To briefly summarize my point in my previous post, the barriers to entry of bottom-up equity quantitative analysis has dropped dramatically over the last couple of decades. As a result, the competitive advantage of using multi-factor bottom-up quantitative stock selection techniques has eroded considerably. We are all using the same databases and the same tools. Is it wonder why we wind up in the same crowded trade and bottom-up stock selection alpha is becoming such an arms race that no one can win?


Back to first principles: Modeling human behavior
To find the ever elusive alpha, it is important to go back to first principles and ask: Why does quantitative analysis work?

Unless you can convincingly answer that question, you will not find an enduring alpha.

The roots of quantitative analysis came out of the anomalies research literature written by finance academics starting in the 1970s. Remember the low P/E anomaly? The P/B anomaly? Small cap and neglect effect? That research was followed by inquiries into earnings expectations and surprise, etc. Investment managers took many of those insights and implemented them in a systematic way in their portfolios. Thus quantitative analysis was born.

What many quants never understood or forgot why buying low P/B stocks gave you better returns. Stocks with cheap valuations, as measured by low P/B, usually have something wrong with them fundamentally – a "yuck" factor. Buying them required an investor to hold his nose from smelling the "yuck" in the portfolio. Quantitative analysis gave you the discipline to buy those stocks.

It was true in those early days and it is true now. The value of quantitative techniques is the systematic application of a principle that exploited human behavior.

Many quants have forgotten the human behavior modeling part of building models.


Still an alpha in modeling human behavior
I can suggest a couple of ways to build quantitative alpha. Both of them require work and real change in the genetic disposition of how quants are trained and think.

The first is the geeky solution.

Today, most bottom-up multi-factor models use common factors like P/E, P/B, EV/EBITDA, etc. While that is a useful technique for valuing stocks from 30,000 feet up, why not use the powerful of the computer to get much closer to the ground?

We know that industry analysts analyze their companies differently.