MILES Model Selections

This model (named after the generic answer you get when you ask someone how much better an avreage Premier League side is than an average League One side) is built on the fundamental principles of the Elo system popularised by chess, but has added tweaks and extras to apply it to the markets. As well as giving each team/player/horse etc an actual Elo rating, which goes up and down based on the strength of opponent they win or lose against over time, it also derives a theoretical rating based on a combination of opening and closing lines (and by proxy the movement between those two sets of odds). These two sets of ratings are then weighted together (the weights dependent on things like sport, sample size and recent unusual line movements) to give a final MILES value that can then be used to compute future prices. Crucially, MILES can be updated once a fixture or race has been priced too to take into account new information and in essence try to 'predict' where the lines may close. Ultimately, CLV is generally a very good barometer of performance in most liquid markets, so using a model like this should in theory help to derive this even on events with very little line movement

MILES isn't just designed to calculate win prices however, I have tried to build it in a way so that it can calculate all manner of derivative markets, based on the same methods as above. In the main, whilst finishing it off I have been using it for spreads and totals across most team sports + tennis and darts, but I've also begun to build in areas such as player props, BTTS, corners and half markets. For racing (horses, dogs and motorsport) it can generate fair prices for both win and places as well as several other niche markets like matchups and F1 retirements. Essentially, if there's enough pricing and results history for a certain side market, MILES will be able to estimate a value for it and this prediction will only improve as the sample increases. Currently, my results data for football goes back to 1892, with odds data back to 2013/14. Other sports have smaller samples but growing this set of backtest data is on my checklist for the next set of improvements to MILES.

The nature of it is to be a high volume, low ROI type of model, trying to find sustainable edges in liquid day of game markets and slowly price better than the market over time as it uses the bookmaker's own historical prices against them.