When does it make sense for an asset manager to customize its risk model? And why?
There are, after all, a lot of off-the-shelf risk models. For many purposes they suffice, and in some cases they do better than suffice: they render customization an expensive net negative.
That is one of the premises of a recent paper by Zura Kakushadze and Jim Kyung-Soo Liew, who say that pension funds for example “do not require customization but standardization.”
Seven Factors
The report, Custom v. Standardized Risk Models, quickly acknowledges that there also are circumstances in which customize models ought to win out over the SRMs. The choice depends on seven factors, these authors tell us. They try to quantify these and produce some pertinent computer code. I’ll simply list them.
The factors: length of the time horizon; inadvertent alpha neutralization; insufficient industry granularity; the size of the trading universe; the herd effect when SRMs are over-used; profit and loss correlations; and the specifics of the contemplated CRMs.
A word about each:
- For short-term oriented hedge funds, such as the statistical arbs, the SRMs do a disservice by including factors that are irrelevant to that time horizon.
For example, a corporation’s book value (updated quarterly!) is not a consideration that has any value “for holding horizons measuring in days or intraday.” It is worse than useless; it is the source of statistical noise.
- Alpha neutralization
Often the premise behind seeking alpha from a particular strategy runs directly against the advice embedded in an SRM. For example, an SRM will produce a portfolio with a certain capitalization risk exposure, although a given hunter of alpha may be employing a strategy that demands a focus on small cap values. In that case the RM has to be customized to suit the hunt.
- Industry granularity
The SRM coverage universe requires significant chunking of potential issuers into industries/sectors. A particular strategy may be much more particular than this, and may need an RM customize accordingly.
- Coverage universe
Likewise a manager’s strategy may focus on a rather small universe of stocks, excluding telecoms from consideration entirely, then. The RM will require adjustment.
- Herding effect
This speaks for itself, but it may be worthwhile to observe a connection between this point and point one above. Specifically, the herding effect can create a storm that longer-term funds and the managers can simply ignore and ride out.
- Profit and loss
If the P&L correlation of two distinct RMs is quite low, “it is optimal to run a combined strategy using both RM thereby reducing portfolio turnover and market impact.”
- Pros and Cons
This is something of minestrone of additional considerations.
The lead author of this paper, Zura Kakushadze, is the president and co-founder of Quantigic Solutions, a Tennessee-based company that markets the Quantigic Risk Model. He’s also an adjunct professor at the University of Connecticut and a full professor at Free University in Tbilisi. So he does, to speak quite frankly, have something to sell here.
After reading the paper I went to the Quantigic website, and encountered more about the Quantigic Risk Model. It is apparently intended to straddle the line discussed in this article, that is, to be standard in a sense but readily customizable. For example, “QRM risk matrices are not based on any estimation universe; rather, they are computed for a given input universe.” The matrix then can vary from a mere 500 companies, to the whole universe of U.S. exchange-listed companies.
Kakushadze’s co-author, Jim Kyung-Soo Liew, is an assistant professor at Johns Hopkins University, Carey Business School.