.399]
∆((Man_Serv)t-1
19.187
[0.963]
-0.268
[-1.546]
∆(GOV)t-1
0.288
[ 1.903]
-0.001
[-0.408]
∆(OPENESS)t-1
-0.788
[-2.939]*
-0.004
[-1.659]
∆(Product)t-1
-24.553
[-1.011]
0.104
[ 0.494]
∆(Inflows)t-1
-1.808
[ 1.213]
-0.008
[-0.516]
ECM_Ct-1
0.084
[ 0.897]
0.000
[ 0.015]
ECM_Mt-1
27.647*
[ 6.017]
0.006
[ 0.145]
0.707
[-1.013]
-0.018
[-1.910]
C
Summary statistics
R-squared
0.763
0.614
Adj. R-squared
0.663
0.451
S.E. equation
6.079
0.053
t-statistics are in brackets. ECM_Ct-1 and ECM_Mt-1 are the residuals of the Long-run competitiveness equation and
long-run relative manufacturing equation, respectively.
The properties of the residuals of the estimated model have been analyzed (see Table
4.5). The system residual Lagrange-Multiplier test for autocorrelation shows that the
null of no residual correlation cannot be rejected. The null hypothesis of normality for
the residuals is tested through the joint Jarque-Bera test and cannot be rejected at the
usual significance levels.
115
Table 4.5: Residual tests for the VECM system.
Autocorrelation
Test
LM(1)
LM(2)
LM(3)
LM(4)
LM(5)
Normality
Heteroskedasticity:
no cross terms
Chi-sq
^2 (49)= 49.91
^2 (49)= 54.62
^2 (49)= 45.64
^2 (49)= 56.65
^2 (49)= 38.06
^2 (14)= 6.678
^2 (560)= 533.02
p-value
0.4366
0.2695
0.6101
0.3003
0.8708
0.6179
0.7880
Note: Normality is based on joint Jarque-Bera test; orthogonalization is based on Cholesky (Lütkepohl).
As a robustness check, the long run equations have also being estimated through the
Engle and Granger (1987) single equation methodology. The main results regarding the
symptoms of the Dutch disease, available from the authors upon request, remain the
same.
4.3.3.
Results analysis
Perhaps the first important result is related to the fact that there is evidence that
commodity prices positively affect the real exchange rate, as the Dutch disease
hypothesis predicts. In fact, an increase in the commodity price index of 1% produces a
real appreciation of 0.47 dollars, “ceteris paribus”, see equation (4.4) for the long-run
competiveness. The real appreciation of the Colombian peso creates a loss of
competiveness in international markets, since local prices are higher than their
international competitors. Likewise, internal consumers replace the demand of
expensive national goods with cheaper imports.
Regarding equation (4.4), or long-run competiveness equation, the majority of the signs
are those expected in the theory and all the variables are significant. It is important to
emphasize the GOV variable, which represents the share of government consumption in
the GDP, as a one point increase in this relation produces a real appreciation of 0.64
116
dollars, “ceteris paribus”. Theoretically, the expenditure effect boosted by the
government generates an increase in prices, which causes a relative price rise of local
goods compared to international products (real currency appreciation). The variable that
relates to the openness of the economy to foreign trade, OPEN, has a negative
relationship with the real exchange rate, and a one percent increase in the trade to GDP
ratio generates a real depreciation of 2.35 dollars. This result indicates that liberalization
of commercial policy leads to real depreciations as an increase in the openness degree
leads to a convergence of international prices.
Productivity is positively related to the real exchange rate. The model estimates that an
increase of the country productivity of 1% causes a real appreciation of the currency of
approximately 1.3 dollars, “ceteris paribus”. As a consequence of the increase in
productivity, the salaries and the prices of local goods rise compared to those from
abroad, thus creating real appreciation of the currency. Private inflows, Inflow, do not
have the expected positive sign, however.
With respect to the relative manufacturing equation, see equation (4.5), no supporting
evidence can be found that the real exchange rate, RER, directly affects the relative
manufacturing output negatively in the long term. In fact, a one dollar real appreciation
generates an increase in relative manufacturing output of 0.70%. Our result suggests
that the symptom associated to Dutch disease that is related to the relative deindustrialization is not present in the Colombian case. It is important to note the variable
MAN_SERV, is measured as the ratio of manufacturing production to services
production. Thus, our results show that, in the long run, real appreciation does not cause
the relative de-industrialization process in Colombia. The positive long-run relationship
may be explained by:
117
a. There is no reallocation of factors within the sectors due to the movement of the
real exchange rate, so real appreciation does not push production factors from
manufacturing exporting sectors to the production of services.
b. The appreciation periods are often used by manufacturing companies to capitalize
and become more productive. Real appreciation can generate a positive effect on
capacity of industrial production when facilitating access to imported inputs and
capital goods, as well as improve financing conditions abroad. Therefore, the
reduction of competitiveness due to the real appreciation of the exchange rate can
be compensated by an increase in manufacturing companies’ productivity.
c. Abstracting from service production developments, a resource boom implies that
the non-commodity exports must lose share in total exports. We actually see this
since 2007 for the post-2003 commodity boom (see Table 4.6)25. However, it does
not automatically imply a shrinking of manufacturing output. As Colombia is a
small economy that takes the world price of manufactures as given, manufacturing
output could be maintained with the increase of the domestic demand associated
with the resource boom. Hence, competitiveness losses are compensated by gains in
domestic demand. In fact, periods of appreciation coincide with increases in the
GDP growth or GDP growth above the mean (see Figure 4.7). Moreover, for the
manufacturing sector domestic demand is a more important market than the
external one.
For the period between 2000 and 2009, for example, domestic
manufacturing sales represented about 83% of total sales (The National
Administrative Department of Statistics, DANE)
25
Long series on manufacturing exports are not available for Colombia, therefore, we could not include
this variable in our model.
118
Table 4.6: Share of the different economic sectors on total exports. Colombia:
2003-2012.
2003
2004
2005
2006
2007
2008
2010
2012
Manufacturing
53,64
54,48
52,35
53,59
56,33
47,73
37,56
29,84
Agricultural Sector
9,04
8,33
8,09
7,63
7,02
5,66
5,44
4,36
Commodities
37,32 37,19 39,56 38,78
