Regulators understand the potential threat of crowded trades, but they also recognise the difficulty of tracking them. This column suggests a new approach for regulators to monitor crowdedness of selected trades. Fund managers and financial regulators could use data on crowdedness to assess the risk that a financial market may enter an asset bubble.
Whether crowded trades pose a threat to financial institutions has been on regulators’ minds for several years. In 2004, Timothy Geithner, then President of the Federal Reserve Bank of New York, put it this way: “While there may well be more diversity in the types of strategies hedge funds follow, there is also considerable clustering, which raises the prospect of larger moves in some markets if conditions lead to a general withdrawal from these ‘crowded’ trades.” The underlying logic of Geithner’s remarks is simple enough. Market participants may face additional risks if many players want to exit similar positions at the same time. Lasse Heje Pederson (2009) has modelled this behaviour in financial markets. When shouts of “fire” are heard in a crowded theatre, patrons face not only the risk of fire, but also the collateral risk of being trampled by others trying to escape the common threat. A stampeding crowd just might kill you even if the fire does not.
Congressional testimony over the last year shows that regulators understand the potential threat of crowded trades, but they also recognise the difficulty of tracking or identifying crowded trades. As one commentator concluded, “the sad truth [is] that crowded trades are difficult for the government to identify” (Mallaby 2009). While there are some data that analysts occasionally allude to as indicators of crowding, no technique has been proposed for identifying which trades are crowded or quantifying the extent of crowdedness.
In recent research, we propose a new technique for identifying crowded trades and apply it to a set of stylised trading strategies in the currency market (Pojarliev and Levich 2009). The technique is fairly intuitive and simple to implement. When a hedge fund manager bets some of his assets on a gold trade, for example, then his fund’s returns will be correlated with returns on gold and will exhibit a positive and significant style beta versus a gold index. Contrarian managers would have negative style betas versus a gold index. We classify a “crowded trade” as one that many managers are betting on. In other words, a crowded trade is a situation where a large percentage of hedge fund managers have positive and significant style betas for that trading strategy.
We adopt this approach to examine crowdedness for three classic currency trading strategies:
- Carry: where speculators hold long positions in high interest-rate currencies financed by short positions in low interest-rate currencies,
- Trend: where speculators hold long positions in currencies with positive trend financed by short positions in currencies with negative trend, and
- Value: where speculators hold long positions in highly undervalued currencies financed by short positions in highly overvalued currencies.
Our data cover a three-year period up to 26 March 2008 with weekly data on as many as 60 currency funds. We use rolling windows of 26 weekly observations to estimate style betas. We count only those betas with t-values greater than 2 as evidence of a fund being committed to a particular style and express crowdedness as a net figure (number of positive betas less number of negative betas).
Even though our approach relies on some econometric restrictions, we find that the net number of funds with significant style betas to any strategy fluctuates considerably over time. For example, the percentage of funds with significant style betas for carry varies between -7% and 31%, a level reached in spring 2008 just prior to the collapse of the carry trade. Trend crowdedness ranged between 4% and 34% and value crowdedness varied between about +12% early in the sample to about -27% toward the end.
New insight on herding in hedge funds
Our specific empirical results and general methodology raise a variety of questions. One question is: what factors drive changes in crowdedness? Presumably, speculators are attracted to a trading strategy when expected returns are high. Rational speculators ought to be motivated by expected returns, which for all three of our strategies depend at least in part on expected exchange rate changes. We found weak evidence that this was the case. Instead, it appeared that crowdedness tended to increase after strategies were profitable and decreased for unprofitable strategies. In other words, there is some degree of herding among currency speculators who seem attracted to strategies that have been working and could continue to work in the future.
Another question concerns the dynamics of funds electing to crowd into or vacate a trading strategy. Some funds in our sample are dedicated trend or value players, so they are unlikely to change their tune and become carry traders even when those strategies experience good profits. But other fund managers have discretion to switch and become carry traders when it suits them. There is yet another group of managers who are on the sidelines and can start up a fund to follow carry when it looks like a profitable strategy. Our data show that about half of the growth in funds with significant carry positions came from existing funds who switched into carry, while the remainder came from new funds that joined our data sample.
