Analysis of the TennisRatings ATP Model in 2014

I've had a number of enquiries about how my TennisRatings Model performs for recommending pre-match bets - the 'red' value price in the daily spreadsheets.


Model performance in 2014 in all ATP matches: 0.12% ROI
Model performance in 2014 excluding 1st round ATP matches: 3.48% ROI

Model performance in 2014 in all ATP Challenger matches: 1.07% ROI
Model performance in 2014 excluding 1st round Challenger matches: 3.91% ROI

Value Players priced 1.01-2.99 performed better than those priced 3.00+

At this point it's important to make the following points.  

This 'red' value price on the daily spreadsheets is a model price which is then adjusted for some match-up factors, which include but are not restricted to: 

Known fitness issues
Head to head records
Travel/scheduling issues
Ranking discrepancy

This price is a purely statistical evaluation with no subjective factors taken into account.  The above - and other - factors have been backtested to allow a statistical adjustment, as opposed to guesswork.

This is just one area of the daily spreadsheets.  Subscribers will be well aware of the wealth of in-play information that the spreadsheets also provide, enabling various in-play angles to be taken advantage of.  

For extra clarity, Challenger Daily Spreadsheets are purely based on my model price with no situational/match-up adjustments.

Analysing this was a huge undertaking but I felt it was vital for transparency and with this extra demand, it was also pressing to get this done prior to the 2015 season.  Due to the amount of time needed for this analysis, I have yet to assess the WTA but this will be done in the near future.

Assessing the whole season enabled me to obtain a very strong sample size which derived the following overall results:-

1004 bets
£149.38 profit based on prices at time of the spreadsheet release daily, based on a stake to give a level £100 profit
0.12% ROI

These figures are, on the face of it, pretty mediocre, but I did a great deal of further analysis to delve further into the database to generate further information:-

Starting Price Matches Stake Profit ROI
<1.50 150 58039.95 395.82 0.68
1.50-1.99 263 38331.19 714.13 1.86
2.00-2.99 310 23220.74 776.73 3.34
3.00+ 281 8450.95 -1737 -20.56

This information is obviously very useful.  A positive return on investment of 1.58% was generated from starting price range 1.01 to 2.99 (723 bets) with slight underdogs performing the best.  Heavy underdogs performed atrociously and skewed the sample significantly.  Throughout the sample I noticed these players losing 2-1 on a very regular basis, so this may be a large variance issue.   It would appear that backing these players +1.5 sets would be a better strategy than for the outright match win.

I then took a further step.  

Major US-based syndicates, who also use statistical models extensively, maintain that first round matches are to be avoided and that statistical models perform very badly in these matches.  This is logical and understandable, given that there may be some 'false' value on players who turn up out of condition/motivation in first round matches, when it is impossible to quantify these factors.

I decided to filter out all first round matches and this generated the following results:-

Starting Price Matches Stake Profit ROI
<1.50 77 29538.43 802.44 2.72
1.50-1.99 133 19302.05 990.19 5.13
2.00-2.99 176 13047.63 665.98 5.10
3.00+ 169 4789.35 -139.3 -2.91

A startling diffference!  Heavy underdogs, whilst still negative, were much improved, as was all the other price ranges.

Overall, this produced the following overall statistics:-

555 bets
£2319.27 profit based on prices at time of the spreadsheet release daily, based on a stake to give a level £100 profit
3.48% ROI

Clearly these are much better figures and this ROI is very strong indeed.  It is very hard to get much more than 3% ROI over a big sample in pre-match Tennis betting.  Not only this, when just 1.01-2.99 starting prices were taken into account, profit of £2458.57 was generated giving 3.97% ROI.

On this basis, filtering out first round matches from a pre-match betting perspective is an absolute must, and doing so gives the daily spreadsheets a very strong pre-match return overall.

The ATP Challenger Spreadsheets also produced very similar comparisons.

All matches:-

805 matches
£1031.37 profit based on prices at time of the spreadsheet release daily, based on a stake to give a level £100 profit
1.07% ROI

Filtering out 1st round matches:-

460 matches
£2178.68 profit based on prices at time of the spreadsheet release daily, based on a stake to give a level £100 profit
3.91% ROI

Again, first round matches were negative, and matches in the second round onwards were strongly positive, with a similar ROI to the ATP.  

It is clear that from decent samples, the daily ATP & Challenger spreadsheets performed very well in non-first round matches, and the data backs up the US-based syndicates assertions that first round matches are a minefield and should be avoided by statistical modelers, who can gain a strong edge in subsequent rounds.

It may well be that the bookmakers can get better inside info regarding player condition for first round matches, but after a player has won their first round clash, the TennisRatings model has a clear edge over the market with less fitness concerns on players.