Source: National Statistics Office of Colombia ,DANE
36,65
46,61
57,00
65,79
We have checked for the robustness of our results by replacing the variable RER by
COMM in equation (4.5) of the VECM. Our results (signs and significance of the
variables) are not altered by this change, see Appendix G.
Although there is a positive long-run relationship between the real exchange rate and
relative manufacturing output, in the short run the negative effect prevails (Table 4.4).
The loss in competitiveness may be affecting exports and production in the last resource
boom, 2003-2011. In fact, the increase in commodity price has been the more lasting in
our period of study (in eight years the real price of the commodity price index increased
by 225%). Besides, it is important to note the role of oil as intermediate good. In the
short run, as manufacturing uses oil as input, there is a transitory loss of output as a
result of the oil price surge that reinforces the competitiveness losses due to real
appreciation26.Our results are in the line with the theoretical studies of Buiter and Purvis
(1983).
26
Since 1993, oil prices account for more than a half of the share in our commodity index. The importance
of oil as intermediate good is therefore greater during the last commodity boom.
119
Figure 4.7: Colombian real exchange rate and GDP growth.
180
160
140
120
.100
100
.075
80
.050
60
.025
.000
-.025
-.050
1975
1980
1985
1990
Real Exchange Rate
1995
2000
2005
2010
GDP growth
Shaded areas represent the most important commodity price booms.
The horizontal line represent the growth mean for the period, 3.76%.
Source: IMF International Financial Statistics, and authors´ calculations.
Although the parameter associated to productivity has the right sign, it is not significant.
Moreover, the variable INFLOW showed a non significant positive relation with the
relative manufacturing output.
Other variables that significantly affect the relative manufacturing output in equation
(4.5) are OPEN and GOV. One additional point in the ratio trade to GDP, that is the
openness index, generates a 5.5% increase in the relative manufacturing output. Hence,
a greater exposure to international trade benefits the manufacturing production.
According to equation (4.4), the real exchange rate is negatively related in the long run
to trade openness. This means that in the long run one of the determinants of a
depreciated currency is a high degree of economic openness. It is normal, therefore, that
an increase in this ratio, OPEN, upsurges manufacturing production by allowing a
greater access to international markets and making them more competitive. On the other
hand, the estimated coefficients of government expenses share in GDP, GOV, shows a
negative relationship with relative manufacturing output. When the government
120
increases its expenses by one per cent of GDP, the variable MAN_SERV decreases by
2.52%. This result confirms the idea that governments tend to spend more on nontradable goods such as services, health and education, Lartey (2007). It is important to
note that an expansionary fiscal policy negatively affects the competitiveness of
tradable sectors by increasing the real exchange rate. Therefore, public spending is a
major source of pressure on aggregate demand, prices and real exchange rate. In this
regard, the government may indirectly weaken the competitiveness of manufacturing
sector. This is in line with the Stokke (2008) results for the South African economy.
According to his study an increased public consumption due to a resource boom caused
a real exchange rate appreciation and led to an expansion of service at the cost of
industrial tradable sector.
4.4. Conclusions and policy recommendations
The aim of this work was to find empirical evidence of the Dutch disease symptoms in
Colombia, of a country that experienced first a boom in coffee and lately in oil. Since
the variables in the analysis are not stationary, Johansen’s (1995) cointegration
approach was used to establish the equilibrium relations between them in the long term.
A Vector Error Correction Model, VECM, has been estimated to determine whether
commodity prices are related to the real exchange rate and the relative manufacturing
output in time. Other variables suggested by the theory and included in the model are
productivity, the ratio between the national government expenditure to GDP, and the
degree of openness.
The VECM estimates show that there is co-integration evidence between the
commodity prices, the real exchange rate and the relative manufacturing output. In this
way, increases in oil price produce a negative effect on the country’s competiveness,
121
since national goods become more expensive, compared to those of the rest of world, or
they appreciate in real peso terms. However, our results do not suggest that the real
exchange rate significantly affect the relative manufacturing output in the long run. In
fact, periods of booms in coffee and oil match with periods of real appreciations but
relative de-industrialization. Our results show that relative de-industrialization that has
been occurring through the decrease in the size of the manufacturing sector relative to
services is not due to the appreciation moved by commodity prices. Further studies are
needed to analyze what internal reasons drive this pattern.
Our results, which indicate that relative manufacturing has so far been spared the
negative effects of commodity price increases, may provide only temporary relief for
policymakers in Colombia. If commodity prices remain high in the future, the nominal
and real exchange rates will continue to appreciate by putting pressure on noncommodity industries. Then, the competitiveness losses will overcome the productivity
gains from capital replacement and local demand. Against this backdrop, policymakers
would be well advised to implement structural measures aimed to prevent unproductive
and pro-cyclical public expense. Our result shows that public spending is a major source
of pressure on aggregate demand and real exchange rate. In this regard, the government
may indirectly weaken the competitiveness of manufacturing.
Our results present great challenges at economic policy level. What to do then with the
proceeds of resource boom? In the literature there are two main approaches: Hartwick's
rule for sustainability, Hartwick (1976), and the permanent income approach, Medas
and Zakharova (2009). The first is an extreme intergenerational altruism rule.
According to this rule, all revenues from extractive activities are invested in financial
assets abroad, whose returns are the only raw material source of government spending.
The priority of this rule lies in the transfer of most of the commodity wealth for future
