Strategy and multiverses

The way sports betting is normally supposed to work is that you take your hard earned money and put it into the pockets of the bookmakers. That’s pretty easy to do; the odds are constructed to make sure there’as a strong and steady net flow of money from punters to bookmakers. The punters win a bit, but on average, the flow is always one way. Its a fool’s game.

The aim of “sports investing” is to make the money flow in the other direction. To take their money, and put it in your pocket. And that’s not so easy. It’s like swimming against the tide. So how can it be done? There are two man methods for achieving this (without resorting to rigging games). The first is a kind of arbitrage, where you find that different bookmakers are offering different odds on the same game, and you manage to place bets to cover every possibility, such that you win a little whatever happens. The second is to find and bet on cases where the bookmakers are offering odds that are better than the “true odds”, meaning that they believe the chance of a result is lower than it actually is. This second option is what we do,

But what of these “true odds”. To understand the concept fully, we need to consider the concept of multiverses. Imagine that immediately you place a bet, we create 10,000 identical copies of our world. Time moves forward, and in each world, things happen slightly differently. The teams are identical, but the game plays out differently in every world. Different numbers of goals, some wins, some losses some draws. If we count up the times the outcome we backed comes up, and the number of times it didn’t, we can easily calculate the true odds.

Unfortunately, if multiverses exist, they’re certainly not something we can just peer into. So, we have to make other arrangements. In our case, we try to estimate a number of data points for each team, and then run a large number of “multiverse simulations” to estimate the true odds. The downside, unfortunately, is that we can never be sure if our “true odds” were right, because we can only see the actual results of one game. We can only infer whether we are doing it right by whether, on average, we are making the profit we hoped for.

So that’s a quick overview of what we are trying to do. Surprisingly, the multiverse simulation is actually a fairly straightforward thing to achieve, with a bit of computer programming called a Monte Carlo model. Estimating the parameters that feed into it however, is much, much harder. That’s where the mathematical and statistical AI dark arts come in!