Excessive Speculation? Christopher L. Gilbert Academic Director, Doctoral Programme in Economics and Management, CIFREM University of Trento, Italy ICEA, 9 November 2009

Plan of this talk 1.Introduction and background 2.Bubbles in standard and behavioural economics 3.Extrapolative behaviour in commodity futures markets 4.Testing for extrapolative behaviour 5.Institutional index-based investment 6.Some econometrics 7.Brief conclusions

1. Introduction and background

The term “excessive speculation” is a quote from two recent reports by the U.S. Senate Permanent Subcommittee on Investigations

Bubbles In terms of the academic discussion, this relates to the literature on asset market bubbles. Neoclassical economic theory permits only a very specific type of bubble – “rational bubbles” – while behaviouralists see bubbles emerging directly out of the psychological aspects of decision-making under uncertainty. In a 2009 working paper, Peter Phillips and Jun Yu, using an econometric methodology based on rational bubbles, argue “the empirical evidence supports a selective migration of the bubble activity through financial markets as the subprime crisis evolved and liquid funds searched for safe havens”. In his Frank Hahn Lecture at the 2009 RES Conference, David Laibson, from a behaviouralist standpoint, stated “Bubble economics may provide a cohesive explanation of the economic events of the past decade”.

Crude oil Both Phillips & Yu and Laibson cite crude oil as having been subject to a bubble. The time plot of WTI prices is consistent with this view, A number of informed comments around the oil price peak suggested that the price was a bubble. NYMEX, WTI, front contract, rolled first day of delivery month, daily, January 1999 – December 2008.

U.S. Senator Joe Lieberman, June 2008

George Soros June 2008

“There is a growing feeling that the latest sharp upsurge in the price of oil may be a speculative bubble rather than an outcome of market fundamentals”. Meghnad Desai, Financial Times, 6 June 2008 Lord Desai, June 2008

You have asked the question “Are Institutional Investors contributing to food and energy price inflation?” And my unequivocal answer is “YES.” In this testimony I will explain that Institutional Investors are one of, if not the primary, factors affecting commodities prices today. Clearly, there are many factors that contribute to price determination in the commodities markets; I am here to expose a fast-growing yet virtually unnoticed factor, and one that presents a problem that can be expediently corrected through legislative policy action. Michael W. Masters, Masters Capital Management, LLC Hedge fund manager Michael Masters, June 2008

Outline The possibility of bubbles in commodity futures is interesting because futures are traded by professional and institutional and not retail investors. Arguably, these investors should be better informed and less subject to psychological biases than retail investors. I look first at what standard economics and finance theory has to say about bubbles, and then at behavioural theories. I use the crude oil market to see the extent to which these theories help us understand recent market developments.

2. Bubbles in standard and behavioural economics

Tirole’s impossibility theorem and rational bubbles Jean Tirole proved that bubbles are impossible in world of risk-averse, infinitely lived agents with common information and common priors. His argument works through backward induction from the fact that the bubble must eventually burst (otherwise aggregate wealth will become zero or infinite). The concept of “rational bubbles” was introduced by Olivier Blanchard (1979). Investors are aware that the bubble may burst but are compensated by a faster rate of price appreciation conditional on its not bursting. The greater the divorce of the price from its “fundamental”, the faster its rate of increase.

Heterogeneous agents Much modern finance theory uses a framework in which there are three classes of transactors 1.Informed traders (perhaps few in number, perhaps hedge funds) 2.Uninformed traders (lacking information but trading rationally) 3.“Noise traders” (transacting for reasons unconnected with prices, best interpreted as “non-rational” traders).  In these models, the uninformed traders attempt to infer the price implications of the informed traders’ information from price movements.

De Long et al and “semi-rational bubbles” Bradford De Long and co-authors (1990) developed a model of “semi-rational” bubbles. Suppose noise traders have positive net demand and drive the price of an asset above its fundamental value. The rational but uninformed speculators do not know whether the prise rise was due to noise traders or informed traders – they attach a probability to each and so buy, putting further upward pressure on the price. The informed traders know that the price is above the funda- mental but are discouraged from selling by “noise trader risk”. The result can be extrapolative expectations – a change rise in price leading to expectations of further price rises.

