How Quant Funds Work (in 1 paragraph)
How Quant Funds Work in 1 Paragraph
The first investors to take advantage of computers to process data and make investments were ‘quant funds’ that continue today to be successful at scale. The tools at their disposal were statistics and linear models – limited by the computational capacity and the availability of structured data to apply statistics to. Today – the successful quants have become enormous, managing $30+ billion and the smaller players moved towards high frequency trading. At their core, traditional ‘quant’ strategies are based on statistical correlations: i.e. linear models. At a small scale - stand on the surface of the earth, the horizon looks flat – step back to big scale in space and you see the curvature. Same idea with statistical correlations with time scale – in the next day or week the linear correlation can be a fairly good representation of reality but over the next year you’re in the flat earth society. The consequence of this is that the strategies tend to be high-turnover: a 1 week holding period means trading your portfolio 52x per year. Strategies with lots of trading tend to require scale to have low enough trading costs (including the infrastructure to execute) to be practical. But of course scale (>$1B+ AUM) means that you can’t take meaningful positions in small companies to move the needle. This size barrier to entry has meant there have been few new entrants to the quant fund landscape for some time and why, despite having an engineering background, we did not focus on this direction at first. Long-term success is driven by competitive advantage and we had little chance of creating against large incumbents playing their game.
Instead, our approach has been to ‘depth first search’ as investors by being entrepreneurs and activists and invest in smaller companies where we could be the largest and most sophisticated – and energetic! – stakeholder. This gave us a competitive advantage and a number of demonstrated successes. Along the way we have thought hard about how to generalize what we have learned about what drives the change in intrinsic value over time and codify it so that we could continue to screen and search for similar situations later in a systematic way. Lots of investors know how to look for clues to a stock being mispriced relative to its apparent intrinsic value NOW but very few have been in the trenches seeing how intrinsic value can increase or decrease from management decisions in the boardroom.
Machine Learning Changes the Game
Starting several years ago the landscape around the traditional quant funds shifted. New technology has allowed for 1) the ability to work with unstructured data (i.e. natural language) that can be gathered less expensively and 2) nonlinear predictive models. These tools were extremely expensive until the last year or two and impractical to use for the investment process. Now it’s possible to build a machine learning investing system with a small group of talented engineers using open source software and low cost cloud computing. Add to the formula an activist investor, who also happens to be an engineer, to help direct what data factors are important to predict how a company will perform in the future and that is exactly what we have now built. I introduce: Meson Gravity , our machine learning system to predict the long term performance of companies using data.
Our approach, although utilizing computational tools, is fundamentally the same business-focused approach we have been deploying for years employing a long term perspective. The term “quantamental” (quantitative + fundamental) has been recently popularized to describe this class of strategy. We aren’t competing with other quants on trading-like timescales – we continue to focus on small companies where the markets are less efficient and we can compete against other predominantly human, emotional, biased investors. Now we have a machine. Most of our competitors don’t, because it is very hard to build. What usually happens when machines compete against humans at the same game? Year when machines defeated the human world champion: Checkers (1990), chess (1997), Jeopardy! (2011), go (2016), poker (2017).
Mr. Market’s “votes” reflect the current state of the world and clarity of a company’s prospects ahead. The data accounting for this has continued to become more available – from credit card transaction data to Walmart parking lot satellite imagery. I can’t imagine making a strong argument for our ability to compete from a position of strength in this dimension. We have become experts in the “weighing machine” and understanding the effects of “diet” on future weight of a business, to extend the analogy to include time. This has been learned from the inside out after experience on half a dozen corporate boards and numerous other entrepreneurial experiences. I believe understanding what drives a business to change over time in a fundamental way is likely to be out of reach from the pure quants for some time.
This is part 2 of 3 of a larger article on how machine learning and AI has changed what is possible for long term value focused investing. Click below to learn more.
