What a long strange trip it’s been...on ALIS
By Michael Oliver Weinberg, Chief Investment Officer, MOV37 and Protege Partners
Published: 21 January 2018
We have titled this paper with an ode to a compilation album by a band that was founded in Palo Alto and developed a counter-culture. Though the title espouses a new state of mind, it is investment, not consumption driven, as this is 2017 and not 1965.
There is a book written by John Markoff entitled What the Dormouse Said, and those with a penchant for '60s music may recall this is a line from a Jefferson Airplane song based on Lewis Carroll’s classic book, Alice’s Adventures in Wonderland. For those not familiar with Markoff’s book, a primary point is that today's personal computer is largely a derivative of Stanford and its/the counter-culture. It was this counter-culture that revolutionized society in ways that no one then could dream of ex-ante, and are only obvious now, ex-post. In line with these themes, in our website, www.mov37.com, we include allusions to Carroll’s book through the titles of our “ALIS Through the Looking Glass” and “ALIS Down the Rabbit Hole” sections.
“ALIS” is an acronym we have created for “Autonomous Learning Investment Strategies”, which we believe are an emerging “third wave” of investment managers, the first and second being fundamental discretionary and quantitative investing, respectively. ALIS are smaller managers taking of advantage of recent advances in artificial intelligence and machine learning, combined with an explosion in data availability and inexpensive cloud computing, to generate alpha at a fraction of the cost of traditional managers. (For a fuller explanation of ALIS managers, we recommend reading the canonical paper on them by Jeffrey Tarrant, entitled “The Intelligent Investor in an Era of Autonomous Learning”, available on our website.)
Just as the Grateful Dead developed a counter-culture, ALIS managers are also. Whereas the first wave of investing was comprised of MBAs, the third wave is comprised of PhDs. Though some ALIS managers are based in New York and London, as the MBAs and Wall Street or The City are, many are based in Palo Alto and San Francisco, homes of the original counter-culture and Silicon Valley. The counter-culture of the '60s often sprouted from the world’s leading universities, which is where ALIS managers also are germinating.
Reverting to the title of this essay, we have spent the last couple of years traversing the globe, far beyond New York and London, including Israel, Asia, Canada, Silicon Valley, university towns, The South (in the US) and suburbs in our quest to locate the world's best ALIS managers, and it has been a long, strange trip. We have found 200 managers that portend to be ALIS managers. For context, this compares to one of the world's preeminent hedge fund databases that has only identified a fraction of that.
In George Orwell's Animal Farm he states that all animals are equal but some are more equal than others. Similarly, our take on ALIS managers is that they all are equal but some are more equal than others. In the land of ALIS there is a wide spread between the best and the worst, and most managers are far from average.
We shall now share with the reader some of our adventures researching ALIS managers around the world, organized into sections on the genres of ALIS managers we would and would not invest in. To do this, instead of obscuring and altering certain details so that it’s not overtly apparent which managers we are referring to, we will simply speak very generally, just as Barton Biggs did in his book about first wave investing, Hedgehogging.
Which ALIS managers are better? Which ones to avoid?
Genre 1 - Acronym Soup
The best optimize machine learning techniques and focus on depth rather than superficial coverage.
The best ALIS managers understand exactly why they use a particular machine learning technique. They don’t just use off-the-shelf code – they custom-tailor it to exactly suit their needs.
The worst fall into Acronym Soup.
Though we have named the third wave with our own acronym, ALIS, we sensibly took a pause. There is a genre of ALIS managers that over-compensates for a lack of substance by asserting they use every acronym of machine learning, in what we refer to as Acronym Soup.
A typical due diligence meeting with these transgressors might go something like this. MOV37 team: "What type of machine learning techniques do you employ?" Alphabet Soup ALIS managers: "We use SVM, PCA, NN, NLP and KNN" (referring to Support Vector Machines, Principal Component Analysis, Neural Networks, Natural Language Programming or Neural Linguistic Programming and K-Nearest Neighbor, which are detailed on our website).