Why doesn’t Warren Buffett sell? Warren Buffett is the archetypical informed trader. His funds stood aside from the NASDAQ dot com boom but he did not short the market. Why not? The De Long et al model implies that if there are relatively few informed traders and if they have short time horizons (e,g, because of reporting requirements) they will prefer to rise the bubble rather than trade against it even though they know that prices are unrealistic. This tallies with the market wisdom that the easiest way to go bankrupt is to be right but right too soon. Noise trader risk implies that it’s better to stand to one side. The bubble continues unabated.

Behavioural accounts of bubbles These emphasize five factors (e.g. Laibson, 2009) 1.Extrapolation 2.Return chasing 3.Herding (rational and irrational) 4.Overconfidence 5.Over-optimism However, those discussions are generally at the level of the individual retail investor. It is unclear to what extent they translate to institutional investors. Investment in commodity futures is dominated by institutions.

Institutional investors Investors delegate asset allocation decisions to advisors either because they are legally required to do so (some pension schemes), because they regard the advisors as more qualified and/or better informed, or because they lack the time and resources to make their own allocations. To that extent, investor psychology is relevant only in so far as it relates to choice of advisors. What matters is how investment advisors make investment decisions. Investment advisors may be more sophisticated than retail investors – see Jonathan Alvey et al (2007) who compare the experimental responses of students and CBOT traders. However investment advisors are required to compete with each other - an institution which produces poorer returns than competitors will lose market share. Incentives are important.

Do incentives result in herding? Institutional incentives depend on relative, not absolute, returns despite the fact that investors are interested in absolute returns. David Scharfstein and Jeremy Stein (1990) argued that this can result in institutional herding – investment institutions benchmark themselves against common indices and each aims not to underperform relative to those indices. The consequence is that asset allocations differ only marginally from those in the benchmark. Robert Weiner (2006) has found evidence of herding in the NYMEX heating oil market Idiosyncratic allocation is possible for new players, who can gamble in order to attract funds, and for a few funds associated with funds with unassailable reputation (Berkshire Hathaway, Tiger) but are dangerous for most institutions.

3. Extrapolative behaviour in commodity futures markets

The CFTC, CEA, CTAs and CPOs CTAs are Commodity Trade Advisors. CPOs are Commodity Pool Operators. CTAs and CPOs are regulated in the USA by the Commodity Futures Trading Commission (the CFTC) under the CEA (the Commodity Exchanges Act) which governs all futures trading activity. CTAs place retail investor funds in CPO funds CTAs were registered in 2002 with total assets under management of $162bn (Liang, 2004). There were also 597 funds of funds with $343bn. Rich institutions, including many large financial institutions, will prefer to invest in hedge funds. Liang estimates there were 1597 hedge funds in 2002 managing $1580.

Technical Analysis CTAs are require to declare their investment strategies. The vast majority (probably over 90%) declare that they follow non- discretionary “technical” strategies. Much smaller proportions follow contrarian, fundamental or mixed strategies. Trend spotting methodologies differ and CTAs effectively compete on their trend-spotting methodologies. They correspond closely to the uniformed speculators in the De Long et al model. A common strategy involves short and long moving averages. If the short average crosses the long average from below, this is a buy signal; when it crosses from above, the signal is sell. It is difficult to construct trading rules which generate positive risk-adjusted post-sample excess returns. Nevertheless, there is a concern that CTAs collectively generate extrapolative behaviour which may result in bubbles.

Hedge funds Hedge funds have three common features they are only open to the very rich, they have high leverage, and they are currently exempt from reporting requirements. This is all they have in common. Hedge fund strategies and asset allocations differ enormously: Some may use technical analysis but most also do fundamental analysis. Hedge funds correspond more closely to informed speculators in the De Long et al framework. Importantly, all except the top hedge (closed) funds compete to attract and retain funds and have short term reporting requirements.

