Investigation of the Linkages among Agricultural, Oil, and Exchange Rate Markets

Transcription

1 Investigation of the Linkages among Agricultural, Oil, and Exchange Rate Markets Julieta Frank University of Manitoba Philip Garcia University of Illinois at Urbana-Champaign CAES Risk Management and Commodity Market Analysis Workshop April 12, 2010

2 Background Substantial changes in agricultural prices plus the increasing importance of energy markets has motivated the development of more comprehensive models to understand the market and macroeconomic linkages. Exchange rates are often viewed as good indicators of commodity prices because they capture market expectations of future price dynamics, the effect on future exports, and in future exchange rate values. However, forecasting power of exchange rates is unclear. Direction of causality from exchange rates to prices is mixed and highly dependent on the specification of the model used.

3 Background Failure to include important markets can hamper estimation of the linkages between markets. VAR framework used to identify market relationships may be problematic due to overparametrization, with consequences in the error matrix and in impulse response interpretation. Understanding of market relationships and their magnitude could be used to develop marketing strategies and to provide policy makers with added insights into the effect of new policies to related markets.

4 Objective Identify market relationships in a systematic manner, in light of the relatively recent observed market changes Investigate corn, wheat, cattle, hogs, and crude oil prices, and exchange rates. Examine time series properties of the price series Use restricted VAR framework based on model likelihood measures

5 Crop prices Wheat ($/10kg) Corn ($/bu)

6 Livestock prices Cattle ($/cwt) Hogs ($/cwt)

7 Exchange rate and Crude oil prices US dollar index Crude oil ($/barrel)

8 Methods Identification of breakpoint Zivot-Andrews unit root with one breakpoint Dickey-Fuller unit root test for each period Test for cointegration Johansen rank test

9 Methods VAR Model Δp t = A 1 Δp t-1 + A 2 Δp t A j Δp t-j + u t where: Δp t = (Δp 1t,, Δp Kt ) is a vector of K series K= {corn, wheat, cattle, hogs, crude oil, exchange rate} A j : (K x K) matrix of coefficients u t = (u 1t,,u Kt ) is a vector of K white noise residual series u t ~ (0, Ʃ u ), and Ʃ u = E(u t u t )

10 Methods VEC model Δp t = Пp t-1 + Г 1 Δp t Г j-1 Δp t-j+1 + u t where: П: (K x K) matrix of long-run coefficients such that П = αβ, α: (K x r) matrix of speed of adjustment coefficients β: (K x r) matrix of cointegrating relations Г j : (K x K) matrix of short-run coefficients u t = (u 1t,,u Kt ) is a vector of K white noise residual series

11 Methods Model selection (VAR and VEC) Number of lags based on AIC and diagnostic tests Sequentially elimination: deletes regressors which lead to the largest reduction of the AIC until no further reduction is possible (only a single regressor is eliminated in each step). Estimation VAR: Feasible GLS because restricted model has different number of regressors in each equation VEC: Two stage estimation, first estimate the cointegration matrix using Johansen s reduced rank, then feasible GLS Diagnostic tests VAR and VEC: univariate and multivariate versions of Portmanteau, LM, ARCH-LM, and JB

12 Impulse response function Methods Use MA representation of vector autoregression y t = Φ 0 u t + Φ 1 u t-1 + Φ 1 u t Coefficients reflect the responses to impulses hitting the system Coefficients are estimated recursively using VAR coefficients, Φ s = Ʃ s Φ s-j A j VECM is written in VAR (levels) form to compute the impulse response function Different orderings of the variables produce different response coefficients; use most influential variable first Confidence intervals are constructed using bootstrap methods (use 1,000 draws w/replacement to generate bootstrap residuals)

13 Methods Forecast error variance decomposition (FEVD) Compute variance of the forecast error using coefficients of the impulse response matrix, σ k2 (h) FEVD gives the percentage contribution of variable j to the h-step forecast error variance of variable k

14 Data Weekly cash prices for period Jan 98 Nov 09 All prices converted to logs Corn (C): Central Illinois Wheat (W): St. Louis Cattle (LC): Texas/Oklahoma (USDA) Hogs (LH): Iowa/Minnesota (USDA) Crude oil (CL): West Texas Intermediate (WTI) Exchange rate (EX): US dollar relative to basket of world currencies

15 Results Point of break: Sep 06, periods 1 and 2 Unit root Period 1: W, LC, LH, CL, EX non-stationary; all differenced series are stationary Period 2: All prices are I(1) Rank Test Period 1: no cointgrating relationship Period 2: 1 cointegrating relationship

16 VAR Model Period 1 ΔEX ΔCL ΔC ΔW ΔLC ΔLH Const ΔEXt *** *** ΔEXt ΔCLt *** ΔCLt * ΔCLt *** ΔCt *** ΔCt ** ΔWt * 0.169** ΔWt ** ΔWt ** ΔLCt *** 0.426*** ΔLCt * *** ** ΔLCt *** ΔLHt *** ΔLHt * * 0.095** * ΔLHt ***

17 VAR model Period 1 Predominately, the markets are mostly affected by the lags in their own prices Effect from EX to C emerged, with no effect on other ag markets Crude oil has limited effect on ag markets Significant effects among ag markets

18 Shock of Exchange Rate Period 1 Corn Wheat Cattle Hogs

19 Shock of Crude oil Period 1 Corn Wheat Cattle Hogs

20 Point estimates: Impulse response Period 1 A one standard deviation increase in the exchange rate leads to a decrease in corn, wheat, and cattle prices which quickly dies out after 5 or 6 weeks. A cheaper US dollar leads demand (price) increases. A one standard deviation increase in crude oil leads to an initial decrease of ag prices with a subsequent increase, reaching equilibrium levels after 7 or 8 weeks. Increase in oil prices is transmitted to ag markets via increases in production costs.

21 Forecast error variance decomposition Period 1 Corn Wheat Cattle Hogs

22 VEC Model Period 2 ΔEX ΔCL ΔC ΔW ΔLC ΔLH α *** *** β *** * *** ΔEXt *** ΔEXt ** ΔEXt * 0.88** 0.226* ΔCLt ** * 0.064*** ΔCLt ** * ΔCLt * 0.238*** ΔCt ** 0.197*** ΔCt * ΔWt *** ΔWt ** ΔWt ** ΔLCt ** ΔLCt *** ΔLHt *** ΔLHt *** ΔLHt **

23 VEC model Period 2 EX has no effect on ag markets (only marginally in LH), but has larger effect on CL relative to period 1 The effects of CL appear to be more pronounced in W, LC, and LH, and limited in C Ag markets seem to have a reduced own lag structure

24 Shock of Exchange Rate Period 2 Corn Wheat Cattle Hogs

25 Shock of Crude Oil Period 2 Corn Wheat Cattle Hogs

26 Point estimates: Impulse response Period 2 In corn, cattle, and hogs, a one standard deviation increase in the exchange rate leads to price decreases that die out at a lower equilibrium level. Corn and wheat show an initial positive response to a shock of crude oil that dies out after 4 or 5 weeks. In livestock, the initial positive response dies out at a higher level.

27 Forecast error variance decomposition Period 2 Corn Wheat Cattle Hogs

28 Conclusions Agricultural markets have undergone a structural change in both the level of prices and in their dynamic relationship with other markets Relationships among agricultural, oil, and macro markets changed In period 2 positive shocks of EX may cause persistent price reductions in ag markets. Effect of crude oil is more pronounced in period 2. This may be due to competition with biofuels in output markets In the second period, EX shocks explain a fraction of about 10% of the variance in corn prices.

29 Thanks!