There are some top ALIS managers who do employ many of these techniques together, however, in our experience to date, the top decile of ALIS managers generally are most proficient in one of them, which is dominant in their strategy. In addition, these techniques are typically not being used through off-the-shelf code, but are custom-tailored by the managers. There typically is an inverse correlation between the number of ALIS techniques employed and the quality of the machine learning, and fund.
For fun, to violate our aforementioned acronym stinginess, we will extemporaneously create a new one, MLP. Henceforth, an MLP is a Machine Learning Panacea. Unlike a Black Swan, it doesn't exist. Different machine learning techniques are ideal for solving different classification challenges.
Genre 2 - To PhD or Not To PhD.
The best are highly educated.
As a member of a Shakespeare Club, we love to pay tribute to the bard. We therefore title this genre with a bastardized take on “to be, or not to be”. One of our fundamental ALIS premises is that an ALIS manager only needs one or two PhDs, not a hundred or multiple hundred, as some of the world's best second wave quantitative, or computational finance managers employ.
However, while ALIS managers don’t need a large quantity of PhDs, quality is critical. ALIS leverages the strength of man plus machine, which will outperform either man or machine individually. Generally the world's top-decile ALIS managers have PhDs from the world's best schools, though not necessarily Ivy League, but often technical. The degrees may be in statistics, particle physics, epidemiology, machine learning and robotics to name a few.
Though we have identified one ALIS manager without a PhD, he is exceptional, went to a top school, studied computer science, was a gamer and hacker (though in a benign way) at an early age, and effectively educated himself in many of the machine learning techniques and technical trading of the markets. This manager is exceptional. The rest of the top decile ALIS managers generally have 1 or 2 PhDs.
The worst have less education and experience.
We did a meeting with an ALIS manager who discussed his PhD. However, upon closer examination, we found that the manager hadn't completed his dissertation and therefore didn't have one. This manager’s non-PhD was from a school in a state with a nice landscape, but that was about it.
Though as previously stated, we prefer (and generally require) PhDs to MBAs for ALIS managers, there is one crucial caveat. An ALIS manager with a PhD who doesn't understand investing is an automatic pass. As one of the world's great investors we worked for once said, "If I could short that fund I would." And that applies to some ALIS managers.
For example, we met an impressive ALIS PhD with strong machine learning. However, there were two primary and irreconcilable problems with the fund. The first was the fund naked shorted options. That is a strategy we learned early on in our careers to avoid at all costs due to the left-tail risk. The second issue is the Prime Brokers convinced the manager to turn liquid large capitalization securities into illiquid swaps. Perhaps this is revenge of the MBAs. In any case, it is an example, where the man in the man plus machine formula must at heart be a strong investor and understand markets and investing. Merely having a PhD is not enough.
And sometimes there may be plenty of PhDs, years of research and inordinate amounts of money invested, but nothing to show for it. Because Silicon Valley has revolutionized and disrupted many industries ranging from the taxi business to advertising, there often is a view that Wall Street or investing is no different. But it is, at least in our opinion.
For example, we have seen ALIS managers in the unsecured lending space with (and without) PhDs who have come up with models that were effectively short a put, and worse, levered them up only to lose large amounts when the loans stopped performing. More troubling are managers in this space making loans with no credit, distressed or work-out experience. We believe that these are likely to lose large amounts of principal in the next economic downturn.
Genre 3 - Back-test Heaven
The best know that history doesn’t always repeat itself.
The top-decile managers that we have identified all have actual track records, ranging from a year to a few years. The actual returns are impressive, not only in terms of level of return, but also due to the quality of the returns. They are low beta, high alpha, uncorrelated to indices and are generated by idiosyncratic sources. These return streams are not predicated on easily and inexpensively replicable factors. They also are not long-only or long-biased. With top decile ALIS managers, one is paying for alpha, not beta.
Moreover, because ALIS managers are typically small and emerging managers with low cost structures, they are amenable to investor friendly and aligned fees. Some top decile ones have adopted the 1/10/20 fee schedule that we, Jeffrey Tarrant, Adil Abdulali and I, espoused in a Pensions & Investments article entitled “A Perfect Solution.”
The worst live in the past, in back-test land.