122
generations. The benefits of this approach, which is followed by Norway, are that
neither losses of competitiveness, nor relative de-industrialization may occur. The oil
revenues are placed in a State Oil Fund, which reports directly to the central bank and
those who invest resources in stock markets worldwide. Only the net income flow is
delivered to the government.
The permanent income approach consists of spending the extractive wealth at a gradual
pace so as to ensure the ongoing participation of each generation according to a social
welfare criterion, or a fiscal rule previously defined by the society. Chile has a fiscal
rule that takes into account periods of prosperity and periods of "lean" and is based
mainly on the fiscal balance. In early 2009, this system allowed Chile to use the savings
from previous budget surpluses to conduct anti-cyclical policy and deal with the impact
of the international recession on the economy. The appreciation and relative deindustrialization can be avoided in this approach by saving a substantial fraction of the
commodity related revenue windfalls.
Moreover, the approach provides fiscal
stability27.
Although the two previous approaches, based on savings, have great benefits, the
temptation to spend is great for a developing country like Colombia. Colombia’s
natural resource wealth can help seize the growth opportunity by serving directly as a
key source of growth if properly managed. So the challenge of policy lies in saving
enough to prevent relative de-industrialization and to direct public expenses into
productive activities. But this opportunity will only be realized if windfall earnings are
managed judiciously within a long-term horizon, avoiding the “voracity effect” – a
more than proportional increase in discretionary fiscal spending in response to a
27
In a recently article on Dutch disease for the Russia case Benedictow et al.(2013) found that saving oil
revenues in, for example, a wealth fund may reduce procyclicality of fiscal policy and the risk of boom–
bust cycles.
123
positive revenue shock, such as a commodity revenue windfall-, Tornell and Lane
(1999).
References
Acosta, P. & Mandelman, F. 2009, "Remittances and the Dutch disease", Journal of
International Economics, vol. 79, no. 1, pp. 102-116.
Algieri, B. 2011, "The Dutch Disease: evidences from Russia", Economic Change and
Restructuring, vol. 44, no. 3, pp. 243-277.
Alberola E. 2003, "Misalignment, liabilities dollarization and exchange rate adjustment
in Latin America", Bank of Spain Working Paper 0309, July.
Arellano, C., Bulir, A., Lane,T. & Lipschitz, L. 2005, "The Dynamic Implications of
Foreign Aid and Its Variability,” IMF Working Paper WP/05/119, June.
Beine, M., Bos, C. S., & Coulombe, S. 2012, "Does the Canadian Economy Suffer from
Dutch Disease?",Resource and Energy Economics, vol. 34, no.4, pp. 468-492.
Bénassy-Quéré A., Duran-Vigneron P., Lahrèche-Revil A. & Mignon, V. 2004,
"Burden Sharing and Exchange-Rate Misalignments within the Group of Twenty",
CEPII research center, Working Paper 13, September.
Benedictow, A., Fjærtoft, D., & Løfsnæs, O. 2013, "Oil dependency of the Russian
economy: An econometric analysis", Economic Modelling, vol. 32(C), pp. 400428.
Bruno, M. & Sachs, J. 1982, "Energy and resource allocation: A dynamic model of the
‘Dutch Disease’", Review of Economic Studies, vol. 49, pp. 845-859.
Buiter, W.H. & Purvis, D.D. 1983,"Oil, disinflation, and export competitiveness: A
model of the Dutch Disease", in J.S. Bhandari and B.H. Putnam (eds.), Economic
Interdependence and Flexible Exchange Rates, MIT Press, Cambridge, Mass.
124
Chatterji, M. & Price, S. 1988, "Unions, Dutch Disease, and unemployment", Oxford
Economic Papers, vol. 40, no.2, pp. 302-321.
Corden, W.M. 1984, “Booming Sector and Dutch Disease Economics: Survey and
Consolidation,” Oxford Economic Papers, vol. 36, no. 3, pp. 359–380.
Corden, W.M. 1994, Economic Policy, Exchange Rates and the International System,
Oxford University Press.
Corden, W.M. & Neary, J.P. 1982, "Booming Sector and de-Industrialisation in a Small
Open Economy", Economic Journal, vol. 92, no. 368, pp. 825-848.
Coudert V, Couharde, C & Mignon V. 2011, "Does Euro or Dollar Pegging Impact the
Real Exchange Rate? The Case of Oil and Commodity Currencies", The World
Economy, vol. 34, no. 9, pp. 1557.
Davis, G. 1995, "Learning to love the Dutch disease: Evidence from the mineral
economies", World Development, vol. 23, no.10, pp.1765-1779.
DeLoach, S. 2001, "More Evidence in favor of the Balassa-Samuelson Hypothesis",
Review of International Economics, vol. 9, no.2, pp. 336-342.
Edwards S. 1984, "Coffee Money and inflation in Colombia", World Development, vol.
12, pp. 1107-1117.
Edwards S. 1989, Real Exchange Rates, Devaluation and Adjustment: Exchange Rate
Policy in Developing Countries, Cambridge: MIT Press.
Egert, B. 2008, "Dutch Disease Scare in Kazakhstan: Is it real?",Open Economies
Review, vol. 19, no. 2, pp. 147.
Engle, R. & Granger, C. 1987, "Cointegration and Error Correction – Representation,
Estimation, and Testing ", Econometrica, vol. 55, no. 2, pp. 251-276.
Gelb. A. 1988, Oil Windfalls: Blessing or Curse? New York: Oxford University Press.
Guha, P. 2013, "Macroeconomic effects of international remittances: The case of
developing economies”, Economic Modelling, Vol. 33, pp. 292-305.
125
Hartwick, J.M. 1977, "Intergenerational equity and investing rents from exhaustible
resources”, American Economic Review, Vol 66, pp. 972-974
Ismail, K. 2010, "The Structural Manifestation of the ‘Dutch Disease’: The Case of OilExporting Countries" IMF Working Paper WP/10/103, April.
Johansen, S. 1991, "Estimation and Hypothesis Testing of Cointegration Vectors in
Gaussian Vector Autoregressive Models", Econometrica, vol. 59, no. 6.
Johansen, S. 1995, Likelihood based inference on cointegration in the vector
autoregressive model. Oxford University Press, New
York.
Kamas, L. 1986, "Dutch disease economics and the Colombian export boom", World
Development, vol. 14, no. 9, pp. 1177-1198.
Kang, S. & Lee, H. 2011, "Foreign Direct Investment and De-industrialisation",
TheWorld Economy, vol. 34, no. 2, pp. 313-329.
Kronenberg, T. 2004, "The Curse of Natural Resources in the Transition Economies",
The economics of transition, vol. 12, no. 3, pp. 399-426.
Krugman, P. 1987, "The narrow moving band, the Dutch Disease and the competitive
consequences of Mrs. Thatcher on trade in the presence of dynamic scale
economies", Journal of Development Economics, vol. 27, pp. 41-55.
Kukk, K. 2007, "Fiscal Policy Effects on Economic Growth: Short Run vs Long Run".
Working Papers TTUWPE No 167, Tallinn School of Economics and Business
