Comovement and the. London School of Economics Grantham Research Institute. Commodity Markets and their Financialization IPAM May 6, 2015

Transcription

1 London School of Economics Grantham Research Institute Commodity Markets and ir Financialization IPAM May 6, / 35

2 generated uncorrelated returns Commodity markets were partly segmented from outside financial markets and from each or. Abbildung: Inflation-Adjusted Performance of in UK December 1969-December Gorton and Rouwenhorst (2006). 2 / 35

3 Why investing in commodity markets? Hedging strategies (lock-in future prices); as a method of diversification commodities had low positive return correlation with each or; commodities had negligible correlations with or financial markets. Abbildung: Correlations of commodity futures returns with stocks, bonds, and inflation, July 1959-December Gorton and Rouwenhorst (2006). 3 / 35

4 , new asset class Since early 2000s, commodity futures have emerged as a popular asset class. The increasing presence of investors allocating money in commodities (derivatives) initiated so called process of financialization of commodities markets. 4 / 35

5 Emerging of new patterns After 2004 we observe an increase in price co-movements within commodities. Correlations of oil and grains. Corn Soy Beans 5 / 35

6 Emerging of new patterns Correlations of oil and soft. Coffee Cocoa 6 / 35

7 The paper in a nutshell - 1 We test differences in price (co)movement building on ory of co-movement by Barberis et. al. (2005). We exploit difference between indexed and off-index commodities considering 25 commodities attempt to complement work by Tang and Xiong (2012). We consider three indexes: S&P GS CI, DJ-UBS CI (heavy energy) and Thomson Reuters CI (equally weighted components). We provide new evidence in support of commodity financialization view. 7 / 35

8 The paper in a nutshell - 2 Using high-frequency data, we compute realized variance and evaluate changes in return volatility and correlation between indexed and off-index commodities. We provide new evidence in support of price volatility spillover effect. 8 / 35

9 Theory of co-movement Availability of commodity index funds and commodity ETF facilitate participation and trading of commodities (new asset class). Changes in funds allocation can cause commodities in fund/index to move toger (Barberis and Shleifer (2003)). One would expect price co-movements of commodities in index to be greater than those of off-index commodities. Barberis et al. (2005) found that stock s listing can significantly increase return correlation of that stock with index. 9 / 35

10 Index inclusion and exclusion Barberis et al (2005) investigate wer addition to S&P 500 (index) leads to a shift in correlation structure of returns. After inclusion: S&P beta should increase and non-s&p (rest of world) beta should decrease. After exclusion: S&P beta should decrease and non-s&p (rest of world) beta should increase. Changes in co-movement should be stronger in more recent data. 10 / 35

11 Alternative view of co-movement drivers Traditional ory: co-movement in prices reflects co-movement in fundamental values. in prices is delinked from fundamentals when: Category view: investors group assets in categories and allocated funds at category level rar n at individual asset level. Habitat view: Traders choose to trade only a subset of all available asset. 11 / 35

12 Our approach Motivated by Barberis et al. (2005) and Tang and Xiong (2012), we test this ory of co-movement. An increase in price co-movement between indexed commodities and index indicates existence of a category/habitat view. When investigating commodity financialization process, choice of 2005 as breakpoint is innocuous. Monthly, weekly, daily and high-frequency data allow construction of reliable measure of changes in return co-movement. 12 / 35

13 Reduced-from model of co-movement 2n commodities grouped into two categories, X and Y. Funds are allocated at level of se categories. Let P i,t := P i,t P i,t 1 represent return. A representation of commodity returns is: where and ( ux,t P i,t = ɛ i,t + u X,t, i X, P j,t = ɛ j,t + u Y,t, j Y u Y,t ) N (( 0 0 ) (, σu 2 1 ρu ρ u 1 )) ɛ t = (ɛ 1,t,..., ɛ 2n,t ) N(0, Σ), i.i.d. over time 13 / 35

14 Testable predictions Prediction 1 Suppose that commodity j, previously a member of Y, is reclassified into X. Then plim of OLS estimate of β j,x in regression P j,t = α j + β j P X,t + ν j,t, (1) as well as plim of R 2 of this regression, increases after reclassification. The presence of commodity j in category X increases covariance of its return with return on category, P X,t, and its beta loading on that return. 14 / 35

