We first saw program trading, where computers were preprogrammed to execute a stock trade based on predetermined conditions, in the 1980s. Today, over 1/3 of stock trades worldwide and well over 1/2 of stock trades in the US are executed by algorithms.
Algorithmic venture capital investing is far from common place today, and many argue it never will be. However, it would be hard to argue that data does not play an important role in the industry.
Pioneers in this space already use algorithms to make decisions. Matt Oguz at Palo Alto Venture Science ($200M fund) looks at 13 different variables for each prospective company, like technology, IP, people/team, location, and competition (source). WR Hambrecht Ventures works closely with Thomas Thurton of Growth Science and combined predictive modeling with Clayton Christensen’s disruption theory (source). David Koatz and David Kienzle at Correlation Venture ($166M fund) weighs the track record of the entrepreneurs, investors, and advisors heavily (source). Google Ventures ($1.5B under management) looks at data from academic literature, past experience and due diligence of founders and startup. (source).
Looking at survivorship of companies after a 10-year period, Thurston says his algorithm has a 66% hit rate, but it’s still too early to know if these algorithms can consistently out perform humans, and if so, in which areas. It would make sense that this algorithmic approach works better for later stage investing where there are more consistent and comparable metrics to feed into the algorithm, as opposed to early stage companies where there are fewer metrics to feed into the algorithm.
Some might suggest that this algorithmic approach would work best for niche categories, allowing the algorithm to specialize. Circle Up a crowdfunding site for consumer packaged goods employs an algorithm to help help evaluate over 500 deals a month (source). Deep Knowledge Ventures appointed an algorithm to it’s board, capable of making investment recommendations of age-related disease drugs and regenerative medicine companies based on a companies’ financing, clinical trials, intellectual property, and past funding rounds (source).
While these algorithms can process more information and deliver results quicker, most (if not all) funds still employ some human element to screen results before making final investment decisions (source).
By making enough investments, it’s not difficult to see how an algorithm might help a fund outperform the average venture fund. It seems difficult, however, for an algorithm based fund to be a top-performing fund, whose returns come from outliers – which are by definition the most difficult to identify via an algorithm.
Perhaps the role of algorithms in venture capital investing is not to replace human decision making, but merely augment it.