Administration, Tallinn University of Technology.
Lartey, E. 2007, "Capital Inflows and the Real Exchange Rate: An empirical study of
sub-Saharan Africa", Journal of International Trade& Economic Development, vol.
16, no. 3, pp. 337-357.
Lartey, E. 2008, "Capital Inflows, Resource Reallocation and the Real Exchange Rate",
International Finance, vol. 11, no. 2, pp. 131-152.
126
Lartey, E.K.K. 2011, "Financial Openness and the Dutch Disease", Review of
Development Economics, vol. 15, no. 3, pp. 556-568.
MacKinnon, J. Haug, A & Michelis, L. 1999, "Numerical Distribution Functions of
Likelihood Ratio Tests for Cointegration", Journal of Applied Econometrics, vol.
14, no. 5, pp. 563-577.
Magud, N & Sosa, S. 2010, "When and Why Sorry About Real Exchange Rate
Appreciation? The Misssing Link between Dutch Disease and Growth ".IMF
Working Papers WP/10/271, December.
Medas, P. & Zakharova, D. 2009, "A Primer on Fiscal Analysis in Oil-Producing
Countries".IMFWorking Paper WP/09/56, March.
Meisel, A. 1998, "Dutch Disease and Banana Exports in the Colombian Caribbean,
1910-1950".Cuadernos de Historia Económica y Empresarial, No. 26, April.
Morshed, A. & Turnovsky, S.J. 2004, "Sectorial adjustment costs and real exchange rate
dynamics in a two-sector dependent economy", Journal of International
Economics, vol. 62, pp. 147-177.
Neary, J.P. 1988, "Determinants of the equilibrium real exchange rate", American
Economic Review, vol. 78, no. 1, pp. 210-215.
Neary, J.P. & Purvis, D. 1982,"Sectoral shocks in a dependent economy: long-run
adjustment and short-run accommodation", Scandinavian Journal of Economics,
vol. 84, no. 2, pp. 229-253.
Nkusu, M. 2004, “Aid and the Dutch Disease in Low-Income Countries: Informed
Diagnoses for Prudent Prognoses”, IMF Working Paper WP/04/49, March.
Otero, J. & Smith, J. 2000, “Testing for cointegration: power versus frequency of
observation — further Monte Carlo results“, Economics Letters, Vol. 67, no. 1, pp.
5-9.
Prati, A. & Tressel, T. 2005, "Aid Volatility and Dutch Disease: Is There a Role for
Macroeconomic Policies? "IMF Working Paper WP/06/145, March.
127
Puyana, A. 2000, "Dutch Disease, Macroeconomic Policies, and Rural Poverty in
Colombia", International Journal of Politics, Culture, and Society, vol. 14, no. 1,
pp. 205-233.
Salter, W. 1959, "Internal and External Balance: The Role of Price and Expenditure
Effects", Economic Record, Vol. 35, pp. 226-38.
Suescún, R. 1997, "Commodity Booms, Dutch Disease, and Real Business Cycles in a
Small Open Economy: The Case of Coffee in Colombia”, Borradores Semanales de
Economía. Subgerencia de Estudios Económicos del Banco de la República de
Colombia . no. 73.
Swan, T. 1960, "Economic Control in a Dependent Economy", Economic Record,
Vol.36, pp. 51-66.
Stokke, H. "Resource boom, productivity growth and real exchange rate dynamics : a
dynamic general equilibrium analysis of South Africa", Economic Modelling,
Vol.25, no.1, pp.140-160
Tornell, A. & Lane, P. 1999,"The Voracity Effect", American Economic Review,vol.89,
no.1, pp. 22–46.
Van der Ploeg, F. 2011, "Natural Resources: Curse or Blessing?," Journal of Economic
Literature, vol. 49, no.2, pp. 366-420.
128
Appendix A: Summary of Dutch disease literature of Colombia
Period of
analysis
Booming
commodity
Methodology
Country specific studies
Meisel (1998)
1891-1950
Coffee
Descriptive analysis
Kamas (1986)
1975-1980
Coffee
Descriptive analysis and OLS
Puyana (2000)
1985-1997
Oil
discoveries
Descriptive analyisis
Suescún (1997)
1952-1992
Coffee
Real Business Cycle (RBC) model
Edwards (1984)
1952-1980
Coffee
Two and three stage-least-squares
Evidence
Coffee boom led to a squeezing out of the banana exports.
Symptoms of the disease were observed mainly by the appreciation
of the real exchange rate, and a decrease in exports and production
in the banana sector.
Non coffee exports were hurt by real appreciation. Growth rates for
non-coffee and manufacturing exports fell during the period. In the
OLS estimation, the real exchange rate affects positively but
significantly the manufacturing output.
Oil discoveries impact negatively on the agricultural sector.
Find evidence of de-industrialization in a transient coffee shock.
Coffee price changes have been negatively related to the rate of
devaluation and closely related to money creation and inflation.
Cross country studies where Colombia is included
Davis (1995)
1970 and 1991
Oil
Descriptive analysis
According to the author, Colombia performed better, in terms of
development indicators, than more commodity-dependent
economies.
Ismail (2010)
1977-2004
Oil
Pooled least-square estimation with country
Find de-industrialization for Colombia in response to shocks in oil
prices. However, the shocks have a lesser impact on Colombia due
to the country´s lack of financial openness.
129
Appendix B: Commodities production and crude oil
discoveries in Colombia.
Colombian Coffee Arabica Production, 60 kg bags (2005=100)
Colombian Coffe Arabica Production
160
140
120
100
80
60
1975
1980
1985
1990
1995
2000
2005
2010
Source: United States Department of Agriculture, USDA
Colombian Crude Oil Production, 1000 barriles/day (2005=100)
200
160
120
80
40
0
1975
1980
1985
1990
1995
2000
2005
2010
Source: US Departament of Energy, DOE.
130
Colombian Coal Production. Million Short tones
100
80
60
40
20
0
80
82
84
86
88
90
92
94
96
98
00
02
04
06
08
10
Source: Source: US Departament of Energy, DOE.
Major oil discoveries in Colombia
Year of
Discovery
1918
1940
1941
1946
1954
1960
1962
1963
1969
Oil
reservesMillionbarrels
800
270
300
300
200
300
140
240
320
Chuchupa
1972
7 GasCubicTerapies
Apiay-Suria
Caño Limón
San Francisco
Cusiana
Cupiagua
Guando
1981
1983
1985
1989
1993
2000
215
1250
150
750
510
130
Gibraltar
2003
Name
La Cira-Infantas
Tibú
Casabe
Velásquez-Palagua
Yariguí
Provincia-Payoa
Río Zulia
Orito
Castilla
630 Gas cubicGigapies
15 Millionbarrels
Source: ECOPETROL
131
Appendix C: Foreign Direct Investment in the petroleum
sector. Colombia
Colombia BOP direct invest in colombian petroleum sector. Vlaue /millions. Dollars
1,400
1,200
1,000
800
600
400
200
0
-200
-400
96
97
98
99
00
01
02
03
04
05
06
07
08
09 10
Source: Central Bank of Colombia. Banco de la República.
132
Appendix D: Data sources
Real Exchange Rate RER. Authors´ calculations based on the Exchange Rate Index
2005=100 /Index Number /averages /exchange rate index /Cnt: Colombia /Source:
IMF, Wash; CPI low & Middle Income Urban /Index Number /Base year: 2005
/averages /Cnt: Colombia /Source: IMF, Wash; and CPI all Items City Average /Index
Number /Base year: 2005 /averages /Cnt: United States /Source: IMF, Wash.
Relative Manufacturing output Man_Serv: Ratio Manufacturing to services value