“Song von der Ware”, (“The Trader’s Song” or “Supply and Demand”) updated Weiss ich was ein Reis ist? Weiss, ich, wer das weiss? Ich weiss nicht was ein Reis ist. Ich kenne nur seinen Preis. By the way, what is rice? Don’t ask me what rice is. Don’t ask advice. I’ve no idea what rice is. All I know is price. Copper, nickel, lead and zinc! Just don’t ask me what I think. Delta, gamma, vega, rho: That’s the sort of think I know. Contangos, backs, whatever tracks. In the end, the trend’s your friend. Market fundamentals stink. Bertolt Brecht, Measure Taken (Die Massnahme) (1930)

4. Testing for extrapolative behaviour

Weakly explosive processes Extrapolative expectation formation can result in explosive prices processes. The standard intuition is that explosions are easy to detect and cannot persist. Peter Phillips has focussed attention on weakly explosive processes in which the departure from a random walk is o(T -1 ) where T is the sample size, i.e. Phillips shows that one can test for extrapolative behaviour using the ADF test but, in contrast to standard tests, referring to the right tail of the distribution.

Month-by-month ADF tests on the oil price Ten years of month-by-month regressions (120 months) yield only three positive estimated  coefficients, none of which is statistically significant. There is no evidence of extrapolative behaviour. Estimated  coefficient from daily ADF(1), month-by-month. Critical values bootstrapped on null using three months’ data.

The Phillips, Wu and Yu recursive procedure Peter Phillips, Yangru Wu and Jun Yu (2009) have devised a methodology for dating bubbles. They use this to date the excess exuberance in the NASDAQ bubble. The procedure works through recursive regression over sample [1:  ] for  = T 1, …,T. They date a bubble as starting for the first value of  for which  > c.v. and ends when  < c.v. ADF(1) recursive t statistic for WTI crude oil and bootstrap critical values. No bubble identified.

5. Institutional index-based investment

Index investment The past two decades have seen the emergence of a new set of transactors – commodity investors. It is these investors that George Soros and Michael Masters accuse to have driven up oil prices. Index investors set out to replicate an index – usually the S&P GSCI or the Dow Jones UBS index – or a sub-index of one of these. These are generally transacted as fixed-for-floating swaps in which the investor swaps the invested sum for the value of the index. The investor is long the index so the index provider (typically an investment bank) is short. The index provider will invest in commodity futures to offset his risk exposure.

Why add commodities to an investment portfolio? Total commodity returns (e.g. on the GSCI) compare favorably with equity returns, although they are slightly more risky. The Sharpe ratio, which measures the risk-return payoff, is comparable with that on equities and better tan the bonds ratio. But … this is the past. Recent returns have failed to match these prospects. EquitiesBondsCommodity Futures Average5.6%2.2%5.2% s.d.14.9%8.5%12.1% Sharpe ratio Annualized monthly excess returns, July 1957 – December Source: G. Gorton and K.G. Rouwenhorst (2004).

Commodity investment and commodity speculation There is a clear conceptual distinction between commodity speculation and commodity investment. Commodity investors are motivated by the potential returns on the entire “commodity class”, not on specific. Until recently, they have been almost entirely long. They are typically in all important markets, and the size of their positions relates to the size and importance of the markets. They roll their positions forward as contracts approach maturity.  Index funds can be large in relation to total market size  Masters states that while traditional speculators are liquidity providers, index investors absorb liquidity.

