Machine Learning Stocks

There are no “new new” things in finance, there are no “unique” opportunities and the stock that you labored so hard to pick is no “diamond in the rough.”

The perils of static classification

In the olden days, stocks were classified either based on the industry they operated in (i.e., sectors) or based on their market cap (i.e. cap-weighted indices.) But neither of them do justice to the underlying risk or technical behavior of the stocks themselves. Not all stocks in an index are similar, nor do they have a 100% overlap with other stocks within the same sector. Buy say you really liked a stock, like ACCELYA for example, but found its valuation unattractive, how do you go about finding a substitute?

Finding substitutes using clustering

This is where machine learning can help. Using clustering algorithms, stocks can be grouped based on parameters different from just industry or market-cap. What if stocks are grouped based on risk metrics (alpha, beta, Sharpe…) and technicals (RSI, ADX…)?

This was the question we set out to answer and we are now proud to present the results in the “Quant” tab in the equities page. Here’s how the Quant section looks like for Accelya:

It shows that Accelya, although a retail value favorite, is not alone when it comes to its risk profile and technicals. There are other fish in the sea that is worth a gander.

This might be a scary thought for most investors: their favorite stocks are not alone, and there might be cheaper substitutes to the stocks that they like.

Geek Note

All factors are equally weighted and normalized.
A lot depends on whether the underlying data is actually clustered or not. For example, if you only cluster based on risk metrics, then by definition, most stocks will cluster around Alpha=0 and Beta=1.
Risk metrics look back 365 calendar days whereas technical metrics look back over shorter time-frames.
In case you are wondering, there is no information content in observing “cluster migration.”