When financial gurus make sports bets

They were asking for trouble. The investment bank UBS wanted to put together a pamphlet for the FIFA World Cup, highlighting investment opportunities in host country Russia and talking about the economic impacts of the tournament. It is becoming a tradition for such shops to show off their quantitative firepower in these settings by doing a little sports forecasting. Their offices are hipdeep in math-heavy PhDs: they all understand beasts like ELO models and Monte Carlo simulations. And many professional investors have an unconcealed appetite for gambling. Making book in the schoolroom and playing poker are canonical enticements to the finance world.

So UBS created a World Cup forecast, and from a PR standpoint, it worked: the bank’s predictions, assembled by “a team of 18 analysts and editors,” were covered widely in the financial press. Of course they had done just as well sowing headlines with the 2014 tournament, for which they had made Brazil the favourite. Whoops. But who remembered that? Surely this time UBS’s anointed side, Germany’s formidable Mannschaft, would not disappoint ...

Well, even if you aren’t following the World Cup, you can probably guess how this story has already ended. Germany ended up finishing last in Group F, being beaten in the last round-robin match by (already-eliminated) South Korea as Sweden and Mexico sailed through.

No one is going to lose a UBS job over this, in case you were concerned, and perhaps no one should. The UBS model output each team’s probabilities of winning the tournament — but since they don’t play the tournament 10,000 times to match UBS’s proud boast of having simulated the games 10,000 times, the bank did not, strictly speaking, make a failed prediction. The headlines had said “UBS claims Germany will win”: what UBS actually said was that Germany had a 24-per-cent chance of winning. (i.e., “Germany will probably not win.”)

On the other hand, that claim looks at least a little like investment advice. The UBS pamphlet is pretty specific about this: it has a sidebar headed “How we’ve applied the insights from our investment process to the prediction of football matches.” If you believed that Germany had one chance in four to win the tournament, you could have bet them to win before han data price of more like five to one. That’s a decent return. Or, er, it would have been.

UBS’s bad luck has provoked some catcalling from the financier-philosopher Nassim Nicholas Taleb, who likes to make sport of model-builders in the investment world who make assumption-filled forecasts without having “skin in the game.” So it’s worth remembering that the betting markets, an embodiment of skin-in-the-game, liked Germany almost as much, at 5-1, as the jerks from the bank did.

Still, Taleb is probably on to something. Followers of quantitative sports analysis are still hooting over stats guru Nate Silver’s notorious forecast of the 2014 World Cup, which used similar methods and gave Brazil a whopping 45-per-cent pretournament chance of victory. Silver emphasized strongly that this number was investment advice: he published his 45-per-cent figure right alongside the betting-market price for Brazil, like a dare.

The UBS numbers, I think, hint at the nature of the problem. Its model gave Germany a 90-per-cent-plus chance of emerging from Group F and reaching the knockout stage of the 2018 World Cup. Even though it was the thing that had a 10-per-cent chance of happening that happened, this figure may not have been terribly unrealistic. What looks more troublesome, even in retrospect, are the numbers for Germany in the knockout rounds. UBS had made Germany a 3-1 favourite in the Round of 16, conditional on surviving the group: it then had Germany a 3-1 favourite, again, in each of the quarter-finals and semifinals, and suggested that if Germany made the World Cup Final, it would still be a 2-1 favourite in that game.

Is it credible that Germany, given that it reached the final, would be that much better than the other finalist, whoever that ended up being? I suspect this is a form of double-counting of unreliable pre-tournament information, and would suggest that almost no team should ever be given anything like a 24-per-cent initial chance of winning any World Cup. It was apparent minutes into Germany’s first group-stage game against Mexico that the national side was not its old, bold self: German commentators had already started devising quips about “the end of an era” at the half. (It is funny how often eras end at World Cups. Curiously, it seems to coincide with the reigning champion’s elimination.)

Surely the quick obsolescence of even a deep betting market’s initial estimates is a hint that pre-tournament “priors” based on recent international form should have only a feeble influence on a model of an entire tournament. High-stakes international soccer fixtures just don’t happen often enough to give trustworthy probabilistic priors, especially since the World Cup is packed into a few weeks, and teams can be bushwhacked by private quarrels or injury or political distractions or bad grub.

But modellers like UBS and Silver need strong priors, or else their pre-tournament advice would be, well, too boring to attract anyone’s attention. “Yeah, lots of teams can win this thing.” You couldn’t put that in a pamphlet and have anyone read it. And even for a bettor exploiting relative longshots, the wait for a payoff over a “long run” of quadrennial tournaments might be measured in lifetimes.