Tuesday, March 25, 2008

Have we quants been brainedwashed by Barra?

This is a line from Star Trek Next Generation when the Enterprise encounter the Borg and the Borg say something like:

We are the Borg, you will be assimilated – resistance is futile.

As a quant I feel like that sometimes when I encounter Barra and its software. The head of a prop desk once complained to me that everyone is using Barra and they were all getting the same solutions. So when the hedge that the software suggested turned sour, the effect was worse because it seemed that everyone else was rushing for the same exit at the same time.

There are also second order effects. Barra has taught us that the sources of equity risk are industry and common factor (style, size, etc.) and it has affected many quants' analytical frameworks. I was recently at a meeting of quantitative analysts when someone presented some equity analysis. There was general agreement was that the solution wasn’t very well-risk controlled as no self-respecting equity quant would compare two stocks (one in Autos, the other in Media) in the same sector against each other (Consumer Discretionary) as they would normalize for industry effects.

Have we been brainwashed to control risk by industry and common factor (and not much else), after being exposed to the Barra risk framework all this time or is this just another example of a crowded trade?

Comments welcome.


Walt French said...

I, for one, would welcome some new risk management Overlords! There are lots of insights in the world, and MSCI/Barra certainly has no monopoly.

But I've also sampled three other services over time, and built my own multi-factor model. All have had their Nifty Wrinkles... I was especially proud of my industry factors and mispricing factors that were self-aware about the risks they embedded, rather than being pure "alpha."

But I wouldn't expect that if 20% of the market each were using a different model, the carnage of last summer would've been much different. As spelled out in some of the "Dear Client" letters the WSJ posted, the imploded strategies all had various valuation, momentum, earnings and size bets that they were NOT hedging in Barra's or any others' frameworks. These bets were seldom built on Barra's factors, but rather some proprietary variations of cash flows, earnings estimates, asset valuation, etc. that were like Barra's "Earnings," "Value," etc.

Optimizing to emphasize a systematic theme that's unlike any factor in your risk model is a pretty good way to concentrate risk. For example, my old model had a factor that measured short interest ratio, and I could optimize to emphasize mostly it, and no other factor. Barra was ignorant of it, so if I built a portfolio that bet heavily with (against) the short sellers, it looked like a bunch of uncorrelated, stock-specific, low risks. In fact, though, this type of herd-following is a reasonably important theme, and "works" (or not) across most of its stocks in any period. The portfolio will have much higher tracking error than such a model will show.

You can have the same bad experience with systematic themes that are like your risk model's factors, too, as in last summer's meltdown. In that case, you'll concentrate your risk into the names where your factor is just enough different from Barra's. Either way, you're guaranteed by the Law of Error Optimization (just made that up!) to have much more risk than you thought. A well-specified risk model will measure risks of random portfolios, or those created exactly ON the risk factors themselves. Others need elaborate mechanisms to supplement the risk matrix, if you want a good risk forecast.

The fact that there were huge performance hits associated with being on the wrong side of these postures is actually a testament that Barra's broad themes actually DO capture real-world risks. Alas, they don't take responsibility for the wisdom of your doing so.

Rather fat-tails on the risks; it'd have been nice to have understood those. And yes, those could've arisen from panic selling. But if you believe that (I do), you must also believe that the upside many "quant strategies" enjoyed was ALSO driven up as several hundred $billions flowed into them. Most quant managers are smart enough not to push that self-belittling line of reasoning too hard.

Cam Hui, CFA said...

It seems that virtually every consultant I have encountered uses Barra. I would say, therefore, that from a marketing and client service viewpoint Barra is an important piece of software just for that reason.

Putting on my investor hat, however, I find it to be a very scary application for the reasons I mentioned before.

Jenn Bender said...

Hi Cam,
I appreciate your honesty! I can't comment on how many clients are using our models and in what ways (Legal would have my head!) but I CAN say that it's not use of the same risk models that causes increased correlation in times of meltdown/liquidity crunch. One of my colleagues is working on this problem and he can prove it's not the risk model. However, I think other things can cause strategy correlation - same universe, same optimizer, same constraints, same rebalancing, etc... And that's not addressing correlation in the tails which is a whole nother ballpark! Your friend at Barra, Jenn

Cam Hui, CFA said...


I am not blaming the risk model for August 2007.

The issue is that if everyone has the same problem, e.g. if I am long S&P 500 but I want to hedge against Russell 1000 (a close fit), everyone gets the same solution. If something happens in the elements of that hedge (such as a stock surprise) what was termed specific risk, or residual risk in Barra terms, becomes a Barra systemic risk.

The second issue is Barra has taught everyone (I think correctly) that most of the risk comes from industry and a few common factors (Style, Size, etc.) Such a framework, while useful, stifles creativity. In other words, this framework creates the "box" and you need to think "out of the box" once in a while.