Causal dynamics within the food-versus-fuel nexus: parametric and - - PowerPoint PPT Presentation

Faculty of Economics and Business

Causal dynamics within the food-versus-fuel nexus: parametric and nonparametric causality testing in US spot markets

Thesis presentation Sebastian Fischer May 22nd, 2013

Introduction Data Methodology Key Findings Literature Link Recent development Motivation

FAO Food Price Indices 2000−2012

◮ Upward trend in agricultural commodity prices since 2000 ◮ Biofuel mandates strengthen link between energy and commodity

markets → exponential growth in biofuel production

◮ Financialization of agricultural commodities → own asset class ◮ Food price crisis between 2006-2008; price levels reissued in 2011 ◮ Impacts on terms of trade, real incomes and fiscal positions 2 / 12

Introduction Data Methodology Key Findings Literature Link Recent development Motivation ◮ Link between energy and agricultural commodity markets ◮ Structure of this (potential) link ◮ Changing dynamics after introducing policies in favour of biofuels ◮ Concrete policy advice with respect to biofuel legislation ◮ Existing literature focuses on weekly observations, Granger causality

within the parametric (linear) framework and limited number of variables

3 / 12

Introduction Data Methodology Key Findings Literature Link ◮ Two samples on nominal spot price series of key agricultural

commodities and liquid fuels: log-levels log(pt) and log-returns rt = log(pt) − log(pt−1)

◮ 01/2000-12/2012; n=3,268; P1 2000-2006, P2 2007-2012

corn, wheat, soybean, sugar, diesel, gasoline, crude oil

◮ 08/2007-12/2012; n=1,368; P1 2007-2009, P2 2010-2012

diesel and gasoline replaced by ethanol and biodiesel

◮ Similar characteristics as returns from financial markets: zero mean,

skewed and excess kurtosis. log(pt) ∼ I(1), rt ∼ I(0); high positive (linear) correlations among fuel and commodity series; significant correlations between energy and commodity price series, but smaller magnitude.

4 / 12

Introduction Data Methodology Key Findings Literature Link ◮ Parametric Granger causality testing based on Granger (1969) by

means of VAR or VECM specification due to Engle and Granger (1987)

◮ Evidence for nonlinearity by using BDS test of Brock, Dechert and

Scheinkman (1996)

◮ Nonparametric Granger causality testing due to Diks and Panchenko

(2006): focus on residual series of conditional mean models

◮ Pairwise and full-variate implementation ◮ Stepwise procedure: Conditional mean model → linear Granger

causality → residual based testing for nonlinear Granger causality uncovers strictly nonlinear relations → multivariate GARCH-filter → remaining Granger causality is not due to risk transmissions

◮ Conditional second moments captured by multivariate GARCH of

Engle and Kroner (1995): full GARCH-BEKK(1,1) µt = E(rt|Ft−1), ǫt = rt − µt = H1/2

t

zt and Ht = E(ǫtǫ′

t|Ft−1) = CC′ + K

• k=1

A′

kǫt−1ǫ

′

t−1Ak + K

• k=1

B′

kHt−1Bk,

(1)

5 / 12

Introduction Data Methodology Key Findings Literature Link ◮ Causality in its most general form can be written as

(Yt+1, · · · , Yt+k)|(FX,t, FY ,t) ≁ (Yt+1, · · · , Yt+k)|FY ,t). (2)

◮ The approach of Granger (1969) amounts to Wald-test in

Xt = c0 +

p

• i=1

αiXt−i +

p

• i=1

βiYt−i + u1t Yt = c1 +

p

• i=1

γiXt−i +

p

• i=1

δiYt−i + u2t.

