macrocephalopod

You're on the executive committee at a $10bn hedge fund with a 10% annual vol budget. You've hired 100 long/short sub-PMs to manage money for the fund, each running a concentrated portfolio of 20-50 stocks.

For reasons well explained by @quantian1 in the latest @trashfuturepod, despite being nominally completely independent, the PMs will end up doing similar things. They read the same research, speak to the same analysts, and have broadly similar education and outlook.

In particular there are certain kinds of positions that they love to get into -- buying stocks that have been going up, that have been recently upgraded by analysts, where sentiment is very positive etc. More mechanically they will end up taking positions in highly shorted ...

stocks (a boring truism because the way that stocks become highly shorted is that lots of hedge fund PMs think it's a good idea to short them). These are your factor exposures.

Some of the factors (e.g. momentum, short interest) are quite "crashy" and will lead to large simultaneous drawdowns across all PMs at the same time -- something you *really* want to avoid because you are a hedge fund so you are levered up to your eyeballs.

The PMs do have skill though! Their individual stock picks are pretty good, it's just that they come with all this factor exposure which (a) is not well compensated given the risk (b) causes the PMs to be more correlated with each other than you'd like.

What do you do? You build a factor model and use it to hedge out as much of the factor risk as you can. Specifically, you want to find stocks that are not held by many PMs but have similar factor exposures to the stocks they do hold, and use them to hedge the factor risk.

This costs, both in t-costs to adjust the hedge and because the hedge is short factors with positive expected return -- but it reduces the volatility of the portfolio by more than it reduces the return, so the Sharpe improves and you can apply leverage to juice the return.