15 Testable predictions Prediction 2 Suppose that commodity j, previously a member of Y, is reclassified into X. Then plim of OLS estimate of β j,x in regression P j,t = α j + β j,x P X,t + β j,y P Y,t + ν j,t (2) rises after reclassification while plim of OLS estimate β j,y falls. P X,t in Equation (1) is not a clean measure of sensitivity to u X,t. A substantial part of variation comes from news ɛ t ; P Y,t controls for such news. 15 / 35

16 Commodity Exchange GSCI DJ-UBS Th. Reu Energy WTI crude oil NYMEX Heating Oil NYMEX RBOB NYMEX Natural Gas NYMEX Grains Corn CME Soy beans CME Chicago wheat CME Kansas wheat c KCTB Soybean oil CME Minn. wheat MGE Soybean meal CME Rough rice CME Oats CME Softs Coffee ICE Cotton ICE Sugar ICE Cocoa ICE Lumber CME Orange Juice ICE / 35

17 (cont d) Commodity Exchange GSCI DJ-UBS Th. Reu Livestock Feeder cattle CME Lean hogs CME Live cattle CME Pork bellies CME Metals Gold NYMEX Silver NYMEX Copper NYMEX Platinum NYMEX Palladium NYMEX / 35

18 description We consider 25 commodities (US traded) - except Kansas wheat, Minn. wheat, Palladium. Futures contracts rolled over before expiry to next available contract. Sample data spanning period 9, April 1998 to 24, March ,222 trading day. We have 1-minute frequency and consider only overlapping trading hours from 10:30 to 14:00 NYT. Soybean meal, Rough Rice, Oats (Grains), Lumber and Orange Juice (Softs) and Pork Bellies (Livestock) are off-index commodities. 18 / 35

19 Estimation results - univariate regression We estimate change in βs before and after January 2005 (daily, weekly and monthly). We test Prediction 1 for every commodity (vs. mean of βs and vs. pooled βs). Statistically significant increase in co-movement between indexed non-energy commodities and heavy-energy indexes. Statistically significant increase in co-movement between indexed non-energy commodities and equally weighted index. No significant increase in co-movement between off-index non-energy commodities and equally weighted index. 19 / 35

20 Univariate regression S&P GSCI UBS-DJ CI Th. Reuters CI Commodity β R 2 β R 2 β R 2 WTI crude oil Heating Oil RBOB un. gas Natural Gas Corn Soy beans Chicago wheat Soybean oil Soybean meal Rough rice Oats Coffee Cotton Sugar Cocoa Lumber Orange Juice / 35

21 Univariate regression S&P GSCI UBS-DJ CI Th. Reuters CI Commodity β R 2 β R 2 β R 2 Feeder cattle Lean hogs Live cattle Pork bellies Gold Silver Copper Platinum Tabelle: and denote positive significant difference from zero at 10% and 5% level in two-sided test. Dark grey cell negative and significant at 10%; light grey cell denote difference is not significant at 10%. Rest significant at 1% level. 21 / 35

22 Univariate regression (only oil for energy) S&P GSCI UBS-DJ CI Th. Reuters CI Commodity β R 2 β R 2 β R 2 WTI crude oil % % % Corn % % % Soya beans % % % Chicago wheat % % % Soybean oil % % % Soybean meal % % % Rough rice % % % Oats % % % Coffee % % % Cotton % % % Sugar % % % Cocoa % % % Lumber % % % Orange Juice % % % 22 / 35

23 Univariate regression (only oil for energy) S&P GSCI UBS-DJ CI Th. Reuters CI Commodity β R 2 β R 2 β R 2 Feeder cattle % % % Lean hogs % % % Live cattle % % % Pork bellies % % % Gold % % % Silver % % % Copper % % % Platinum % % % Tabelle: and denote positive significant difference from zero at 10% and 5% level in two-sided test. Dark grey cell negative and significant at 10%; light grey cell denote difference is not significant at 10%. Rest significant at 1% level. 23 / 35

24 Bivariate regression We estimate daily change in βs before and after January 2005 considering index and off-index returns as explanatory variables. Statistically significant increase in co-movement between indexed non-energy commodities and index. No significant increase in co-movement between off-index commodities and index and off-index. between indexed non-energy commodities and off-index increases with heavy-energy indexes... and does not increase with equally weighted index misspecification issue? 24 / 35

25 Bivariate regression S&P GSCI UBS-DJ CI Th. Reuters CI Commodity β ind β off β ind β off β ind β off WTI crude oil Heating Oil RBOB un. gas Natural Gas Corn Soya beans Chicago wheat Soybean oil Soybean meal Rough rice Oats Coffee Cotton Sugar Cocoa Lumber Orange Juice / 35

