Market microstructure theory does not distinguish between real-money investors and hedge funds. But the two are differentially situated with respect to dealers. Real money investors are large institutional investors like pension funds or endowments whose mandate makes them de facto patient investors — they want to achieve superior returns on their capital over the long run. They rely on dealers to get exposure to risk assets in order to meet their investment goals. Hedge funds have considerably shorter time horizons. They are primarily risk arbitrageurs relying on dealers not only to take the other side of their directional bets but to also provide them with leverage. Indeed, they are the primary consumers of dealer balance sheet capacity.
Fortunately, the CFTC reports derivatives positions of the three species of intermediaries separately. This is an extremely useful dataset that allows us to get a handle on two questions of macrofinancial interest. First, how much predictive information is contained in the net positions of these three intermediaries? Second, how do the three species of market-based intermediaries respond to each other? What are their response functions? Ie, How do they respond to position-taking by the others? The answer to the first may provide a useful way to nail down the dynamic price of systematic risk — the essential component of any serious dynamic systematic trading strategy. The answer to the second can help us better understand the structures that condition the wholesale market for risk. They can also help us anticipate their footfall.
In what follows, we’ll look at a particular corner of this world: intermediary position taking in half-a-dozen of the world’s main hard currencies. FX options and futures account for nearly half of all derivatives by value (the other half is largely interest rate derivatives). Derivatives on hard currencies account for the lion’s share of FX derivatives by value — hard currencies being those that strengthen in bad times. The exchange rates of the hard currencies against the dollar are largely determined by the relative interest rates prevailing along the respective yield curves.
When a gap opens up between the yields on safe assets denominated in different hard currencies, it automatically creates carry trade opportunities. Deviations from uncovered interest rate parity attract arbitrage capital which eventually closes the gap. Position-taking by market-based intermediaries thus emerges as a mediating variable between fundamental macro/monetary cycle-driven movements in interest rates on safe assets in the global system on the one hand and movements in the exchange rates of hard currencies on the other. In what follows, we can imagine that intermediaries observe these gaps and take positions accordingly.
We document two patterns of interest. First, we document that dealer (net) positions in FX derivatives, and to a lesser extent, hedge fund positions — but not the positions of real money investors — contain predictive information about next week’s return on the hard currencies. One possible explanation of this pattern is that dealers and hedge funds are more informed than real money investors. Or, if real money investors are just as informed, dealers and hedge funds have a relatively faster response rate to market developments. In either case, dealers and hedge funds are the bearers of information that gets impounded into hard currency exchange rates.
Second, we document that dealer and hedge fund FX derivative positions are strongly anticorrelated, while the positions of real money investors are much more independent. The obvious interpretation of this pattern is that, on average, dealers and hedge funds take opposite sides of the trade. This implies that dealers absorb hedge fund order flow on their own books without bothering to run a matched book. It also implies that dealers must constantly try to keep up with hedge funds in price discovery, for otherwise, they would systematically lose money on making these markets.
In order to demonstrate these patterns, we present two graphs and one table for each of the six hard currencies under consideration. The first graph is a heatmap showing the predictive t-statistics in a vector autoregression of the change in net positions of the three species of intermediaries along with the log returns on the hard currency against the dollar. The second graph is a heatmap showing the residual autocorrelation between the four series. And the table documents the F-statistics and p-values for Granger causality tests. The data is at the weekly frequency from June 2006 to Feb 2021. We obtain the positions from the CFTC and the exchange rates from Fred.
We begin with the Australian dollar. The predictive t-statistic for dealer net positioning (top right square in the heatmap) is t = -3.13, which is highly significant. That for real money investors is t = -1.88, which is not. Note that the signs of the t-statistics are only informative in a relative sense.
The residual autocorrelation heatmap shows that hedge fund positions and dealer positions are very strongly anticorrelated. As we shall see, this is a general pattern across hard currencies.
Finally, we formally test the hypothesis that dealer net positioning predicts log returns on the Australian dollar. The test statistic is large and highly significant: F = 9.77, P = 0.002. Note also that position-taking by real money investors in the Australian dollar predicts next week’s position-taking by dealers.
Next, we look at the Canadian dollar. This is an exception to the result you were promised. The reason is probably that CAD returns are governed by returns on crude. But for the sake of completeness, we document the evidence anyway. The position-taking of hedge funds and especially real money investors predicts next week’s positions of the dealers. But none contain predictive information about returns on the Canadian dollar.
The residual autocorrelations do show the promised pattern, however. Hedge fund positions and dealer positions in CAD derivatives are strongly anticorrelated.
Finally, the Granger tests show that returns on the Canadian dollar are not predicted by the positioning of any species of intermediaries. However, we have evidence that dealers respond to hedge fund and real money positioning. And all three respond to innovations in the Canadian dollar.
Next up, the euro. Interestingly, hedge fund positioning predicts returns on the euro (t = 2.26) but dealer positioning does not (t = -1.58). This is the only case where hedge fund positions are more informative about future returns on the hard currency. It cannot be ruled out that the pattern is driven by an exceptional event.
The residual correlations exhibit the now-familiar pattern. Hedge fund and dealer positioning on the euro is strongly anticorrelated.
The Granger causality tests confirm the exceptional pattern for the euro. Hedge fund positioning predicts the return on the euro. Meanwhile, both dealers and hedge funds respond strongly to real money wagers on the dollar-euro exchange rate. The interpretation is that they try to front-run and make money off the slower moving real money investors.
The pound sterling brings us back to the standard pattern. Only dealer positioning predicts returns on the quid against the greenback.
Hedge fund and dealer positions are strongly anticorrelated, as before.
And Granger causality tests reveal that only dealer positions contain predictive information on the exchange rate between the dollar and the pound sterling. Meanwhile, both dealers and hedge funds respond strongly to innovations in the quid.
The Swiss franc is up next. Perhaps due to its special status as a small regional safe asset provider, the Swiss currency is especially exposed to intermediary wagers. Both dealers (t = 9.54) and hedge funds (t = 6.66) strongly predict returns on the Swiss franc.
Residual covariation in intermediary positions is very strong between dealers and hedge funds.
Finally, Granger tests reveal that both dealer net positions (F = 91.09) and hedge fund positions (F = 44.39) are highly significant predictors of returns on the Swiss franc.
Lastly, the yen. Again we find that dealer net positions strongly predict log dollar returns on the yen (t = 5.70). This is not the case with the wagers of other intermediaries.
Dealer and hedge fund positions are once again strongly anticorrelated.
Finally, the Granger tests reveal a very large test statistic for dealer positioning as a predictor of log returns on the yen (F = 32.47).
The two patterns we have documented above suggest that it is really hard to beat dealers. That is, it is hard to be more informed than them so that the very definition of informed investors in market microstructure theory is hard to meet in core markets. The patterns also suggest that dealers absorb the order flow from hedge funds on their own books without bothering to run a matched book, at least in FX markets. We have also documented the predictive information contained in intermediary positioning. We have shown that the net derivative positions of dealers contain predictive information about future returns on the hard currencies, as do hedge fund positions to a lesser extent. But real money positions do not. This is clearly due to the empirical fact that dealers and hedge funds are the main players in the carry trades but real money investors are not.
Note that these patterns may or may not generalize to other derivatives. Much work remains to be done before we understand how macro information is impounded into asset prices. A promising line of inquiry would be to try to understand how intermediary positions in interest rate derivatives and exchange rate derivatives cointegrate. For after all they must. They must cointegrate because both are driven by the logic of carry.
——————
P.S. Come to think of it, this post replicates the results of Etula et al. (2010).