Don’t Be Fooled: This is No Hedge Fund Replication

On Wednesday, the Wall Street Journal ran a blog on replicating hedge fund returns.  The technique proposed by Tim Edwards, an analyst with S&P Dow Jones Indices, could not be simpler.  He constructed a model consisting 50% of the returns on the S&P U.S. Aggregate Bond Index and 50% on those of the S&P Global 1200 Index, rebalanced monthly, charging a 1.5% annual management and a 15% performance fee.  He compared it with the HFRI Fund-Weighted Composite Index, which is constructed from the reported returns of hedge funds worldwide.  The blog provided this graph of the model’s monthly returns against the benchmark:

No analysis was provided, but the graph is all that is needed to show that the series are tightly correlated: from the look of it, probably >0.9.  Note that the replication was consistently and significantly more volatile than the index it allegedly replicates.

Mr. Edwards is quoted saying “The average hedge fund looks like a fixed blend of cheap investments, at high cost.”  He has not demonstrated this.  What he has shown is that an unweighted index aggregating >2,000 managers’ returns will, over short periods, resemble broad market indices.  It does not analyze the funds’ performance over periods meaningful to investors, and over which the funds’ cumulative returns and volatility would differentiate them from an index.  Mr. Edwards’s model replicates the mean fund and chain-links its return to that of the next month’s mean return: it does not replicate any particular fund, so there is no sense in which it replicates an “average” hedge fund.  Volatility lowers compounded returns, so his model underperforms even that notional fund on both absolute and risk-adjusted bases over periods of a year or more.

That an average of many funds’ returns resembles such a model over short periods is not surprising.  The World Federation of Exchanges’ June report counted 42,759 companies worldwide.  The overwhelming majority of them would not attract a fund’s attention, so hedge funds’ portfolios inevitably overlap.  In a given month, returns on the Russell 1000 are similar to the average of all large cap U.S. equity funds, too.  But hedge funds’ returns disperse widely around the mean and are negatively skewed: the return on the median fund (the one outperformed by as many funds as it outperforms) is higher than the mean return of all funds.  Based on the graphic with which we are provided, Mr. Edwards’s technique creates a roughly sixtieth percentile hedge fund clone, in a period of consistently rising markets that favor a long-only comparison.  This is hardly impressive.

There are many issues regarding indices that purport to measure hedge fund returns.  They include survivorship bias, whether the “best” funds report their returns at all, and whether funds that do report are honest (it is in their interest to report returns which flatter their actual performance).  And quite apart from issues with indices, there are plenty of reasons to be skeptical about hedge fund.  The extent to which leverage parades itself as alpha is the most important one.

I am skeptical about “hedge fund beta,” and consequently doubtful about any approach to hedge fund replication based on aggregate fund returns rather than the returns of a specific fund.  The high dispersion of hedge fund returns indicates that the mean of those returns has limited value as a series to replicate: any data can be correlated with sunspots, Elvis sightings, etc., and data-mining will always find a series it resembles.  A specific fund at least provides a meaningful basis for comparison with other financial series, and one that might be worth the trouble of using factor analysis or replicated trading strategies, to replicate.  Strategies based on an algorithmic model of a fund's trading strategy may also hold some promise.  I am nevertheless skeptical.  But I am not skeptical about crude strategies such as that proposed by Mr. Edwards: it is not a replication at all.

Research-o-Matic!

Last week the Wall Street Journal ran an article entitled “Can You Tell the Difference Between a Robot and a Stock Analyst?”  No doubt the question was rhetorical, but at the risk of seeming literal-minded, I would say that, much of the time, the answer is “No.”  Too much alleged “research” is so devoid of analytical value-added that, if it is not written by machines, it might as well be.  Although I suspect robots have a stronger grasp of grammar and less of a predilection for TV taglines than many human analysts, performing a Turing Test on Wall Street research is becoming increasingly difficult: shaky grammar and an unhealthy obsession with TV may soon be the only sure indicators of flesh-and-blood authorship.

Having been a sell-side analyst in the early 1980s, I can vouch for the fact that, in those days, analysts were human.  There were several respects in which we were not only noticeably different from, but unquestionably superior to robots: 

·         Analysts aspired to style.  Few of us could write as gracefully as Barton Biggs or some of the other research luminaries of the day, but I think all of us wished to.  We did not revel in solecisms such as ‘going forward’ or ‘proactive’ and did not try desperately to sound like sportscasters.  Granted, concern for style was not an unmitigated good: I often suspected that my English colleagues devoted more care to how than to what they wrote.  But, as I discovered when I moved to the buy-side and my days were filled with reading research, an effort at style is a kindness to the reader.  Robots are sadly neglectful of this act of charity. 

·         Analysts made satisfactory scapegoats.  Every portfolio manager knows that if something goes wrong with a position, it is the analyst’s fault.  Bullying human analysts is enjoyably therapeutic for a disappointed fund manager: they flinch when shouted at, offer ludicrous excuses and sometimes (though rarely) can be brow-beaten into an apology.  Robots are useless as scapegoats: it is impossible to reduce them to tears and if you kick them you merely put your toes at risk. 

