It has become commonplace for users of online platforms to expect that their attention i.e. time spent using the platform, converts to loyalty -- in the form of an artificial intelligence algorithm that knows them better over time e.g. auto-populating search fields, recommending preferred clothes to wear, books to read, or food to eat. Yet, when it comes to applying such sophisticated algorithms to financial markets, why aren't such quant funds always outperforming the market?
Artificial Intelligence is most useful where the problem set is narrowly defined, i.e., it is well known what is being optimized and how, and where the fuzzy data needs the structuring at scale that AI provides. A narrowly defined problem may be – given this particular set of personal characteristics about a person, should they be allowed to borrow this particular amount of money based on prior examples. A poorly defined problem may be – predict the price of a stock tomorrow given thousands of inter-correlated data points and their price history. It all boils down to the reliance of quant investment strategies reliance on pattern recognition: models look to correlate past periods of superior returns with specific factors including value, size, volatility, yield, quality and momentum. Such approaches have several fundamental weaknesses: (1) hindsight bias — the belief that understanding the past allows the future to be predicted, (2) ergodicity -- the lack of a truly representative data sample used in the model, and (3) overfitting -- when a model tries to predict a trend in data that is too noisy i.e. too many parameters or factors. Logically, over time the anomalies that these quant strategies are relied upon to exploit should dissipate, given the swift pace at which technology, competitors, and data moves to correct such anomalies. This is not stopping the likes of augmented analyst platform Kensho (acquired by S&P Global for $550 million), crowdsourced machine learning hedge fund Numerai, and the industry-leading quantamental funds of BlackRock. There is an inherent contradiction in that the approach exploits inefficiencies, but requires market efficiency to realign prices to generate returns.
With Cryptocurrencies, the strategies are different. Native Cryptocurrencies i.e. Ether and Bitcoin, are considered unconstrained assets, with limited correlations to other assets. Additionally, the data sets and factors that need to be considered when trading Cryptocurrencies are far fewer — many of which are speculative and co-dependent, resulting in far more predictable patterns than in financial markets. Because most of Cryptocurrency trading is autonomously and algorithmically driven, patterns are more easily discernible and human trading behavior often sticks out in stark contrast to established market behavior.The issue of course is not the opportunity to profit — it’s the magnitude of such profits. Currently, Cryptocurrencies simply do not have the volume and liquidity necessary for autonomous trading strategies to be deployed in large quantums. Percentage returns for algorithmic Cryptocurrency trading may be significant, but beyond certain volumes, especially when assets under management start approaching the hundreds of millions of dollars, traders need to get far more creative and circumspect in deploying funds as the opportunities are far fewer at larger order sizes.
For now at least, AI and machine learning are still some ways away from consistently beating the financial markets, but with a bit of tweaking they may be a lot closer to beating the Cryptocurrency markets. Evidence of this is already beginning to show -- in 2018 Swiss asset manager GAM's Systematic Cantab quant fund lost 23.1 percent, as well as, Neuberger Berman is considering closing their factor investing quant fund over poor performance. All this whilst Cryptocurrency quant funds returned on average 8% over the same period. While the prospect of searching for phantom signals that eventually disappear could dissuade some people from working in finance or Cryptocurrency trading — the lure of solving tough problems coupled with the potential to dip into the $200 billion opportunity means that there will always be more than enough people who will try.