• John Hawver

Warren Buffett, the OG Quant

Last week I decided to improve my equity ranking model. The model has been a stalwart of my investment process, but in the time since I last updated it, I’ve learned a few things that I wanted to incorporate. This is what finance nerds do late at night. This time however I had a moment of clarity and “learned” something that has been in front of my nose for quite a while, so I thought I’d share that insight.

During this same week, I had also been conversing with a good friend of mine about an early-stage investment he had made that worked out very well. This factors into my insight and I’ll explain towards the end.

Let’s start with my equity model. This model began years ago as a simple stock screen and has evolved dramatically since. Its current version is basically a multi-factor predictive regression. If that sounds scary, it shouldn’t; I’m basically taking historic data for each stock, adding some “factors” (features) that I think are more indicative of that company’s health, and then seeing if those factors have some meaning for the future returns of the stock. A load of fancy math goes into this process, but at its core, it is a simple idea; if you think something influences the future, then test it.

So what did I change? Well, my data set has grown considerably larger since I last updated this model, and so have my machine learning modeling skills. So I set out to apply both of those. A core tenet of machine learning (and don’t let this term scare you either, it is simply a fancy way of saying statistical modeling) is to split your data into a training set and a test set. This is a basic scientific process; the statistical model is “trained” on a set of data and then fresh data, the test set that the model has not seen, is applied. The results from these two are compared. If your model is good it should work almost as well in your test set as it did on the training set, ie it should retain as much “intelligence” as it can.

What I learned from improving the model opened my eyes. The last model version was not as good in the test set as the new one by a long shot. That was a good result, I had improved the predictability. But as I examined the different forward horizons that the model was predicting, it became apparent that there was a limit to how far it could accurately predict into the future. That limit was somewhere at about three months. This applied to both my last model and my new one. Three months was a little disappointing so I examined monthly horizons out to a year; there was a definite fall off in predictability after three months. On the positive side, the model was doing an excellent job of finding undervalued stocks that were mean-reverting back to their fair value; these stocks had a clear margin of safety, which if you’ve read my early posts, this is core to how I think about investing.

Now, there could be many causes for that short horizon, my data-set, while large, is not huge. The factors that I’ve constructed may not be the best. There could be more areas to improve my process (there always are). The list of possible causes is long. So I stepped away from my work for a while and thought about it, and that’s when I had my realization. I had learned what Buffett had learned fifty (yes, 50) years ago. I’m such a quick study (note sarcasm).

Warren Buffett began his investment career as a value investor who liked to get the last “puff” of a worn out cigarette butt. It’s how he ended up buying Berkshire Hathaway. But he soon evolved into a “long-term” investor. He recognized the scale and efficiency in finding high-quality businesses and owning them for a long time; essentially index investing, which he currently preaches quite clearly. My model was finding lots of undervalued cigarette butts with a few puffs left. The model found, in stacks of data, what Warren had discerned all by himself. Warren was the “OG” quant.

My insight didn’t end there. If you do a survey of the current quantitative investing landscape and look at statistical arbitrage funds, high-frequency trading firms, news/event-based trading (which deserves its own post), etc, there is a clear trend towards investing (more accurately, trading) at smaller and smaller time horizons. Access to better data, better computers, and better modeling techniques has driven this short-termism. And I think they have all found what I have found: there’s a limit to how far you can predict mathematically for individual securities. There is simply too much noise in the financial markets to do so accurately.

Back to my friend's early-stage investment. I put some thoughtful work into trying to value his investment. (A core belief I’ve learned as a trader is that you shouldn’t trade/invest in something you can’t value.) But I couldn't arrive at more than a foggy number. Which disappointed me. In stepping back to look at my own model above, I also thought about this investment. How did it work out so well? How had my friend been able to make such a successful leap of faith with so little data? Well, just like my realization about Buffett, it occurred to me there is another very good way to invest long-term, but it is harder; invest in economy-changing trends early. This takes some insight into what is changing and then finding a company with the able leadership to execute upon that change. Think Amazon in 1998. Or Google in 2003. Or Intel in 1990. Or Apple in 2002.

So, my final insight came from asking this question: What techniques can an individual (or professional) investor use to invest long-term? There are really only three that I understand right now (hopefully I’ll continue to learn and catch up with Buffett!):

1. Factor investing. Like my model, use current data to base your investments on future predicted returns. This is not a very long-term method but should produce reliable returns.

2. Index investing. This is especially effective if you vary your weightings between stocks and bonds based on how expensive the stock market is. Index valuation is an excellent long-term predictor of future returns, even up to five years. I include equity value indices in my reports and a long-term index strategy (Navigator) based on valuation and risk tolerance in my investment reports.

3. Early trend investing. This is the hardest of the three methods and what venture capitalists attempt to do.

I hope this insight is helpful to you. I post my investment reports on the site, they should help you with investing in the first two methods. For the third, well, keep your eyes open; things are always changing and hopefully you can be fortunate enough to recognize those changes and courageous enough to act when you do.


©2020 by Mud Muscle and Markets - Disclaimer