Streetwise Professor

Scott Irwin at UIUC brought to my attention this paper by Kenneth Singleton*: “Investor Flows and the 2008 Boom/Bust in Oil Prices.”  As one would expect from Singleton, it is carefully done and thoughtful.  It is appropriately cautious about what can be achieved, given the complexity of the problem and the available data.  He characterizes the empirical part of the paper as “guidance” to efforts to create “much richer structural models.”  Knowing intimately the challenges of this structural modeling, I can say that even given extensive guidance, the challenges to actually incorporating “price drift owing to learning and speculation based on differences of opinion” into models of intertemporal resource allocation (e.g., dynamic models of storage economies).

The headline result in Singleton’s paper is that he finds a statistically significant and positive association between excess returns on oil futures contracts of different maturities and (a) the lagged thirteen week change in imputed positions of index investors, and (b) the lagged thirteen week change in managed-money spread positions.  He also finds that the one-week change in repo positions on Treasury bonds by primary dealers predicts price moves  Unlike Hong and Yogo (a paper I discuss below), he finds that open interest has no predictive power over oil futures excess returns, once these (and some other) variables are controlled for.

Two issues arise here.  The first is the interpretation of the results.  The second is their policy implications.

Singleton appeals to two explanations: limits to arbitrage (LTA) and differences of opinion (DOE).

LTA explanations make sense.  Speculators are subject to various constraints: informational frictions, information costs, fixed costs, etc., limit the amount of capital speculators have.  Moreover, this capital is subject to economic shocks, and various institutional features (e.g., the use of Value at Risk or other similar mechanisms for constraining speculation by agents) can lead to self-reinforcement of these shocks.  Furthermore, hedger demands can vary randomly due to shocks to their balance sheets and to broader financial market conditions.  This can lead to changes in risk premia in equilibrium. (A paper by Acarya, Locstoer, and Rmadorai explores these issues.  This is really an extension of Hirschleifer the Younger’s work on segmentation of commodity markets and the broader financial markets due to fixed costs of market participation.) (My work with Martin Jermakyan on electricity derivatives pricing documents (a) large premia for risks that should be diversifiable, and (b) substantial declines in these risk premia as constraints on speculation declined.  We use LTA-type explanations to explain these results.)

I have worked through the numerical analysis of a structural storage model in which the market price of risk varies randomly as a way of incorporating LTA factors into a dynamic storage model.  Quite intuitively, shocks to the market price of risk affect prices.  When the market price of risk falls, the commodity price rises and interestingly, inventories rise too.  Indeed, the inventory effect is more noticeable in simulations than the price impacts.  The intuition is straightforward: when the market price of risk falls, due, for instance, to an easing of constraints on speculators, it is cheaper to hedge inventory risks, so it is cheaper to hold inventories, and inventories go up.  The only way for inventories to go up is for spot prices to rise to reduce consumption and increase output.

I view this work as complementary to Acharya et al.  They model the factors that affect risk aversion of hedgers and speculators, but have only a crude model of the storage economy.  I treat the risk aversion affects exogenously, but have a more rigorous model of the storage economy.  It would be nice to both in one model, but that’s not really tractable.

But my reduced form model is sufficient to demonstrate that LTA-type phenomenon which lead to fluctuations in the market price of risk in a particular commodity market can be associated with price and inventory movements and with changes in speculative trading.

The DOE explanation is more, well, speculative.  They can rationalize departures from rational expectations equilibrium pricing, and hence can explain some anomalies.  But testability is problematic.  Moreover, per Hong and Stein’s 2007 JEP survey article, the main thing that separates DOE from LTA is that the former predicts much higher trading volumes.  That doesn’t map in that clearly with Singleton’s findings.

Moreover, although Singleton does a good job at presenting the state of understanding in rational expectations-type models of storable commodities (including some nice cites to my work), and mentions evidence relating to quantities, I still don’t think he confronts fully the fact that commodities are different than speculative stocks–which are the main motivation for much of the DOE literature.  Commodities are consumed and produced in the here and now, meaning that distortions in prices affect consumption and production decisions.  That should show up in quantity data.  To figure out exactly how, you’d need to integrate a model with storage and differences of opinion among market participants.  That’s a very tall order.  Similarly, you’d like to have the model make predictions about the entire forward curve.  Another tall order.

If you think that due to DOE, speculators are causing prices to be high, it is pretty clear that this will reduce consumption.  For something like oil, the effects on output are complicated.  If producers have no ability to allocate output intertemporally, the effect of the price rise is straightforward: output should go up.  Thus, if prices are forced up by speculators, this should lead to a rise in inventories.  But for oil, producers may shift output over time.  Producers can store underground by deferring production (and perhaps exploration and development), and their decisions on intertemporal allocation depend on the price today relative to what is expected to be the price in the future.  Thus, you’d need to look at a DOE model of a forward curve that incorporates intertemporal resource allocation through mechanisms other than storage.  As noted before, a tall order.