◮ Diks and Panchenko (2006) restate (2) in terms of conditional

distributions with (X, Y , Z) = (X lx

t , Y ly t , Yt+1) as

fX,Y ,Z(x, y, z) fY (y) = fX,Y (x, y) fY (y) · fY ,Z(y, z) fY (y) (3) and derive test statistic Tn, which is asymptotically distributed as standard normal: Tn(ǫn) = n − 1 n(n − 2) ×

n

• i=1

( fX,Y ,Z(Xi, Yi, Zi) fY (Yi)− fX,Y (Xi, Yi) fY ,Z(Yi, Zi))

6 / 12

Introduction Data Methodology Key Findings Literature Link Causality Robustness Conclusions

Full-variate Nonlinear Causality Sample 1 after GARCH-BEKK filtering1 P1 Diesel Gasoline WTI Corn Wheat Soy Sugar Diesel

• ***

*** Gasoline

• WTI

*** *

• Corn
• **

Wheat **

• Soy

*

• Sugar

*

• P2

Diesel Gasoline WTI Corn Wheat Soy Sugar Diesel

• **

** * ** Gasoline

• ***

* * ** WTI * *

• *

Corn * ** **

• **

* Wheat * ***

• **

Soy ** ** ***

• Sugar

*** ** ** **

• 1 ***, ** and * indicate rejection of the null hypothesis at the 1%, 5% and 10%

level, respectively. If i denotes the row and j denotes the column, causality running from X to Y is represented by the ij-th element; element ji refers to causality running from Y to X.

7 / 12

Introduction Data Methodology Key Findings Literature Link Causality Robustness Conclusions

Full-variate Nonlinear Causality Sample 2 after GARCH-BEKK filtering1 P1 Biodiesel Ethanol WTI Corn Wheat Soy Sugar Biodiesel

• ***

* ** * Ethanol *

• **

WTI *

• ***

Corn

• *

Wheat **

• Soy

*** *

• **

Sugar * * **

• P2

Biodiesel Ethanol WTI Corn Wheat Soy Sugar Biodiesel

• Ethanol
• WTI

**

• Corn
• ***

* Wheat ***

• *

Soy

• Sugar
• 1 ***, ** and * indicate rejection of the null hypothesis at the 1%, 5% and 10%

level, respectively. If i denotes the row and j denotes the column, causality running from X to Y is represented by the ij-th element; element ji refers to causality running from Y to X.

8 / 12

Introduction Data Methodology Key Findings Literature Link Causality Robustness Conclusions ◮ Various multivariate GARCH models have been used to capture

conditional second moments: full and diagonal GARCH-BEKK, Dynamic Conditional Correlation-GARCH (DCC) due to Engle (2002)

◮ Impact of considering three different lag lengths: Akaike (1973) and

Schwarz (1978) information criterion plus fixed lag length of l=5 due to data frequency

◮ Decreased data frequency: weekly data using 5-day averages ◮ Results indicate importance of taking full variance-covariance structure

into account, but no clear-cut solution with respect to lag length selection

◮ Results based on weekly data show decreased absolute number of

causal relations and average significance, but these should be interpreted with caution

9 / 12

Introduction Data Methodology Key Findings Literature Link Causality Robustness Conclusions ◮ Increased causal dynamics during recent years, although

GARCH-BEKK filtering removes many of the causal relations

◮ Might indicate more integrated and efficient markets ◮ Results for sample 2 suggest that link between biofuel and agricultural

commodity markets is weaker than suggested

◮ Policy advice: commodity price buffers might dampen the impacts,

especially for developing countries; re-adjust policy frame to lower pressure on supply side

◮ Improve infrastructure on biofuel data and extend research on

nonparametric causal relations

10 / 12

Introduction Data Methodology Key Findings Literature Link ◮ Stelios D. Bekiros and Cees G.H. Diks.

The relationship between crude oil spot and futures prices: Cointegration, linear and nonlinear causality. Energy Economics, Vol. 5, 2008. Cees Diks and Valentyn Panchenko. A new statistic and practical guidelines for nonparametric granger causality testing. Journal of Economic Dynamics and Control, Vol. 30, 2006. Zibin Zhang, Luanne Lohr, Cesar Escalante, and Michael Wetzstein. Food versus fuel: What do prices tell us? Energy Policy, Vol. 38, 2010. David Zilberman, Gal Hochman, Deepak Rajagopal, Steve Sexton, and Govinda Timilsina. The impact of biofuels on commodity food prices: Assessment of findings. American Journal of Agricultural Economics, 2012.

11 / 12

Introduction Data Methodology Key Findings Literature Link

Data, program code and slides will be made available at http://uva-thesis.de.vu/. Thank you for your attention!

12 / 12