added (current MN LCU) /Cnt: Colombia. Source: Data Service & Information. World
Bank Statistics (2005=100).
The Balassa Samuelson or Productivity effect, Product. Proxied by the ratio of the
consumer price index (CPI) to the producer price index (PPI).
CPI low & Middle
Income Urban and PPI / Wpi /Index Numbers /Base year: 2005 /averages /Cnt:
Colombia /Source: IMF, Wash.
Government expenditure GOV: General government final consumption expenditure
(% of GDP) /Cnt: Colombia /IMF, Washington. (2005=100).
Real commodity prices, Commo: Authors´ calculations based on both oil and coffee
prices and taking into account their relative weight on commodity exports since 1972 to
2011. The commodity prices are: The Cushing, OK WTI Spot Price (Dollres per
Barrel). Source: Energy Information Administrator (2010), and the external price of
Colombian coffee (exdock) /Cents per pound of 453.6 gr./annual average /
sublime/source: Colombian Coffee Growers Federation. We deflected by US CPI
(2005=100).
Openness, OPEN.Trade (% of GDP) /Cnt: Colombia/Source: IMF, Wash.
Private Inflows, Inflow. Private capital flows, total (BoP, current MN US$) /Cnt:
Colombia/Source: IMF, Wash.
133
Appendix E: Graphs
Ratio Manufacturing to Ser vices, Man_Serv.
Commodity Pr ice Index, Commo.
300
250
Balassa-Samuelson, Product.
.6
4.7
.5
4.6
.4
4.5
.3
4.4
200
150
100
50
.2
1975
1980
1985
1990
1995
2000
2005
2010
4.3
1975
1980
Private Inflows, Inflow.
1985
1990
1995
2000
2005
2010
1975
Openness, OPEN
6
5
1980
1985
1990
1995
2000
2005
2010
Government Expendi tur e, GAS_GOB.
110
24
100
20
90
16
80
12
70
8
4
3
2
1
0
60
1975
1980
1985
1990
1995
2000
2005
2010
2005
2010
4
1975
1980
1985
1990
1995
2000
2005
2010
1975
1980
1985
1990
1995
2000
2005
2010
Real Exchang e Rate, RER.
180
160
140
120
100
80
60
1975
1980
1985
1990
1995
2000
134
Appendix F: VECM system short-run coefficients of the
rest of the variables of the model.
∆(GAS_GOV) ∆(OPEN)
∆(GOV)t-1
∆(OPENESS)t-1
∆(Commo)t-1
∆(RER)t-1
∆((Man_Serv)t-1
∆(Product)t-1
∆(Inflows)t-1
∆(Commo)
∆(Product)
∆(Inflow)
-0.162
-0.082
0.012
0.000
0.005
[-0.70163]
[-0.58550]
[ 1.70384]
[-0.40311]
[ 0.24583]
0.348
-0.021
-0.024
-0.001
0.019
[ 0.85182]
[-0.08564]
[-1.91560]
[-1.04233]
[ 0.54776]
-3.641
1.754
-0.003
0.055
0.056
[-0.49206]
[ 0.39052]
[-0.01539]
[ 3.36804]
[ 0.08988]
0.115
0.000
-0.009
-0.002
0.035
[ 0.39799]
[ 0.00153]
[-0.96770]
[-2.43241]
[ 1.41664]
-18.324
-5.007
1.299
0.019
2.205
[-0.60313]
[-0.27160]
[ 1.39339]
[ 0.27704]
[ 0.86218]
32.382
5.676
0.666
0.076
-0.364
[ 0.87400]
[ 0.25249]
[ 0.58611]
[ 0.92239]
[-0.11663]
5.846
2.401
-0.016
-0.018
-0.222
[ 2.14826]
[ 1.45409]
[-0.19381]
[-2.97418]
[-0.96718]
-0.905
0.397
0.029
0.010
0.147
[-0.55703]
[ 0.40245]
[ 0.58784]
[ 2.82502]
[ 1.07193]
-0.164
0.042
0.000
-0.002
0.009
[-1.14754]
[ 0.48507]
[-0.00444]
[-5.91250]
[ 0.77427]
-1.541
-1.280
0.198
-0.003
-0.054
[-0.21978]
[-0.30093]
[ 0.92211]
[-0.17246]
[-0.09232]
C
ECM_Ct-1
ECM_Mt-1
t-statistics are in brackets. ECM_Ct-1 and ECM_Mt-1 are the residuals of the Long-run competitiveness equation
and long-run relative manufacturing equation, respectively.
135
Appendix G: Estimation results when RER is replaced by
COMMO –equation (4.3)
RER=f(PRODUCT,COMM,GOV, OPEN, INFLOW)
MAN_SERV = F (COMMO, PRODUCT, OPEN, INFLOW,GOV)
Johansen cointegration tests
Data Trend:
Test Type
Trace
Max-Eig
None
No Intercept
No Trend
2
2
None
Intercept
No Trend
2
2
Linear
Intercept
No Trend
2
2
Linear
Intercept
Trend
3
2
Quadratic
Intercept
Trend
3
2
*Critical values based on MacKinnon-Haug-Michelis (1999)
Long-run relationships
CointegratingEq:
CointEq1
CointEq2
(RER)t-1
-1.000.000
0.000000
(Commo)t-1
4.650.759
0.182518
-101.052
(0.14989)
[ 4.60232]
[ 1.21771]
(Man_Serv)t-1
0.000000
-1.000.000
(GOV)t-1
1.592.157
-0.024006
(0.34849)
(0.00517)
[ 4.56872]
[-4.64430]
(OPENESS)t-1
(Product)t-1
(Inflows)t-1
C
-4.247.109
0.052513
(0.68556)
(0.01017)
[-6.19509]
[ 5.16426]
-0.915238
1.238.306
-764.294
-113.364
[-0.01197]
[ 1.09233]
-1.038.388
-0.138734
-425.211
(0.06307)
[-2.44206]
[-2.19969]
1.922.805
-3.844.675
Standard errors in ( ) & t-statistics in [ ]
136
Speed of adjustment
Error
∆(RER)
∆(Commo)
∆(Man_Serv)
∆(GOV)
∆(OPENESS)
∆(Product)
∆(Inflows)
Long-run
-0.038835
-0.002799
0.000730
-0.224123
0.057172
-0.001643
0.013739
competiveness
(0.08708)
(0.00359)
(0.00081)
(0.12063)
(0.07232)
(0.00028)
(0.00951)
equation
[-0.44595]
[-0.78077]
[ 0.89868]
[-1.85799]
[ 0.79058]
[-5.83122]
[ 1.44519]
Long-run
2.069.808
0.122526
0.028360
-1.113.563
0.820926
-0.064051
0.702042
relative
-649.994
(0.26761)
(0.06065)
-900.365
-539.774
(0.02103)
(0.70958)
[ 3.18435]
[ 0.45786]
[ 0.46757]
[-1.23679]
[ 0.15209]
[-3.04629]
[ 0.98938]
Correction
manufacturing
equation
∆=1-L is the difference operator where L is the lag operator such that  = −1 .
t-statistics are in brackets and the p-value for the LR-test on the restrictions is 0.26.
*denotes significance at 5% level
VECM system short-run coefficients
∆(RER)
∆((Man_Serv)
Coefficient
t-statistics
Coefficient
t-statistics
∆(RER)t-1
-0.497371
[-2.37899]
-0.002760
[-1.41476]
∆(Commo)t-1
7.002.812
[ 1.33440]
0.043388
[ 0.88598]
∆((Man_Serv)t-1
4.470.435
[ 2.09442]
-0.372724
[-1.87131]
∆(GOV)t-1
0.474128
[ 2.74073]
-0.000591
[-0.36601]
∆(OPENESS)t-1
-0.871656
[-2.93791]
-0.004172
[-1.50676]
∆(Product)t-1
5.024.515
[ 0.19518]
-0.138463
[-0.57638]
∆(Inflows)t-1
1.609.648
[ 0.76659]
-0.011857
[-0.60512]
C
0.312818
[ 0.26780]
-0.024684
[-2.26454]
SummaryStatistics
R-squared
0.676692
0.393168
Adj. R-squared
0.572772
0.198115
S.E. equation
6.847.933
0.063902
137
Residual multivariate serial correlation analysis: Lagrange Multiplier (LM) Test
Lags
LM-Stat
Prob
1
2
3
4
5
6
7
8
9
10
49.91849
54.62263
45.64108
53.65937
38.06780
28.57306
32.72669
46.97774
48.38471
38.00304
0.4366
0.2695
0.6101
0.3003
0.8708
0.9913
0.9643
0.5555
0.4980
0.8725
Null Hypothesis: no serial correlation at lag order h.Probs from chi-square with 49 df.
Residual multivariate normality tests: Null Hypothesis: residuals are multivariate
normal.
Component Jarque-Bera
Joint
11.85549
df
Prob.
14
0.6179
Orthogonalization is based on Cholesky (Lutkepohl)
VEC Residual Heteroskedasticity Tests: No Cross Terms
Joint test:
Chi-sq
df
Prob.
533.0200
560
0.7880
138
Chapter 5
Conclusions
In this concluding chapter I will first present an executive summary of my key findings
in the section Overall Conclusion. Then in the section Future Development, I make
suggestions for potential future research in the final remarks. Finally, I report on the
seminars, congresses, awards and articles that resulted from this thesis.
5.1. Overall Conclusion
This thesis contains three essays that relate, in one way or another, to the evolution of
raw material prices: the forces that drive commodity prices on an international scenario,
