1. 1 Returns in commodities futures markets and financial speculation: a multivariate GARCH approach Joint with Matteo Manera and Ilaria Vignati Università di Pavia 29/11/2011

2. 2 Motivation From 2000 onwards we observe a number of severe changes in financial markets. There has been a sharp increase (at least until the crisis) in: –Energy prices –Food prices –The numbers of financial participants (both hedgers and speculators) in the futures markets. These stylized facts have lead to claims that: –speculators drive energy and food prices –speculators affect commodities’ volatility –oil/energy prices induced an increase in food prices

3. 3 Energy futures prices

4. 4 Agriculture futures prices

5. 5 Motivation The aim of this paper is to answer the following research questions: Is financial speculation significantly related to returns in energy and non-energy commodities? Do macroeconomic factors explain returns in energy and non-energy commodities? Are there spillovers between energy and non-energy markets?

6. 6 Literature review oil prices and macroeconomic factors The economic theory suggests few factors that should affect commodities futures returns: –Treasury bill yields –Equity dividend yields –Junk bond premium (Sadorsky 2002, Chevallier 2009)

7. 7 Literature review oil prices and speculation Several papers suggest that the increasing presence of speculators in oil future markets could explain the spike in prices in 2007-2008 (Masters White 2008, Medlock III and Myers Jaffe, 2009) However, empirical evidence generally shows that there is not a relationship between the two phenomena (Irwing Sanders 2010, Büyükşahin Harris, 2011)

8. 8 Literature review oil prices and other commodities prices Oil prices have triggered the increase in food and agricultural prices through costs of energy inputs (Alexandratos, 2008, FAO, 2009) Increasing interaction between oil/energy and food/non-energy prices is due to the growing demand of bio-fuels (Du et al., 2009, Baffes and Haniotis, 2010) “Excess co-movement hypothesis” theory (Pindyck and Rotemberg, 1990, Deb, Trivedi and Varangis, 1996) Spillover effects in futures markets have implications in terms of cross hedging and cross speculation (Malliaris and Urrutia, 1996)

9. 9 Data description Dependent variable: returns of continuous series of futures in: –4 energy commodities (oil, gasoline, heating oil, natural gas) –5 non-energy commodities (corn, oats, soybean oil, soybeans, wheat) Time period: 1986-2010 Frequency: weekly Source: Datastream, CFTC (U. S. Commodity Futures Trading Commission), FRED (Federal Reserve Economic Data)

10. 10 Data description explanatory variables Macroeconomic factors: –Return on the annual yield on the 90-day T-bill –Returns of S&P 500 Index –Junk bond yield= (return on the annual yield on Moody’s long-term-BAA-rated corporate bonds) – (return on the annual yield on Moody’s long-term- AAA-rated corporate bonds)

11. 11 Data description explanatory variables Speculation: Working’s T index (Working 1960) = proxies the excess of speculation relative to hedging SS = Speculation Short SL = Speculation Long HS = Hedging Short HL = Hedging Long

12. 12 Has speculation in energy futures market increased?

13. 13 Has speculation in agriculture futures market increased?

14. 14 Speculation (Working’s T index) across different commodities 1986-2005: energy commodities have mean values lower than non- energy one For oil and natural gas the increase is bigger than for other commodities 2006-2010: means increase

15. 15 The model specification: returns of each commodity are a function of –Treasury bill yields –Equity dividend yields –Junk bond premium –Measures of speculation (Working’s T index) The econometric strategy: 1.Test for stationarity of series 2.Estimate the model using OLS and test for autocorrelation and ARCH effects in the residuals 3.If present, move to a GARCH (1,1) and/or ARMA specification Then we extend the analysis to a multivariate GARCH model The econometric model

16. 16 Stationarity tests Macroeconomic variables and futures generally present a unit root, and are transformed taking the first difference of the logs to gain stationarity Working’s T index is stationary and considered in levels

17. 17 Results Univariate GARCH(1,1) Apart from crude oil and oats, Working’s T index is negative or not significant

18. 18 Multivariate GARCH The analysis developed so far is univariate. It is interesting to estimate a system where returns for different commodities are jointly estimated, allowing for conditional variances and covariances (spillover effects) Possible contribution of this econometric exercise: –Indentify if and how returns on oil/energy futures prices are related to food/non-energy commodities futures prices Model implemented: CCC model and DCC model, which allows for time-varying conditional correlations

19. 19 Multivariate GARCH General multivariate GARCH: = mx1 vector of dependent variables C = mxk matrix of parameters = kx1 vector of independent variables = Cholesky factor of the time varying conditional covariance matrix of the disturbances = mx1 vector of i.i.d. innovations = conditional correlations R = matrix of unconditional correlations of the standardized residuals

20. 20 Multivariate GARCH CCC model: DCC model: is a diagonal matrix of conditional variances in which each evolves according to a univariate GARCH process defined as in the univariate analysis as R = matrix of time-invariant unconditional correlations of the standardized residuals = matrix of time-varying conditional correlations = mx1 vector of standardized residuals = parameters for dynamics of

21. 21 Multivariate GARCH We group the commodities in two subgroups: 1.“Fuels”: it includes the four energy commodities and soybean oil (spillovers between energy markets and a bio-fuel) 2.“Agriculture”: it includes the five agricultural commodities (spillovers between food markets and a bio-fuel) We consider one more group: 3.“Agriculture + factor of energy commodities”: it includes the five agricultural commodities and a factor of the energy ones (spillovers between energy and agricultural markets)

22. 22 Results CCC model – “Fuels” group

23. 23 Results CCC model – “Fuels” group Conditional correlations:

24. 24 Results DCC model – “Fuels” group

25. 25 Results DCC model – “Fuels” group Dynamic conditional correlations – Mean tests:

26. 26 Dynamic conditional correlations graphs “Fuels” group

27. 27 Results CCC model – “Agriculture” group

28. 28 Results CCC model – “Agriculture” group Conditional correlations:

29. 29 Results DCC model – “Agriculture” group

30. 30 Results DCC model – “Agriculture” group Dynamic conditional correlations – Mean tests:

31. 31 Dynamic conditional correlations graphs “Agriculture” group

32. 32 Do energy markets influence non-energy commodities? It is not feasible to estimate jointly fuels and agriculture markets due to the large number of parameters to be estimated (no convergence achieved) We summarize energy markets into one variable (“energy”) using principal factor analysis This allows to investigate if “energy” influence agriculture prices, as often claimed

33. 33 Results CCC model – “Agriculture + factor energy” group

34. 34 Results CCC model – “Agriculture + factor energy” group Conditional correlations:

35. 35 Results DCC model – “Agriculture + factor energy” group

36. 36 Results DCC model – “Agriculture + factor energy” group Dynamic conditional correlations – Mean tests:

37. 37 Dynamic conditional correlations graphs “Agriculture + energy factor” group

38. 38 What’s next? Recursive estimates allow to investigate the effect of Working’s T index over time

39. 39 Conclusions Speculation −Working’s T index is poorly significant in both univariate and multivariate models suggesting that speculation is not relevant in explaining commodities’ returns Macroeconomic factors −equity returns are generally positive and significant Spillovers −There are spillovers within the two groups of commodities −Between groups (“energy” factor): high conditional correlations between energy factor and agricultural commodities after 2005

40. 40 Thanks!