From the book "Finding Alpha", written by a popular quant fund WorldQuant, explains many techniques about quantitative investing but intentionally omits many of the caveats and applications of those techniques.

In the book, they introduce a concept of 'factor grouping', which is as simple as grouping stocks by their sectors/industry.

Sometimes a stock's return is demeaned by the group mean or ranked within the group.

Magically, this technique yields very good results and make an unprofitable 5-day reversion strategy to something that actually generates profit.

The book says "It works", but doesn't explain why. Why does it work? What's the intuition behind all the 'grouping' techniques?

  • 1
    $\begingroup$ This is commonly done, it can be seen as an attempt to achieve a better balance of industries in the final portfolio. For example if you simply buy high B/M stocks you might end up mostly with utility stocks and no tech stocks. If you buy the highest B/M in each industry you will have better balance, including some tech stocks. There are numerous published articles that compare for ex. "momentum" vs "within industry momentum". Of course there is a risk of overfitting by trying numerous variants like this. $\endgroup$
    – noob2
    Oct 13 '20 at 12:02
  • $\begingroup$ @noob2, Thanks for the explanation. However, is there any specific papers/books that explicitly discuss these topics? Not limited to sector/industry but extends further to any other factors in general. $\endgroup$ Oct 13 '20 at 17:53
  • $\begingroup$ I don't know any general treatment, I have seen grouping on Industry, Country, or Asset Class. For industry, one paper is Asness: Predicting Stock Returns Using Industry-Relative Firm Characteristics (2000). There are many others since then. $\endgroup$
    – noob2
    Oct 13 '20 at 18:49
  • $\begingroup$ @noob2 Thanks for the reference. $\endgroup$ Oct 14 '20 at 13:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.