the new role of uncertainty in determining such prices, and the impact of commodity
booms in a commodity-dependent economy such as Colombia.
The first essay reports interesting results, evaluating non energy price co-movement and
the variables that determine such co-movement in the short term. The study found that
after late 2003, not only is there greater synchronization among non-energy commodity
prices, but also, notably, a considerable increase in the role of uncertainty as a
determinant of commodity price co-movement. These findings are important because
139
they evidence the transformation that is taking place in the commodity market,
especially in short term price determination, possibly enhanced by the financialization
process. Since late 2003, market uncertainty has played a larger role in explaining nonenergy fluctuation than fundamentals such as the real exchange rate and the real interest
rate, and this confirms that the commodity market is in transformation. As previously
explained in the thesis, since late 2003 there has been an extraordinary increase not only
in the investment driven towards the commodities market, but also in the number of
participants or financial investors, which is known as financialization. Although this
essay does not examine a direct relationship between the financialization of the
commodity market and non-energy raw material price formation, it does give evidence
which confirms that, at least in a short term context, real prices are behaving more as
speculative assets.
Given the increase in the common evolution of non-energy raw material prices after
2004, in the second essay the predictive content of the co-movement either of a large
range of commodities, or the co-movement within a specific category of raw material
prices is evaluated. The essay reports success in using small scale factor models in
forecasting the nominal price of non-energy commodity changes on a monthly basis.
Therefore, communalities of commodities in the same category, estimated by the
Kalman filter, can be useful for forecasting purposes. Notably, category communalities
in oils and protein meals, as well as metals, seem to substantially improve the
forecasting performance of the random walk model.
In contrast, co-movement in
extensive data of commodity prices, estimated through principal components, has poor
predictive power over non-energy commodity prices, compared to the small-scale
factors.
These findings are important since they prove that the co-movement of
140
commodities with similar characteristics has much better predictive power over nonenergy commodity prices, than the co-movement of a large range of commodity prices
data.
Finally, the last essay tackles the issue of raw material prices from a different angle.
This article does not concentrate on analyzing the global determinants of the common
patterns of prices as the first article, but evaluates the implications of the evolution of
these prices in a small commodity-exporting country such as Colombia. Specifically,
the article focuses mainly on evaluating whether, in the long run, resource booms have
brought about the following negative implications in the Colombian economy: a real
appreciation of local currency, which harms competitiveness, and a decrease of the
manufacturing sector related to that of non-tradable goods or services (relative deindustrialization). The results of this paper are important not only because they give
solid empirical evidence against the de-industrialization hypothesis due to the negative
effects of a resource boom, but also because the results give clues to policymakers on
how to prevent future negative effects of commodity price booms, since they show that
the major source of pressure on real exchange rate is public spending.
5.2. Future development
The empirical results found in this thesis regarding the importance of market
uncertainty in non-energy price determination, provides a solid foundation for
beginning a line of research attempting to explain the new role of market uncertainty
and expectations in the commodity markets at all levels. I will give some key issues of
a research agenda derived from the thesis. First, an avenue for future research relates to
141
the use of high frequency data to analyze the influence of macroeconomic and financial
variables, especially market uncertainty, as well as a stock market index in non-energy
commodity prices. As raw material prices may have characteristics of a financial asset,
they may react more quickly to all kinds of information. Therefore, it would be
interesting to know the extent and magnitude of shocks in market uncertainty, and in the
stock index over non energy commodity prices on a daily basis. Second, I identify a
lack of theoretical research aimed at explaining the empirical relationship between
uncertainty and the commodity market that documented in the thesis. Specifically, there
is a need for a unifying theory of short term price formation that takes into account the
new importance of co-movement between different commodities (and between
commodities and other assets) and uncertainty, as well as traditional factors such as
macroeconomic and commodity-specific fundamentals.
Third, as the second essay
evaluates the predictive content of co-movement in non-energy commodity prices, the
logical extension would be to assess the role of uncertainty in commodity markets as a
possible source of predictability. A fourth issue of interest is to evaluate the volatility
spillovers between the uncertainty measure, the equity market and the commodity
market. In the first essay, the first moments or return linkages among variables are
evaluated, however, volatility spillover effects are also important for analyzing market
integrations. More importantly, although I evaluate the long term effects of a resource
boom in the real exchange rate and the relative manufacturing output, the role of
volatility spillovers between the uncertainty measure, the commodity market, and
finally the output growth in commodity-dependent countries have not been researched.
Commodity prices are known as having high volatility related to manufacturing, which
prompts the following questions: does financialization in the commodity market
142
increase such volatility? And how would that impact terms of trade and the economy in
general in a commodity-dependent country?
5.3. Dissemination of results
The different essays that constitute the central body of this dissertation have been
presented at various international and national meetings. In what follows, I provide a
list of the congresses and seminars where the essays have been presented.
Congresses