These types of theories would be necessary to understand fully what is driving energy prices.  I don’t think such an understanding is necessary to evaluate the implications of Singleton’s analysis for specific policy proposals–notably, position limits.

I imagine that the usual suspects will seize on these results to say: “Aha!  The smoking gun showing that speculation impacts prices.  Proof that we need position limits.”

Uhm, not so fast.  To the extent that the results reflect limits-to-arbitrage type effects, where shocks to hedger and speculator balance sheets drive variations in the market price of risk, position limits would be counterproductive.  If the association between prices and speculative positions reflects variations in constraints on speculators’ ability to absorb risk, or on hedgers’ demands to shed risk, constraining risk transfer will not improve things.  Indeed, it will make them worse.

The implications of difference-of-opinion-type explanations are hardly supportive of commodity position limits either.  For one thing, if you really believe that DOE results in price distortions, and this phenomenon is ubiquitous, why focus obsessively on commodities?  Why not restrict speculation in the entire investment universe?  But just how would you do that, exactly?  How would you surmount the knowledge problem and get the amount of speculation right?

But even ignoring that issue, position limits are not a discriminating tool that would eliminate the source of DOE-based distortions, to the extent they exist in this commodity market or that.  They are not a magic wand that somehow homogenizes the opinions of market participants.  Presumably even differences of opinion among hedgers, and between hedgers and speculators not constrained by position limits, would result in price impacts.  In other words, this is a ubiquitous feature of all speculative markets, not just commodity markets in an era of index funds and big money managers.

There is some evidence that speaks to this.  A paper by Hong and Yogo finds that open interest in commodity markets has predictive power over futures excess returns.  They also present a model based on gradual information diffusion (a common feature in DOE models) that can explain this result. The empirical result is hard to explain in a rational expectations context.

But note that the data in Hong-Yogo goes back to the mid-1960s.  Well before commodity index funds.  Well before hedge funds became major players in commodities.  Well before the “financialization” of commodity markets.  Meaning that if you believe that DOE and departures from rational expectations, purely fundamental-based pricing explain the result, it cannot be attributed to relatively recent financial innovations.  The Hong-Yogo results say there was no Edenic past in which commodity pricing was rational, only to be destroyed by the snake of financialization that slithered into the garden in the 2000s.

Meaning further that measures targeted at selected types of traders will not restore the non-existent Eden.  Indeed, given that the implications of these behavioral models are quite sensitive to the composition of traders, the information that each type of trader relies on, etc., it is quite possible that restrictions can lead to even greater departures from “rational” pricing.   I would surmise that if you had good data on the trading of certain categories of entities, from these earlier eras you would be able to find that the trading of some type or types of traders would have predictive power, just as Singleton finds that the trading of index funds and spread traders has predictive power.  So if you constrain index funds and spread traders through position limits, it is quite likely that the old-school categories of traders would represent a greater proportion of trading activity, and recoup their predictive power.

In brief, if you believe the DOE-type stories, you believe that these effects are ubiquitous: they are in the water.  You can’t make them go away by targeting one class of trader.  If you have two traders–you’ll have a difference of opinion.

This relates to an earlier SWP theme.  It is really necessary to be careful in distinguishing between the documentation of speculative effects, and the appropriate policy response.  To justify any policy aimed at curbing speculation, it is necessary, but not sufficient to demonstrate that speculators have somehow affected prices.  You also have to show that they have distorted them; with LTA, for instance, changes in speculation may affect prices, but not distort them in any reasonable sense of the word.  Moreover, you have to show how the specific policy will reduce that distortion.  That depends crucially on the source of the speculative distortion.

DOE-based stories are at least on far firmer ground than the typical hydraulic explanations of how speculation distorts prices.  But as I argue above, they provide weak support for position limits.  LTA-based stories provide even weaker support: indeed, plausible versions of LTA suggest that position limits will make problems worse, not better.  (For instance, if you believe that small noise traders distort prices, constraining big smart money will tend to exacerbate the effects of small noise traders.  Although this kind of story could provide a rationale for constraining ETFs that reduce the costs of noise trading–but can you design position limits that constrain noise-trader dominated ETFs but not big smart money traders?)

So my final take on Singleton’s paper is that the result is interesting, not necessarily that surprising, consistent with some non-behavioral explanations of price movements in which financial constraints and fundamentals affect prices–and a very weak basis for policy recommendations, and a particularly weak basis for justifying position limits.

Not that the usual suspects won’t try.

*That would be Kenneth Singleton the Stanford economist, not Kenneth Singleton, former MLB player.