As technology disrupts industry after industry, it is logical to expect the hedge fund industry will be similarly disrupted. We pose the question: How will that disruption happen and where will it come from?
Some fundamental managers believe their brains are irreplaceable and their instincts are their alpha source — and more reliable than a black box. Systematic investors would argue there is no bigger black box than a human brain. Unlike the unpredictability and behavioral biases of human behavior, systematic strategies are strictly rule-based and might even be programmed to exploit these weaknesses.
The debate, however, draws attention away from where the real disruption will come. In the book "The Innovator's Dilemma," Clayton Christensen showed why disruption comes from outside industries, not within them. This, historically, has been true with hedge funds as well. The fundamental hedge funds that disrupted traditional asset management didn't come from within traditional asset managers. The quantitative hedge funds that disrupted fundamental hedge funds didn't come from within fundamental hedge funds. Similarly, the next wave of alternative investment strategies will come from outside the industry; not within it.
What will the next wave of managers look like and where will it come from? We believe we know, because over the past three years we have found and met with more than 200 of them. We call them managers specialized in "autonomous learning investment strategies" or ALIS. They are usually run by millennials, some of whom were hackers or gamers, whose brains seem to be wired differently from the earlier generations. They believe they can use advances in artificial intelligence, record low cloud computing storage and processing power and new, unstructured non-financial data sources to run investment strategies at a fraction of the cost of established managers. With the newer AI techniques, ALIS managers have the "learning" taking place within the machine, with the machine gaining experience and teaching itself at an exponentially faster pace than a human, on thousands of securities, potentially with thousands of data points per security. This means that instead of dozens, or in some instances more than a hundred Ph.D.s needed to run a quantitative strategy, one now only needs a few.
We think of these ALIS managers as being a derivative of the counter-culture (which is why the title of this article references the powerful and prescient Pink Floyd song). Just as it would be hard to imagine Elon Musk working at Ford or Jeff Bezos aspiring to run Walmart, most ALIS managers see themselves as being outside of the Wall Street world of MBAs, investment banks and hedge funds. Very few are in the commercial hedge fund databases. To find these managers requires networking through technology and academic circles. They don't aspire to be the established quantitative and fundamental managers, they aim to disrupt them. While the quality and investibility of ALIS managers varies wildly, we're confident the best ones will succeed.
But are they truly different in how they invest? February was an important reference point. Markets shifted gears and volatility spiked. Commodity trading advisers fell 4.8%, according to the Eurekahedge CTA/Managed Futures Hedge Fund index. Yet the top commodities/futures ALIS managers, according to our ranking system, were up materially in February. Digging into why this was shows ALIS managers aren't just smaller and leaner versions of established firms, they are investing in a different way.
Specifically, in comparing top ALIS managers with established CTAs, we saw five differences:
- Not being reliant on one strategy;
- Machine learning using incremental/new data;
- More emphasis on shorts;
- Not using short option strategies (explicity or implicitly); and
- Didn't over-fit.
We expand on each of these differences below.
Whereas many traditional CTAs might be reliant on one strategy, such as trend following or mean reversion, top ALIS managers may be more dynamic, driven by whichever strategy is best suited to the particular market environment. These decisions may, in turn, be driven by the machine learning's ability to learn from incremental/new data. Moreover, because ALIS managers use machine learning — which, in turn, can find more complicated and non-linear patterns than simple linear regressions, a common CTA technique — they may extract more alpha from the same markets.
In addition, the models might place more emphasis on the shorts. For example, just as a hedge fund that we know and think highly of pays its analysts more for short alpha than long alpha (because it is deemed more scarce and valuable), ALIS managers might program their systems to place more emphasis on short alpha for the same reasons.
Another way to explain how ALIS managers might place more emphasis on shorts is to use an insurance industry analogy. In insurance there are high frequency/low severity businesses, such as auto insurance, and low frequency/high severity businesses, such as catastrophe insurance. Although catastrophes are low frequency, because they are high severity, an ALIS manager, specifically the Ph.D., would train the system to effectively place a greater importance on those events, which in layman's terms would be akin to bear markets or left tail movements.
The ALIS models are essentially not making the unrealistic assumption that market returns follow normal distributions; rather, they are accounting for the fact that market returns do not follow normal distributions, and in fact have fatter tails. That x sigma moves occur more frequently than the traditional risk models would predict.
ALIS managers that are within the top quintile achieved strong absolute and risk-adjusted returns, i.e., Sharpe ratio, within Mov37's database of more than 200 managers in February and May 2017, when CTAs were down and this supports our viewpoint.
Lastly, the best ALIS managers do not over-fit, or extrapolate patterns that may have existed in the past, but will not exist in the future. Whereas traditional CTAs might under-fit due to the use of more rudimentary statistical techniques like linear regressions, best-in-class ALIS managers may use machine learning techniques that enable them to extract greater alpha from the same (or more) data through better fit models.
In our view, the best ALIS managers are not short-option strategies, explicitly or implicitly. In February, for example, when volatility exploded, many top CTAs were implicitly short volatility, akin to a short-options strategy, even if not explicitly short volatility or options.
We believe these data points are likely indicative of our thesis that the world's leading ALIS managers may deliver idiosyncratic alpha, uncorrelated to long-only, hedge fund, factor, smart beta, risk premium or other alternative betas. For readers unfamiliar with ALIS, Welcome to the machine.