Annual Meeting of the Association of Southern European Economic
Theorists, ASSET 2013.
November, 7th, 8th and 9th 2013 in Bilbao, Spain.
“What drives co-movement in commodity prices?”.

20th Annual Conference of the European Association of Environmental and
Resource Economists, EAERE.
June, from the 26th to the 29th 2013 in Toulouse, France.
I presented “Long-term links between raw materials prices, real exchange
rate and relative de-industrialization in a commodity dependent economy.
Empirical evidence of `Dutch disease’ in Colombia”.

XVI Encuentro de Economía Aplicada, EEA.
June, 6th and 7th 2013 in Granada, Spain.
I presented “Dutch Disease in Colombia. Empirical evidence”.

10th Eurasia Business and Economics Society conference, EBES.
May, 23th, 24th and 25th 2013 in Istanbul, Turkey.
I presented “What drives co-movement in commodity prices?”.
143

6th CSDA International Conference on Computational and Financial
Econometrics.
December, 1th , 2th and 3th 2012 in Oviedo, Spain.
"Evidence of excess of co-movement in commodity prices at high
frequencies"
Seminars

Seminario de investigación. Departamento de Análisis Económico:
Economía Cuantitativa. Universidad Autónoma de Madrid.
December, 5th 2013.
“The dynamics of co-movements and the role of uncertainty as a driver of
non-energy commodity prices”.

Seminario del Máster de Análisis Económico Aplicado. Universidad de
Alcalá.
November, 29th 2013 in Alcala, Spain.
“What moves non-energy raw material prices?”.

Seminario de Investigación de Economía. Pontificia Universidad Javeriana
de Cali, Colombia.
July, 24th 2013 in Cali, Colombia
“Evidencia empírica de enfermedad Holandesa en Colombia”.

Seminario de investigación. Departamento de Análisis Económico:
Economía Cuantitativa. Universidad Autónoma de Madrid.
April, 2th 2014.
Thesis seminar
144
Working papers

Poncela, P., Senra, E. and Sierra, L. (2012) Is the boost in oil prices affecting
the appreciation of real exchange rate?: empirical evidence of `Dutch
Disease´ in Colombia, Documentos de Trabajo, No. 694, Fundación de las
Cajas de Ahorros, FUNCAS, Madrid.

Poncela, P., Senra, E. and Sierra, L. (2013) The dynamics of co-movement
and the role of uncertainty as a driver of non-energy commodity prices,
Documento de Trabajo No. 733, Fundación de las Cajas de Ahorros,
FUNCAS, Madrid.
Awards

Premio Estímulo a la Investigación, FUNCAS, 2013.

Grant assignment of 1.500 euros for participation in the EAERE 20th Annual
Conference.
145
Conclusiones
En el presente capítulo se suministra un resumen de las principales conclusiones en la
sección Conclusiones Generales. Posteriormente, en la sección Desarrollos Futuros, se
describen las líneas de investigaciones futuras derivadas de los trabajos contenidos en
esta tesis. Por último, se reportan los seminarios, congresos y artículos de investigación
derivados de la tesis.
Conclusiones Generales
La tesis contiene tres capítulos que se relacionan con la evolución de los precios de las
materias primas: las fuerzas que impulsan los precios de los productos básicos en un
escenario internacional, el nuevo papel de la incertidumbre en la determinación de
dichos precios, el contenido predictivo del movimiento conjunto en los precios, y el
impacto de períodos de auge en los precios de las materias primas en una economía
pequeña y dependiente de estos recursos, como es el caso de Colombia.
146
El primer artículo reporta resultados interesantes cuando evalúa el movimiento conjunto
en los precios de las materias primas no energéticas y las variables que determinan tal
movimiento conjunto en el corto plazo. El estudio encontró que a partir de finales del
año 2003, no sólo hay una mayor sincronización entre los precios de las materias primas
no energéticas, sino que también y sobre todo, la importancia de la incertidumbre como
factor determinante del movimiento conjunto
aumentó considerablemente en este
periodo. Estos hallazgos son importantes porque evidencian la transformación que está
teniendo lugar en el mercado de materias primas, especialmente en la determinación de
los precios a corto plazo, tal vez reforzada por el proceso de financialización. El hecho
de que a partir del año 2003 la incertidumbre de mercado ha desempeñado un papel más
importante en la explicación de las fluctuaciones en los precios, que variables
fundamentales como el tipo de cambio real y la tasa de interés real, confirma el hecho
de que el mercado de las materias primas está en transformación. Como se explica en la
tesis, la financialización en el mercado de materias primas, a partir de finales de 2003,
trajo consigo un aumento extraordinario de la inversión impulsada por el mercado de
materias primas y del número de participantes o de los inversores financieros. Aunque
este artículo no prueba la relación directa entre la financialización en el mercado de
materias primas y la formación de los precios de materias primas no energéticas, sí
provee evidencia empírica a favor de que, al menos en un contexto de corto plazo, los
precios reales se asemejan a activos especulativos.
Dado el aumento en el nivel de sincronización de los precios de las materias primas a
partir del año 2004, en el segundo artículo se evalúa el contenido predictivo del comovimiento, ya sea teniendo en cuenta una gran cantidad de series de precios de
materias primas, o el movimiento conjunto
existente dentro de cada una de las
147
categorías de materias primas no energéticas. En el artículo se reportan resultados
exitosos en cuanto a la capacidad predictiva de los modelos factoriales dinámicos a
pequeña escala, los cuales tienen en cuenta las comunalidades de las materias primas
dentro de una categoría. Por lo tanto, el movimiento conjunto de productos dentro de la
misma categoría, que se estima mediante el filtro de Kalman, puede ser útil para
predecir la inflación de las materias primas no energéticas a un horizonte de un mes.
Cabe destacar que para las categorías aceites y harinas proteicas, así como los metales,
la capacidad predictiva de los modelos factoriales dinámicos a pequeña escala es
ampliamente superior frente a los pronósticos generados por una caminata aleatoria. Por
el contrario, movimiento conjunto de una gran cantidad de series de inflaciones en
precios de las materias primas que se estima a través de componentes principales, tiene
escaso poder predictivo sobre los precios de las materias primas no energéticas, en
comparación con los modelos factoriales de pequeña escala. Estos hallazgos son
importantes ya que demuestran que el movimiento conjunto de materias primas con
características similares tiene mejor poder predictivo sobre los precios de los productos
básicos no energéticos frente al movimiento conjunto de una amplia gama de datos de
precios de los productos básicos.
Finalmente, en el último artículo se aborda el tema de los precios de las materias primas
desde un ángulo diferente. Este artículo no se centra en el análisis de los determinantes
globales de los patrones comunes de precios como lo hace el primero, sino que evalúa
las implicaciones de la evolución de estos precios en un país pequeño exportador de
materias primas. En concreto, el artículo se centra en evaluar si en el largo plazo booms
en los recursos exportados han dado lugar a las siguientes consecuencias negativas para
la economía colombiana: una apreciación real de la moneda local, lo que perjudica la
148
competitividad, y una disminución en la producción manufacturera en relación a la de
bienes o servicios no comercializables (desindustrialización relativa).
Los resultados de este trabajo son importantes no sólo porque dan una sólida evidencia
empírica en contra de la hipótesis de la desindustrialización a causa de los efectos
negativos de un auge de recursos, sino también porque los resultados dan herramientas a
los responsables políticos sobre la manera de prevenir futuros efectos negativos de
booms en los productos básicos exportados, ya que muestra que la principal fuente de
presión sobre el tipo de cambio real es el gasto público.
Desarrollos futuros
Los resultados empíricos encontrados en esta tesis sobre la importancia de la
incertidumbre del mercado en la determinación de los precios de materias primas no
energéticas, proporciona una base sólida para iniciar una línea de investigación que se
proponga explicar el nuevo papel de la incertidumbre del mercado en los mercados de
materias primas a todos los niveles. A continuación se presentan algunos temas clave de
una posible agenda de investigación derivada de la tesis. Una primera línea de
investigación futura tiene que ver con el uso de datos de alta frecuencia para analizar la
influencia de las variables macroeconómicas y financieras, especialmente la
incertidumbre de mercado, en los precios de las materias primas no energéticas. Debido
a que los precios de las materias primas pueden tener características de un activo
financiero es normal que reaccionen más rápidamente a todo tipo de información. Por lo
tanto, resulta interesante conocer la longitud y el tamaño de los choques en la
incertidumbre del mercado sobre los precios de las materias primas no energéticas a una
149
frecuencia diaria. En segundo lugar, se identificó un vacío teórico en cuanto a modelos
que ayuden a explicar la relación empírica entre la incertidumbre y los precios de las
materias primas no energéticas, la cual se documenta en la tesis. Para precisar, existe
necesidad de una teoría unificadora sobre la formación de precios a corto plazo que
tenga en cuenta la nueva importancia del movimiento conjunto
entre diferentes
materias primas (y entre las materias primas y otros activos) y la incertidumbre, así
como los factores tradicionales, entre ellos variables macroeconómicas y
factores
específicos, propios de cada mercado.
En tercer lugar, dado que el segundo artículo evalúa el contenido predictivo del
movimiento conjunto en los precios de las materias primas no energéticas, la extensión
lógica sería la de evaluar el papel de la incertidumbre en los mercados de productos
básicos como una posible fuente de previsibilidad. Un cuarto tema de interés está
relacionado con la evaluación de los efectos o conexiones entre la medida de la
incertidumbre y la volatilidad del mercado de acciones y el mercado de productos
básicos. En el primer artículo se analizaron los primeros momentos o los vínculos entre
los retornos de las variables, sin embargo, los efectos de derrame de la volatilidad,
spillover effects, también son importantes para el análisis de las integraciones de
mercado.
Más importante, aunque evalué los efectos en el largo plazo de auges en las materias
primas exportadas sobre el tipo de cambio real y de la producción manufacturera
relativa, la importancia de los efectos contagio, spillover effects, entre la medida de
incertidumbre, la volatilidad en el mercado de materias primas, y finalmente, la
volatilidad en el crecimiento de la producción en los países dependientes de productos
básicos, no se ha investigado hasta el momento. Los precios de las materias primas son
conocidos por tener una alta volatilidad debido a su proceso de producción o extracción,
150
lo cual anima a las siguientes preguntas: ¿la financialización en el mercado de materias
primas ha aumentado tal volatilidad? ¿Cómo afecta la financialización en el mercado de
materias primas los términos de intercambio y la economía en general de un país
dependiente de los productos básicos?.
Difusión de Resultados
Los capítulos que conforman el cuerpo central de la tesis han sido presentados en
diferentes congresos a nivel nacional e internacional. A continuación se provee una lista
de los seminarios y congresos donde estos han sido presentados.
Ponencias en congresos

