Pit.ai puts a financial twist on reinforcement learning to outperform hedge funds

Despite mystery and intrigue, the reality is that most hedge funds don’t make money. This hasn’t stopped a growing list of startups from trying their hands at employing machine learning to tip the scales in their favor. But Pit.ai, a new machine learning-powered hedge fund, adopted into the YC W17 class, thinks it can best Numerai, Quantopian and others with its own unique recipe for automating money making.

Hedge funds employ aggressive trading strategies to “seek alpha,” which is industry jargon for above market returns. These are not your standard trading shops, and over the last decade firms have gone to great lengths to seize data for information arbitrage. There’s no shortage of startups attempting to sell data to hedge funds — geospatial analytics companies, for example, service funds using satellite imagery and computer vision to count cars in retailer parking lots to project earnings before official revenue reports.

Pit.ai founder Yves-Laurent Kom Samo

Pit.ai founder Yves-Laurent Kom Samo explained to TechCrunch in an interview that most of these funds have one thing in common — they’re using information that hasn’t been factored into stock prices to predict returns. Breaking with that paradigm, Pit.ai is using a variant of reinforcement learning to evaluate trading strategies instead.

Traditional reinforcement learning, a type of machine learning, relies on value functions — a construct that should be familiar to anyone who has studied economics.

Imagine that you wanted to create an agent that could drive a race car in a video game. The reinforcement learning approach would require establishing some notion of utility or value for various decisions. This could be a loss of arbitrary points for driving off the road or a gain of points for increasing speed. Strung together in overly simplistic terms, an algorithm can be fine-tuned over many iterations to make optimal decisions by estimating value functions.

Yves-Laurent explains that this strategy falls short in a financial context because it means that learning how to trade requires one to model returns for each decision in each state of the market. Financial markets are incredibly complex systems, so the math goes from science to art to pseudoscience quite rapidly. Instead, Pit.ai evaluates trading strategies themselves, taking into account metrics like Sharpe ratios and maximum drawdown — financial tools for evaluating risk.

Utilizing this strategy, Pit hopes to best industry stalwarts by not only delivering above average returns, but breaking the traditional two and twenty fee structure of the hedge fund industry. Without the need for large analyst teams to search for macro-economic trends and data to exploit, Pit can stay lean and drop management fees altogether, instead opting only to collect carry from its limited partners.

While Pit has yet to raise a fund to trade with from these LPs, it is raising venture capital to support a handful of machine learning experts. Yves-Laurent was a Google Fellow and garnered a PhD from Oxford in machine learning, so he expects to be able to use his network for recruiting. He has been running his models without real money and notes that signs are very promising. Within a year, Yves-Laurent hopes to have pulled together a fund and initiated formal trading.