These results underscore that our approach will be most useful to gauge crowdedness among hedge fund managers rather than long-only mutual fund managers. Hedge fund managers are more likely to use leverage and to have more discretion in their investment choices as they have absolute return mandates. Mutual fund managers are mostly managing against benchmark mandates, trying to outperform an index (like the S&P 500). For example, a long-only mutual fund equity manager is likely to always have exposure to equity beta even when he is bearish on equities. Furthermore, as he does not use leverage, his trades are simply offset by other investors taking an opposite position without impacting the gross market exposure and will not contribute to changes in equity crowdedness. By comparison, a hedge fund manager can leave his original capital position in cash and use the cash as collateral to take positions in various assets such as gold, a foreign currency, or shares of IBM that do not impact the market’s net position in that asset but can impact the gross exposure of outstanding claims dependent on the asset. An increase in gross exposure implies that investors are committing more of their capital to trades that depend on the asset – they are crowding into a trade. Put differently, a hedge fund manager’s baseline position in any asset is zero until he makes an investment. Variation in crowdedness relative to a zero baseline provides information on the market’s changing gross exposure to an underlying investment or trade.
While our approach seems promising, it also harbours some limitations. First, our methodology is necessarily backward looking, in our specific case, estimating crowdedness based on the last 26 weeks of returns. Conceivably a smaller window (say, 20 weeks) or making use of daily data might still find significant results. Daily observations, however, are not readily available for most hedge funds, and daily data may contain accounting and reporting errors that inject noise into the results. Second, our approach does not capture the size of the positions held by fund managers. A manager with only $100 all invested in the carry trade would have a highly significant style beta, but a very small impact on currency trading activity. Third, our approach considers only at a sample of fund managers rather than the market as a whole.
Each of these limitations has a counterargument. First, even though our approach relies on historic data, it captures those funds that have been committed to a strategy for the last six months, up to the present. The crowdedness measure thus in part reflects current data and persistence over the recent past. Second, data on assets under management are available for many managers and so it could be possible to develop a measure of trading intensity to complement our measure of crowdedness. And third, as long as our sample of funds is representative of the broader institutional fund management universe, a sample may give an accurate reflection of the popularity of various trading styles in the market as a whole.
What could fund managers and financial regulators do with information on crowdedness?
Data on crowdedness could potentially be useful to fund managers. Economic intuition suggests that as more speculators move into a trade, prices adjust and expected returns fall for speculators who come afterwards. When investors buy an undervalued currency, at the margin the spot rate appreciates and the currency becomes less undervalued. When investors swap out of low yielding securities for higher yielding ones, there is pressure to shrink the yield differential, lowering expected returns from the carry trade. (See also Jylhä and Suominen 2009). Trend following could be an exception whereby speculators piling into a trend help exacerbate and extend the trend, thereby attracting other trend followers. But under the notion that “trees do not grow to the sky,” trending currencies may become overvalued, making it risky for late arrivers to a trend-following strategy. Knowing that a certain trade had become crowded would likely encourage managers to look elsewhere rather than piling on and adding to the crowdedness and risk of trading style. Indeed, this conclusion is reached by Robert Litterman, managing director at Goldman Sachs Asset Management as reported by Reuters, “Quant hedgies must fish in fresh waters – Goldman,” in December 2009.
Data on crowdedness could potentially be useful to financial regulators. While the private investment community could be self-regulating in the sense that fund managers ought to be wary of entering crowded trades, market regulators could provide an additional brake on private market excesses. Market regulators could independently monitor crowdedness of selected trades with the objective of either making these measures public, or privately counselling individual banks and funds on the risks they are exposed to. Beyond counselling, regulators of course have the option to raise capital requirements for operations deemed to carry greater risks. Measuring crowdedness may offer a useful tool in determining whether a financial market is at risk of entering an asset bubble.
•Jylhä, Petri and M. Suominen (2009), “Speculative Capital and Currency Carry Trade Returns,” working paper, Helsinki School of Economics.
•Pedersen, Lasse Heje (2009), “When Everyone Runs for the Exit,” NBER working paper 15297, August.
•Pojarliev, Momtchil and Richard M. Levich (2009). “Detecting Crowded Trades in Currency Funds“, December.
•Mallaby, Sebastian (2009) “A Risky ‘Systemic’ Watchdog,” Washington Post, March 2.