Annual Meeting of the Association of Southern European Economic
Theorists, ASSET 2013.
7 al 9 de noviembre de 2013 en Bilbao, España.
“¿What drives co-movement in commodity prices?”

20th Annual Conference of the European Association of Environmental and
Resource Economists, EAERE.
26 al 29 de junio de 2013 en Toulouse, Francia.
“Long-term links between raw materials prices, real exchange rate and
relative de-industrialization in a commodity dependent economy. Empirical
evidence of `Dutch disease’ in Colombia”.

XVI Encuentro de Economía Aplicada, EEA.
151
6 y 7 de Junio de 2013 en Granada, España.
“Dutch Disease in Colombia. Empirical evidence”.

10th Eurasia Business and Economics Society conference, EBES.
23 al 25 de mayo de 2013 en Estambul, Turquía.
“What drives co-movement in commodity prices?”.

6th CSDA International Conference on Computational and Financial
Econometrics.
1al 3 de diciembre de 2012 en Oviedo, España.
"Evidence of excess of co-movement in commodity prices at high
frequencies".
Presentaciones en seminarios de investigación

Seminario de investigación. Departamento de Análisis Económico:
Economía Cuantitativa. Universidad Autónoma de Madrid.
5 de diciembre de 2013 en Madrid, España.
“The dynamics of co-movements and the role of uncertainty as a driver of
non-energy commodity prices”.

Seminario del Máster de Análisis Económico Aplicado. Universidad de
Alcalá.
29 de noviembre de 2013 en Alcalá, España.
“What moves non-energy raw material prices?”.

Seminario de Investigación de Economía. Pontificia Universidad Javeriana
de Cali, Colombia.
24 de Julio de 2013 en Cali, Colombia.
152
“Evidencia empírica de enfermedad Holandesa en Colombia”.

Seminario de investigación. Departamento de Análisis Económico:
Economía Cuantitativa. Universidad Autónoma de Madrid.
2 de abril de 2014.
Documentos de trabajo

Poncela, P., Senra, E. and Sierra, L. (2013) The dynamics of co-movement
and the role of uncertainty as a driver of non-energy commodity prices,
Documento de Trabajo No. 733, Fundación de las Cajas de Ahorros,
FUNCAS, Madrid.

Poncela, P., Senra, E. and Sierra, L. (2012) Is the boost in oil prices affecting
the appreciation of real exchange rate?: empirical evidence of `Dutch
Disease´ in Colombia, Documentos de Trabajo, No. 694, Fundación de las
Cajas de Ahorros, FUNCAS, Madrid.
Reconocimientos

Premio Estímulo a la Investigación, FUNCAS, 2013.

Beca de 1.500 euros para participar en la conferencia anual de EAERE en 2013.
153