Index composition The S&P GSCI (left) had a 76% energy weight (55% crude oil) in September The Dow Jones-UBS index had a lower 33% energy weight (13% crude oil)

Index Fund Values and Share of Open Interest, June 2008 $bnShare$bnShare Crude oil %Cocoa % Gasoline %Coffee % Heating oil %Cotton % Natural gas %Sugar % Copper %Feeder cattle % Gold %Live cattle % Silver %Lean hogs % Corn %Other U.S. markets1.4 Soybeans %Total (U.S. markets) % Soybean oil %Non-U.S. markets38.4 Wheat %Overall total199.9 Source: CFTC

6. Some econometrics

Evolution of overall index positions Sharp rises in 2006q1 and 2007q4-2008q2 Very sharp fall in 2008q3 and q4 Flat in 2009q1 with slight rise in late March Source: Authors’ calculations from CFTC Supplementary Commitment of Traders Reports

Index investors and the oil price The chart shows the cross-plot of weekly changes in index positions and percentage changes in the front WTI oil future over the same period. r = January 2006 – December 2008

Granger (non-)causality tests Contemporaneous correlations raise problems of interpretation, so it is preferable to establish the link via Granger causality tests. I perform these within an ADL(3) framework. Hypothesis 1: Changes in WTI Granger-non-cause changes in net index positions. Test statistic F 3,159 = 1.01 (tail probability 38.9%) so we fail to reject. Hypothesis 2: Changes in net index positions Granger- non-cause changes in WTI. Test statistic F 3,159 = 3.35 (tail probability 2.1%). The hypothesis is rejected.  Conclusion: Causation runs from net index positions to the oil price and not vice versa.

IV EMH Estimates The Efficient Markets Hypothesis (EMH) requires that the influence of index investment should be entirely contemporaneous. Granger causality tests pick this up through the autoregressive predictability of changes in these positions. This motivates IV estimation of a simple regression model. I augment this simple model by inclusion of a measure of equity returns (half S&P 500, half Hang Seng)  lnEQ (also treated as endogenous): These results are not strongly conclusive. The corresponding results for non-ferrous metals show greater significance. Inclusion of changes in the dollar exchange rate as an additional control adds little.

What is the mechanism? There are (at least) two possible mechanisms 1.Liquidity effects – large buy orders push prices up, but the transactions do not convey information so price drop back over time. 2.Expectation effects – because, net index purchases are autoregressive, transactions do convey information and hence lead to additional purchases by trend- spotting uninformed speculators.

Permanent or transient? I relate changes in the oil price to current and lagged innovations (unpredictable changes) in net index positions and equities returns: (3.25) (3.55) (3.24) (2.91) Wald F 5, [12.9%] 3.00 [1.3%] Estimation is by OLS and IV using weekly data from March 2009 (165 observations, t statistics in parentheses. If liquidity effects explain the impact of index investment, lagged innovations should have a negative impact. They do not.

Implications These tests eliminate the illiquidity hypothesis and suggests that these trades convey information. Net index positions are strongly autoregressive, and this allows the possibility of changes in positions generating expectations of further changes:

Actual and counterfactual oil prices Index-based investment is seen as having raised oil prices by around 5% through but by around 15% in the first half of 2008

Non-ferrous metals Results for LME non- ferrous metals. Extrapolative bubbles are evident for copper (Feb – April 2004 and April – June 2006) but not aluminium or nickel. Index investment is seen as having statistically significant effects on both aluminium, copper and nickel. Weekly percentage changes in index positions the LME 3 month copper prices. January 2006 – December 2008

LME metals results Wald F 5,152 Wald F 5,152 Al0.785 (4.66) (3.10) 1.12 [35.0%] (2.02) (2.31) 1.89 [10.0%] Cu0.894 (3.55) (4.37) 2.44 [7.6%] (2.01) (1.53) 1.37 [24.0%] Ni0.847 (2.42) (1.74) 0.34 [89.1%] (2.32) (2.87) 2.53 [3.1%] The estimated impact of index investment on aluminium and copper prices is very similar in scale and timing to that of WTI. Nickel impacts are smaller.

7. Brief conclusions

Conclusions Futures markets factors appear to have amplified fundamentally-based price movements over the recent boom. Speculation does not appear directly to be the main cause. There does not appear to be any strong evidence for extrapolative expectations or for bubble behaviour. Index-based investment appears more likely to be the major culprit. This tallies with the views of George Soros and Michael Masters. The maximum impact (spring 2008) was around 15%.