26 Bivariate regression S&P GSCI UBS-DJ CI Th. Reuters CI Commodity β ind β off β ind β off β ind β off Feeder cattle Lean hogs Live cattle Pork bellies Gold Silver Copper Platinum Tabelle: and denote positive significant difference from zero at 10% and 5% level in two-sided test. Dark grey cell negative and significant at 10%; light grey cell denote difference is not significant at 10%. Rest significant at 1% level. 26 / 35

27 Characteristic and demand effects Alternative explanation: indexed commodities cover larger commodity production. We attempt to address this competing explanation with a matching exercise using soy beans (index), soybean oil (index) and soybean meal (off-index). Soybean meal is a solid residue by-product, flour, created after grinding soybean to extract soybean oil. We observed that Soybean meal exhibits much smaller shifts in betas than indexed soy beans and soybean oil. 27 / 35

28 Non-trading effects Indexed commodities are highly liquid and frequently traded (vs. off-index). might have some spurious upward bias due to non-trading effects. We investigate trading activity of each commodity pre and post 2005 and test non-trading hyposis. We use number of trades per day (up and down ticks) as a proxy for trading activity (10:30-14:00 time interval). Grains, for example, show a decrease in trading activity but indexed grain experienced an increase in co-movement and off-index grains do not co-move with index. 28 / 35

29 Directions of trading activity after 2005 Grains Corn Energy WTI crude oil Soya beans Heating Oil Chicago wheat RBOB unleaded gas Soybean oil Natural Gas Soybean meal Rough rice Livestock Feeder cattle Oats Lean hogs Live cattle Softs Coffee Pork bellies Cotton Sugar Metals Gold Cocoa Silver Lumber Copper Orange Juice Platinum Where indicates decrease and indicates increase. 29 / 35

30 in commodity markets We investigates change in risk transmission in commodity markets using intra-day prices. Abbildung: in energy commodity group. 30 / 35

31 Wishart Autoregressive model is measured using a multivariate realized volatility model, Wishart Autoregressive model (WAR). Y t follows a Wishart process when expected value of Y t+1 is given by: E t (Y t+1 ) = MY t M + KΣ. where for case of (2x2) ( ) m11 m M = 12 m 21 m 22 and M takes care of spillover effect. 31 / 35

32 Spillover ratio The conditional variance of asset 1 can be written as E t 1 [Y 11,t ] = a 1 Y 11,t 1 + b 1 Y 12,t 1 + c 1 Y 22,t 1 + d 1. where b 1 and c 1 capture covariance spillover and volatility spillover. The volatility spillover ratio is: SR 2,1,t = c 1 Y 22,t 1 a 1 Y 11,t 1 + b 1 Y 12,t 1 + c 1 Y 22,t 1 + d 1 The covariance spillover ratio is: SR 12,1,t = b 1 Y 12,t 1 a 1 Y 11,t 1 + b 1 Y 12,t 1 + c 1 Y 22,t 1 + d 1 32 / 35

33 Spillover ratio Let define Spillover Ratio Index as SRI t = N j=1,i=1,1 j SR ij,t N(N 1) Spillover Ratio Sector Pre Post Energy Grains Softs Livestock Precious Metals Oil with Agricultural Tabelle: Pre and Post indicate before of after end of 2005, respectively. column denotes difference in spillover. Bold numbers indicate that difference is significant at 1% confidence level 33 / 35

34 Conclusions Building on ory of co-movement by Barberis et. al. (2005), we test alternative ory of price co-movment. We test differences in price (co)movement and provide new evidence in support of commodity financialization view. We extend analysis by computing so-called realized betas and obtaining corroborating results. 34 / 35

35 Andersen, T. and Bollerslev, T. (1998). Answering skeptics: Yes, standard volatility models do provide accurate forecasts. International Economic Review, 39: Barberis, N. and Shleifer, A. (2003). Style investing. Journal of Finacial Economics, 68: Barberis, N., Shleifer, A., and Wurgler, J. (2005). Comovement. Journal of Financial Economics, 75: Gorton, G., and Rouwenhorst, K.G., (2006). Facts and Fantasies about Commodity Futures. Financial Analysts Journal, 62 (2): Tang, K. and Xiong, W. (2012). Index investment and financialization of commodities. Financial Analysts Journal, 68 (6): / 35