The bottom decile of managers usually have phenomenal back-tests, hypothetical or pro forma returns, but no actual returns. They are “heavenly” theoretical return series that were over-fit, and when actual dollars are deployed they become hellish.
There would have to be quite an extraordinary circumstance for us to invest in an ALIS manager that does not have an actual track record. For example, as multiple members of the MOV37 team were investors in Renaissance Medallion, in our opinion one of the world's best funds, if Jim Simons were to start an ALIS fund, we would be more than happy to consider an exception.
However, all too often with ALIS managers, they may not fall in the two aforementioned categories, but they ascend to back-test heaven. There is no dearth of articles on the flaws with back-tests, and the typical commensurate over-fitting, so we will not spend much time on that other than to say, machine learning can inherently lend itself to over-fitting if not used properly.
What we often see with ALIS managers is a back-test that generates, to quote an '80s icon, Crazy Eddie, “Insane” returns, often 20-30% per annum, or Sharpe ratios in the mid-single digits, or higher! Then at some point the managers scrape up enough capital or intestinal fortitude to actually launch the fund. That's when they go from Back-test Heaven to Actual Hell.
We met with one of these managers that crossed the chasm. Interestingly the fund went from the aforementioned heavenly back-test return rate to a mid-single digit actual return rate. Moreover, the Sharpe ratio fell off a cliff as function of the quality of returns having gone from a few mildly poor periods, to many very bad periods.
Along these lines, there is another ALIS manager that we think very highly of but came from a long-only background, lacked short data and consequently had a back-test with a broad index hedge, rather than security specific shorts. We helped the manager source short data, and explained that after a track had developed with security specific shorts, we are happy to revisit.
Genre 4 - The Disreputable
The best have integrity.
We have thoroughly vetted our ALIS managers. They are who they say they are, situated where they say they are, doing what they say they do. They have been forthcoming with the techniques they use without revealing the details of their secret sauce. They are willing to be transparent with trades, because it is next to impossible to reverse engineer an ALIS strategy.
The worst are frankly disreputable.
Though fortunately only a tiny minority of funds in the ALIS universe, we have come across a few which we refer to as the disreputable. We have not spent the time to confirm or disaffirm their disreputable status, because for our purposes, all we know is that they are bad enough to not warrant further due diligence.
One such manager asked us, "How much and how quickly we could invest?" Our response was a highly respectable institutional amount within the next 6-12 months. After the manager had only run a small amount at the aforementioned “insane” rates of return, they went on to tell us, "That the amount you suggested may be insufficient for them to warrant expending their limited time and resources on our due diligence process." They also went on to ask, "Could we not invest more quickly?"
Another manager told us they learned from some of the world's top investment managers, including ones we knew. It turns out they never worked for those same managers. Instead, they were at a service provider who worked for those managers, and said they learned from seeing the positions and trades. By that token, anyone who studies 13Fs can learn from the world's best managers.
We once set up a dial-in conference line for ourselves and an ALIS manager in another country. We asked them, "Do you have an office and are you all there now?" because it wasn't clear that they did. They then went on to say, "Yes and we are there now." However, because we had the dial-in number, we could see that there was not one dial-in from their side, but rather there were multiple. So even if they did have an office, they prevaricated in terms of where they were at the time of the call.
Yet another ALIS manager we know, who fits into multiple genres, including Acronym Soup, Back-test Heaven and disreputable, both in one-on-one meetings and publicly to the media brags about their systems picking up on insider trading patterns. At least they are making it easy for the Securities Exchange Commission, SEC, to find them.
In summary, we have spent the past two years on this proverbial long, strange trip. It has been extremely rewarding separating out the wheat from the chafe, or bifurcating the top decile manages from the ones in the four worst genres. Though we have over-simplified much of it here for the sake of brevity, the research process was long, arduous and complicated. In many instances it took multiple meetings and much analysis to truly differentiate these managers. We look forward to spending the next few years continuing our long strange trip sleuthing out the world's best ALIS managers. Having come so far on this trip, we are certain ALIS will revolutionize investment management as the counter-culture did approximately a half a century ago.
To contact the author:
Michael Oliver Weinberg, Chief Investment Officer at MOV37 and Protege Partners: [email protected]