Yes but when will we program a computer to play ping pong? Photographer: Daniel Acker/Bloomberg
Yes but when will we program a computer to play ping pong? Photographer: Daniel Acker/Bloomberg

The Internet has paid a lot of attention today to this National Bureau of Economic Research working paper called "Buffett's Alpha,"1 whose principal conclusions are:

1. You could build a robot that replicates Warren Buffett's stock-picking performance.

2. The ingredients of that robot are (1) 1.6x leverage, which is what Buffett runs at Berkshire Hathaway Inc.,2 and (2) a statistical factor-weighting investment approach that overweights what the authors call "high-quality" (growing, high-payout, profitable) companies with low betas.

3. AQR Capital Management LLC, the quantitative hedge fund that employs the paper's authors, can build such a robot with trivial ease, really by just pushing a button, it's no problem, watch, they will do it for you.

So that is interesting, I guess, if you like robots. It is also great advertising for AQR obviously, though they'd need to go match Buffett for the next 40 years before anyone would entirely believe in their robot-building skills. Actually it was initially somewhat puzzling to me that the venue for this is an academic paper written by three AQR affiliates, rather than a flashy launch of the AQR Buffett Replicator fund, but you can probably figure out why that is.3 Anyway, they seem to regard the Buffett-replicating exercise as trivial; it vindicates some of the quantitative signals that they like, but for them the point is not replicating Buffett but rather picking the signals that offer the best risk-adjusted returns.

So why write about Buffett at all? You'd sort of think that the goal of the research done at quantitative hedge funds is to find things that make money and then go do those things, and generally you'd be right in thinking that. (Sometimes it doesn't work.)

But there is another, subtler goal, which is to fit those things into a narrative about the world that is comprehensible by humans.4 "My computer told me to buy these 800 stocks and sell those 700 stocks, and historically my computer has been right 63.4 percent of the time" is a terrible story. Computers crash and go haywire and try to murder their creators. Statistical anomalies vanish. Signals work until they don't. Markets mean revert. Correlation without causation is suspect. Building a computer program to grind out gains in the markets using a complex of factor weightings sounds like the sort of over-complicated, short-term, zero-sum, high-frequency, incomprehensible financial engineering that landed us in various messes and that is not particularly in public favor these days.

But Warren Buffett! Everyone likes Warren Buffett. Value investing, buy and hold, Cherry Coke, the whole schtick is so wholesome and appealing and throwbacky. Saying "Warren Buffett's investment process is really just overweighting a couple of statistical factors and levering them 1.6 times" is really a way of saying "levered quantitative hedge fund investing isn't so different from what Warren Buffett does, you know," though of course it is.5 Translating folksy value investing into the language of quant hedge funds is a way not only to analyze the folksy investing but also to make the quant funds feel down-to-earth and accessible. "We're just like Warren Buffett," they can say. "A cold, heartless, non-ukulele-playing binary version of Warren Buffett, true, but the resemblance is still remarkable."

That's good PR, but it's more than that. If your quantitative research is good, it will identify statistical anomalies that have persisted for the last X years, and you will go and trade on those anomalies, and you will make money for the next Y years. But in some sense you can never feel sure of yourself: No matter how long the anomalies have persisted, if they're just brute statistical facts they could always go away tomorrow. Your anomalies have no soul. Warren Buffett appears, from publicly available evidence, to have a soul. If he confirms your anomalies, and thus your investing approach, that tells you that you're onto something in a way that just making money never could.

1 Released by the National Bureau of Economic Research today, but floating around the Internet for years, as with all things. This one, though, I wrote about when it was previously floating around in 2012. But that was over a year ago, statute of limitations, etc.

2 Also the leverage is cheap because Berkshire is highly rated and a lot of it comes from insurance float which is very cheap. Others have noticed this and tried to replicate it.

3 "We can reproduce Warren Buffett's performance in our hedge fund, would you like to invest?"

"Um what are the fees like?"

"Two and twenty of course."

"Hmm. Why don't I replicate Buffett's performance by just buying shares of Berkshire Hathaway for zero fees?"


I guess you launch the Buffett Replicator when he dies or retires, and then it's a horse race as to whether your robot or his successor does a better job of imitating Buffett. Or you sell the robot to the successor I guess. Though, I mean, too late, now they've published it.

4 So AQR's philosophy includes the sentence "Investment strategies -- whether based on academic study or practical experience -- must be grounded in solid economic principles, not simply built to fit the past, and should contain as much common sense as analytical firepower."

In somewhat related AQR news today, AQR founder Cliff Asness published an article called "My Top 10 Peeves," which is almost the least appealing headline imaginable -- I guess the actual least appealing would be "Rich Man Has Long List of Complaints" -- but is otherwise pretty good. I particularly like #6, about how Asness doesn't like people who do various flavors of active management but insist that it's not that at all:

To me, if you deviate markedly from capitalization weights, you are, by definition, an active manager making bets. Many fight this label. They call their deviations from market capitalization -- among other labels -- smart beta, scientific investing, fundamental indexing, or risk parity. Furthermore, sometimes they make distinctions about active versus passive based on why they believe in their strategies. You can believe your strategy works because you're taking extra risk or because others make mistakes, but if it deviates from cap weighting, you don't get to call it "passive" and, in turn, disparage "active" investing. This peeve may be about form over substance -- marketing versus reality -- but these things count.


5 I mean, nobody's claiming that he's actually running regressions and looking at factor loadings. Just like, some of the fundamental analysis he does can be reduced to a regression.