On Investment Modeling, Part 4

I thought part 3 would be the end, but I ended up with one huge and good comment from a friend.  A real friend, not a Facebook friend.  I will respond to it in pieces.

Good article and series. A few comments and/or questions for you. First, there is no doubt that both value and momentum work. And while I happen to personally be a big believer in (certain) trend-following approaches, the way in which Covel interacts with people is childish at best. There is room for professional discourse and disagreement, but he has little interest in being professional about anything…at least in the blogosphere/twitterverse world. So, keep at it…be a gentleman and let the other chips fall where they may. Now, a few other areas:

I try to stay polite.  Sometimes I fail, but thankfully, it is not common.

1. First, I’d be interested to have you elaborate a bit further (or point me to a different post) on your views of the Carhart factors. In my mind, there is no doubt that they are betas…but at the same time, there is also no doubt that those betas can and should be used (carefully) as alpha factors as well. Much as my brain has been trained to think about them as simply betas, I’m more than willing to think about them in both ways. They aren’t mutually exclusive, are they?

I have no other post on this.  This series was meant to bring this idea buried in me to the surface.

I think the thing that set me off here is the value factor.  Value is regarded as a risk factor, when if you own enough stocks with the value factor, you will outperform over the intermediate term.  The same applies to momentum, it is a risk factor, but it tends to outperform.  The same can be said for size, small is usually a winner because of neglect.  Beta tends to be negatively correlated with outperformance.

If the factors were neutral, having zero expectation of future performance, they would be betas.  But that is not true even of “beta.”  Thus, most of the time one can make money by tilting to the moneymaking sides of these factors.

Recently, there was an analysis of Berkshire Hathaway for the last decade, showing Buffett had no alpha.  But if you looked at Berky versus the index, Berky beat it by 6%/year.  Buffett asks whether companies are cheap.  If he buys a company because it is specially cheap, or because its associated “risk” factors are cheap, that should be measured as skill, not taking risk.

If we add enough “risk” factors to the analysis, most alphas disappear.  But the ability of managers to buy when a risk factor is cheap does not go away, as does their ability to be “late followers” and lose money as so many retail investors do.

2. Second, and importantly, I don’t believe that every investment strategy can be boiled down to a fairly simple mathematical/quantitative approximation. Therefore, not every strategy can be tested in a purely academic sort of fashion. Take Covel’s trend-following, for instance. There are obviously many ways to do trend-following, but few are so simple as to easily do an historical test. It’s NOT simply momentum (as you well know!), and while momentum and trend-following certainly have some correlation, it would be a disservice to the TF crowd to view one as a proxy for the other. The buy and sell rules, timing of entry points, level of stop loss, etc. are hugely crucial to the success of the strategy.

You are right here, mostly.  What we test are only the quantifiable aspects of the strategy.  We don’t test, we cant test nuances.  Nuances will get lost in the noise.  We are out to test the first approximation of a strategy, not the strategy itself.

Most managers have an initial screen that winnows down the universe of stocks that they will then use their abilities to analyze.  Managers think that only a subset of stocks are worth their time, because that pool is likely to outperform, now let’s get the best of those.

In this sense, we are not testing the fullness of a manager’s processes, but only his initial quantitative screen.  Processes beyond that are alpha, whether positive or negative.

As one who has sat through many dog-and-pony shows (and you, friend, more than me), most managers fall into buckets off of their screens.  What is their investable universe?  We test that.  We can’t test the fine gradations beyond that — the law of small numbers interferes.

But what I will argue regarding trend following is that there is some measure of momentum that explains over 70% of the results of a wide number of trend followers, much as Buffett could point to the “Superinvestors” and claim that they were all one tribe, though the details differed considerably, much as Covel has done with trend following.  The first approximation of the group element is the important part tested.  Maybe we need to use principal component analysis to tease it out, but we do need to simplify the broad parts of the strategy for testing.  We can test the broad stuff.  Beyond that we are stuck.

3. So with #2 as an assumption, the only way to analyze TF is as a group of investors. Is there survivorship bias? Yes, but you have that with value guys, too. Is it enormously dependent on sticking with the system, even when it’s not working? Yes. Is there a good sample of auditable accounts out there? No…not so far as I’m aware. I think the issue that Covel has is that the majority of the best investment returns people have ever put up are from the momentum/TF crowd. However, one should very clearly separate investing from trading. The trading crowd has the ability to put up ginormous returns, but at what cost? Huge volatility, gigantic turnover, etc. that most people are not willing to live with.

Let Covel and his friends try to raise money from the institutional investment community.  We may admit that momentum works, but not the ability to consistently make money off of price/volume action when managing a large amount of money.  If they do have that skill, we need to explore it, and let the behavioral investors analyze it so that we get a first approximation, a factor, to explain it.

Survivorship bias? You bet that is there.  That is why we test mechanized first approximations to a strategy, not the strategy itself.  We test tribes, we don’t test families, much less individuals.

So in the end, as you’ve said before, it comes down to finding a system/approach that has shown the ability to work well for others and sticking with it through the tough times. No approach to investing/trading will be absolutely perfect every month, and most people lack the discipline to actually make it work over time. They switch from system to system, at the most inopportune times.

Thanks for the good work…keep it up!

You would know better than most that though I am generally a value investor, my own strategies are different because I use industries as my primary screen in investing.  And it is nonlinear — I look at those that are running, and those that are dying, but not the middle.  I consider macro factors that many do not, whether I am right or not.

I am one of those managers that would be hard to measure, if one wanted to measure things precisely; I don’t screen, as most managers do.  But I consider value, momentum, and mean-reversion effects to be givens, while I try to analyze what industries and companies will do well.

And that is a reason why I have not fared well with fund management consultants.  Like Covel, I do not fit their paradigm.  Unlike Covel, I would like to fit their paradigm.

But no, I am happy for the present to attract individual investors who want to outperform on a risk adjusted basis over a 5-year  period.  That is my forte, and I will pursue it with investors as my firm goes live at the beginning of 2011.