A key feature of the markets Walsh and his partners invest in is that they are characterised by randomly “independent events”. Specifically, the occurrence of one event does not influence the chance of the other and they therefore have “finite variance”, or limited downside risk.
“Gambling has the huge benefit of having independent events – I cannot get blown up by the black swans that plague financial markets.”
He says deploying mathematics in “equities markets that may have infinite variance outcomes makes working out probabilities much harder”.
“You don’t know whether you are summing a sequence of fractions that add to one or if they add to infinity, because financial markets have non-independent [or potentially related] events,” he says.
The bankable independence of results in gambling markets is the “component of our strategy that gives me the most security”, Walsh says.
“It is even better in games like black jack, where the events are not only independent but also negatively correlated – your chance of winning goes up if you lost the previous hand because there are an excess of cards remaining that are advantageous to you.”
He is critical of the billionaires printed in financial markets who “often make money in the low-probability, high-opportunity outcomes that are essentially exhibiting ‘correlated parlays’ [where one event significantly influences the probability of another].
“Correlated parlays make people look smart and can create a whole bunch of rich folks, but there was probably nothing but pathologies in the financial data.”
From a profile of David Walsh, a guy who made a fortune using quantitative models to gamble professionally. Walsh seems to have a healthy appreciation for the role luck has played in his success and is redeploying much of his fortune to build an eccentric art museum in Tasmania.
It's his explanation of his success that is worth studying.
Asked about exactly what his team’s “edge” has been over the years, Walsh distils [sic] it down to embracing the wisdom of crowds.
“You can work out some complex algorithm to predict horse racing odds using multinominal logistic regression,” Walsh says. “But the result would significantly underperform the public odds.
“The key is that the public odds must be included in your model. The best models are not predictive models per se, but ‘perturbation’ models that start with the assumption that the public is right and then work out what small errors they might make.
“The public odds are not just an important signal – they are a remarkably efficient signal.”
He cites the example of the former Russian chess grandmaster Boris Spassky, who played and lost to the Russian public in a game of chess.
He describes the public’s ability to make accurate collective decisions as an “emergent strategy”, like birds flocking or democracies (which form not in the mind of one individual, but through the interactions of many).
“I am saying there is wisdom in crowds beyond the point you can model without explicitly incorporating it.”
How does this system work in practice? “Let’s talk about Sydney race night on Saturday,” Walsh explains.
“We might have a model of what we think the probabilities should be that includes the public odds.
“We essentially wager on those events that have better chances than the public thinks, which gives us a positive return expectation.”