·         Analysts collected anecdotes.  I recall a Chairman advising a CEO to tell the newspapers to stop worrying about their company’s cash mountain, because he had misplaced it.  I recall a Yorkshireman talking about competing with the Genovese family.  Visiting companies and spending time with their managers is an endless source of amusing stories, allowing analysts to appear more interesting than they really are.  Robots are hopeless raconteurs ─ they rarely capture the telling detail and they have a habit of telegraphing punchlines. 

·         Analysts lunched.  The decline of this talent may well deserve most of the blame for the increasingly robotic character of sell-side research.  In the quest for Institutional Investor votes, analysts were expected to entertain: we contributed strongly to, and perhaps underwrote Zagat’s success.  Most analysts were dab hands with a martini and some knew their way around a wine list.  In my experience, robots insist on ordering moscato, for which a suitable food pairing is unlikely ever to be discovered. 

Obviously, Wall Street, the City of London and a few other previously favored locations will be a lot duller for the rise of the Grahamatrons and Doddodynes.

Automated research is probably an inevitable development, in keeping with the Street’s pursuit of ever more tawdry cheapness.  Those jolly specialists and market makers who used to mishandle your orders for you have largely been replaced by machines that do it faster and more cost-effectively, but with a sad lack of humor.  Much upstairs trading and portfolio management is now computerized.  The humans who remain in those functions are becoming increasingly cybernetic ─ with software to make good their deficits of life experience, common sense and market history.  I am unaware of any efforts to automate investment banking, although I have long suspected that some calling officers I have known were wind-up toys.  But at least they still know how to make use of an expense account.  Even retail investors, whose function is to pay for brokers’ yachts, are increasingly relegated to obtaining financial market access through machines.  Robo-advisors presumably suggest that they place their orders electronically, keeping humans as much as possible out of the relationship.  What the robots do with their yachts is a good question, and to my knowledge their designers have yet to find ways to build in hand-holding or golfing functionality, let alone the ability to explain that even bonds can lose money.  No doubt they are working on it.  All that remains is to replace the final customers.

Although I do not think that paleontologists any longer believe it, they once theorized that mammals evolved as scavengers and egg-stealers in the age of the dinosaurs.  Something similar may be the role that is left to humans in a thoroughly automated system of transactional finance.  Things were probably headed in that direction even before the machines intruded themselves so pervasively: it surely is no coincidence that the widespread adoption of passive investment techniques was accompanied by the growth of hedge funds and similarly “undisciplined,” opportunistic forms of investment.  We humans can get our own back on the machines by thriving in the environment they create.

A Defense against Predatory Traders?

Institutions have a long history of accusing market makers and other active traders of predatory trading. There is justice in many, but certainly not all, of these complaints. Much of the innovation in U.S. market structure over the last decade or so has been motivated by institutions’ desire to avoid the problems they perceive themselves to have had with traditional liquidity providers. The results have not met expectations: the fragmentation of liquidity into numerous execution venues has in fact increased the ability of computer algorithms to anticipate the routing of large institutional orders. It is entirely legal for a firm to exploit such information, provided that the firm does not have a fiduciary duty to the source of the order.

So it is a matter of considerable interest that RBC Capital Markets has launched Thor™, an order-routing system that is designed (among other things) to ameliorate institutions’ problems with so-called “latency arbitrage.” One of the problems that has resulted from the fragmentation of the equity marketplace is that latency (the time delay between order entry and the actionability of the order at the point of execution) is different for each trading venue to which an institution might choose to route its orders. Thus if it makes use, say, of five trading venues to execute a large order, the portions of its order routed to each venue will arrive at different times. Although the time differences involved are miniscule, they are sufficient to allow high frequency trading algorithms to observe the first order to arrive and to mop up liquidity at the other venues, forcing the institution to buy at a higher price or sell at a lower one than it anticipated.

RBC’s solution is elegant, and somewhat paradoxical: it addresses the problems created by the continual arms race to reduce latency by increasing latency. Orders are routed sequentially to the venues of the institutional customer’s choice, with those routed to the venue that takes longest to access released before those routed to venues with lower latency. The orders consequently appear in the various venues’ orderbooks as close to simultaneously as possible, allowing little or no time for predatory algorithms to anticipate the arrival of orders along the higher latency routes.

The proof of the pudding for this solution will be a matter of how closely it can approach to exact simultaneity. The greater the exactness of timing, the less opportunity for “latency arbitrage,” and precise simultaneity offers no such opportunity at all. Cynics may argue that the financial and computing resources available to high frequency traders will ultimately allow them to defeat this order routing stratagem, but it is difficult to see how, provided that the degree of simultaneity it achieves is sufficiently high. While other forms of predatory trading will still be possible, “latency arbitrage” may soon be a thing of the past.