Make Money Opposing Trading Bots

The title of this post is taken from a search string that brought a reader to this blog. It's an interesting question to ask of Google. How do we answer this query? Well, we need to know when we see a bot and to determine what it is up to.

Often you will see bots on a horse race order book as soon as it appears on Betfair. The stakes will be low, about £2 or thereabouts and odds will be very conservative. Either very low or very high. As soon as someone makes a better offer the bot will jump ahead in the odds with an even better offer and so on, until the bot's offer is taken.

Obviously these bots and their low stakes are programmed by a novice bot trader with no real understanding of betting markets. The bot will either be offering to lay at very low odds or wanting to back at very high odds. The bot has no idea of fair odds for the horse in question, it just wants to get matched as soon as possible before the spread closes.

As soon as the odds are matched the bot will then try and reverse the position on the other side before the spread closes. If the bot owner is lucky the bot will find someone stupid enough to take the odds whilst the book is at over 200%.

These simple bots will never make the owner wealthy because the average bettor is not so naive to risk large amounts of money whilst the overround is so big. Also, there is no point trying to outwit one of these bots when it is offering peanuts. A more sensibly coded bot is one that does repetitive or simple tasks for a knowledgeable player. And then, you may as well treat the bot as though it were human because it probably isn't going to make a serious mistake.

Decrease Your Bet Size After a Loss

If there is anything that Kelly Criterion tells you about your wagering it is this, you must decrease the size of your next bet after a loss and only increase it if you win.

In other words, chase your winnings but never chase your losses.

For many gamblers (and I mean gamblers rather than astute investors) a losing bet sends them into a panic and certain ruin. They start chasing their losses with ever increasing bets in a vain attempt to recoup their losses.

If your last bet was a losing one then the first thing that should come to mind is, "Is the system still working?" In other words is the losing bet percentage in line with the original system. Is it an expected run of bad luck? Or is the system simply a bad one that must be discarded?

Kelly Criterion tells you that you must bet as a function of your chances of success and the size of your betting bank. That is the only way to maximise the growth of your bank.

If your last bet was a losing one, your bank is now smaller, therefore your next bet will be smaller too. Don't get trapped in the fallacy of martingales.

Place Pricing in Horse Racing

There have been many previous studies that focus on different approaches to the problem of pricing a horse to place rather than to win. If we assume that the win market is efficient then at the start of a race the last traded win price for any horse represents the true probability of that horse winning.

Harville (1973) assumed that any horse that won is automatically discounted from placing and the win probabilities of the remaining horses recalibrated to sum to 1. This naive Bayesian approach suffers from various problems.

1) It is rare for a race to finish in order of win probabilities. In other words, the favourite does not always win and the second favourite does not always finish second etc.

2) The "Silky Sullivan" effect. Some horses either win or finish well off the pace. One such horse was Silky Sullivan that sat at the back of the race until the final furlong and would make a spirited attempt to win (or fail miserably).

3) Trainer orders. A jockey will be under orders from the trainer, which cannot be modeled mathematically. A jockey may be told to go all out for the best placing possible. Another jockey may be told to strike out for the win but canter in if the attempt fails. Yet another jockey may be told to treat the race as a training run. And so on.

Further studies by Henery (1981) and Stern (1990/2008) have sought to model the fact that favourites tend to place less than their place odds say they should and long-shots tend to place more often than their place odds suggest.

A comparative study of all three approaches has been carried out by Lo (2008) and demonstrated the superiority of the Henery and Stern models over the Harville model. However, these superior models are computationally taxing and Lo demonstrates a simplified approximation that is easily coded.

Further Reading

 

A Study of Betfair Place Odds - An article demonstrating the futility of using place pricing models to discover pricing errors on the Betfair place market.

References

Harville, D.A. (1973) Assigning probabilities to the outcomes of multi-entry competitions. Journal of the American Statistical Association, 68, pp. 312-316

Henery, R.J. (1981) Permutation probabilities as models for horse race Journal of the Royal Statistical Society, B 43, pp. 86-91

Lo, V.S.Y. & Bacon-Shone, J. (2008) Approximating the Ordering Probabilities of Multi-entry Competitions by a Simple Method. Handbook of Sports and Lottery Markets, pp. 51-65. Elsevier

Stern, H.S. (2008) Estimating the Probabilities of the Outcomes of a Horse Race (Alternatives to the Harville Formulas). Efficiency of Racetrack Betting Markets. Hausch, Lo & Ziemba. pp. 225-235.

Stern, H.S. (1990) Models for Distributions on Permutations. Journal of the American Statistical Association 85, No. 410 (June) : pp. 558-564