Beating the market is hard. Notoriously hard.
A large majority of professional portfolio managers, after deducting their fees, don't do it. And Warren Buffett - one of the greatest investors ever - famously said regular investors shouldn't even try: just invest in the S&P 500 index and call it a day.
A lot of investors are taking his advice, piling billions each year into passive benchmark funds. This trend is putting pressure on active portfolio management firms to up their game, and they're increasingly using AI to do it.
In professional investing parlance, a stock's return above the market average is called its 'alpha.' This is the magic quantity portfolio and hedge fund managers chase to justify their fees.
With stock picking intuition and semi-quantitative research not consistently creating alpha, the investment industry is urgently turning towards AI to beat the market.
The alpha of an individual stock can usually be traced back to a set of factors identified to be the prime contributors to the stock's market outperformance (or underperformance). These are often obvious, like the cost of fuel for an airline or the price of steel for a ship builder. But they can also be more nuanced.
Quantitative investment funds build so-called factor models to identify the multiple causes that determine alpha for each security. This is best done using machine learning to correlate potentially explanatory past data with historical stock returns.
Here's where it gets interesting.
Building factor models is often the easy part. The hard part is determining which direction the factors will move before the market does. That is, in the examples above, where are the prices of fuel and steel headed? To sift for clues, firms are using deep learning at massive scale on big data.
In the past, hedge funds would dispatch people to retailers' parking lots to count cars - the simple idea being more cars predicts more sales. Now imagine you could use deep learning to extract in near real time vehicle or ship locations from satellite or drone data on a massive scale.
Companies like SpaceKnow and Orbital Insights are gearing up to sell this capability to investment funds. Another firm, Quandl, sells access to proprietary datasets containing micro- and macroeconomic insights waiting to be extracted.
This idea of proprietary data feeds is not new. Bloomberg, for example, built its business around real-time market price data and market news. But quantitative investment firms are now taking to a new level what deep learning can do with this type of data, and how it can generate alpha. In the quantitative investment world, this new ecosystem of data is being referred to as Alternative Data, or Alt Data for short.
Quantenstein, based in Europe, is an example of a startup long-term investment fund using this AI approach. It's constructing equity-based portfolios designed to generate alpha while carefully managing risk . (Watch Quantenstein's superb talk at GTC.)
What's interesting is that these AI-powered quantitative techniques are being applied to funds that take a long-term approach geared to general investors.
For hedge funds targeting shorter time horizons, the daily ebb and flow market sentiment can play a large role in short-term market movements. Toronto-based Triumph Asset Management says it's achieving an encouraging 76 percent accuracy rate on initial testing of sentiment analysis based on natural language processing deep learning methods.
Investment services behemoth Blackrock recently stated in a New York Times article that it will be steering its actively managed funds toward this type of AI-based quantitative strategy.
So, the race is on to incorporate AI and deep learning into investment decision making. And with the vast sea of data out there, the whole world is fair game to become the next investment factor model.