Back Testing Tennis Betting System in 5,000 Matches

Back Testing Tennis Betting System in 5,000 Matches

Back in 2009 I had developed a tennis betting system based on rankings. I had used a mathematical formula I found in a publication of 2002. The calculations weren’t difficult and back testing the system using a sample of 600 tennis matches in an Excel spreadsheet proved worthwhile. Apparently though, I must have done that back testing in a bit of a hurry, as there were a couple of mistakes in the conclusions and graphs. I was reminded of that tennis betting system by John S. who sent me an email, wondering why I was worried of placing lay bets at big odds and didn’t omit those specific bets. So, I revisited the betting system this morning finding out that the system is profitable only if bettors lay at short odds – like a bookmaker does – or bet at bigger than average odds. No surprise there I suppose!

tennis-ranking-formulaThe formula that Klaassen and Magnus included in their publication and was the basis of my tennis betting system uses the tennis ranking system and introduces a constant, λ, which is 0.3986 in men’s singles and 0.715 in women’s singles. The spreadsheet I created calculated the probability of each tennis player to win each game and depending on the average odds found in the database, it was able to prompt the value bets I needed to risk my money on. Initially though, my back testing conclusion was that the system was profitable only if bettors bet against those value bets. In other words trying to turn a losing system into a winning one, since if one bet on the value bets, they would lose in the long run.

In that old post, I have published a rather impressive graph towards the article’s end, giving high hopes of the system’s profitability. On the other hand, due to having developed the system quickly, I assumed layers would have a steady profit target, risking different amounts per bet. While they stood to win 10 euros in each match, if they laid at 10.0 odds, they would risk 100 euros! That contradicts the flat betting system I always recommend of when back testing a new idea in gambling. We should always aim to risk the same amount of amount, instead of targeting the same profit out of each bet.

Today I spent some time on that excel spreadsheet, corrected the mistakes and found out that if one risks 10 units per lay bet, they will make some money in women’s matches but will break even in men’s matches. I used once again the sample of that old spreadsheet, meaning I had 600 Australian Open games to work with. I noticed though that the tennis betting system performed much better in WTA matches and that led me to download all 2010 and 2011 WTA games’ details. The database now consists of 5,000 women’s single matches taken place all over the world during 2010 and 2011. The results were astonishing as the forecasting formula calculated value bets of 6.1% +EV!


Ok, been there, done that a million times and by the end of the day I always find a minor detail I have skipped or a small calculations’ mistake I didn’t pay attention of. This time the betting system was profitable due to laying at short odds. The tennis database of includes the average odds of popular bookmakers. Therefore I had been placing lay bets at bookmakers’ short odds, comparing to a betting exchange. Although I included 5% commission in the calculations, the profit was still quite exceptional. Since I wouldn’t be able to lay at those odds, I had to try and back the opponent’s odds and find out how the system would perform.


Obviously the results were disappointing, as I was now backing at somewhat short odds, again comparing to a betting exchange. I was now losing 6% of my money in each bet, resulting to a loss equal with the profit I made earlier. Some can say that the mirrored graphs prove the market’s efficiency in tennis matches. If we laid at bookmakers’ average odds we would win but would break even at a bit higher odds. If we backed at bookmakers’ average odds we would lose, but again break even at a bit higher odds. The market seems to be efficient enough and drive the odds close to the true probability. From that point on, bookmakers adding their vigorish eliminate much of the value punters could find in a zero-sum game. At least that is the case with the specific tennis system.

Maybe I will spend some more time comparing bookmakers’ odds with Betfair’s ones, but I don’t expect the system to have a better ROI of 3% or so in the best scenario. Unless the betting exchange’s odds are much closer to the ones of bookmakers, but I highly doubt it.

Get my Posts in Your Email For Free!

I will deliver my posts straight to your inbox as soon as they're published.

  • Lewis

    Hi Jim,

    This Luis  from Portugal. It´s a pity that people not comment your amazing posts.

    As always, very useful material!!!

    Talking about the lay odds. So, as I understand you can put lay odds on favorite or in the dog.

    So, tell me if your analysis is somethig like this. For example.

    Today we have in France( marselle)

    Nikolay Davydenko – Andreas Beck

    By Klaassen formula you have 62% of chance to Davidenko win

    Max odds in bookamkers = 1.2

    So, we have in bookamers opinion 83% of of chance to Davidenko win.

    My question is:

    a)Do you consider that this could be some value, and in this case you LAY Davidenco, correct?

    b) Do you join some other kind of criteria to chosse if you lay the favorite or the dog?



    • Jim Makos

      Hi Luis, thank you for your kind words! Nice to see you commenting on my post!

      Usually I look for value bets, so I wouldn’t back or lay Davydenko. Value betting theory says that if his odds multiplied by his win-probability percentage resulted to a number greater than 1, then we should back Davydenko. Therefore neither lay nor back in your example. Perhaps you should calculate Beck’s win probability and see if that’s a value bet. However, in case it’s a value back bet, my analysis showed that you’ll lose money if you actually back such a value bet. Therefore I inverted the bets and I would lay that bet, although it doesn’t make sense to bet against value bets!! That’s how the first graph is created, but I strongly believe it’s due to the short odds.

      • Lewis

        Ok Jim, I see!a) So, is it correct to say that in tennis and in this system, you achieve the player  win-probability percentage from Klaassen formula? Coud you give an ideia in terms of %, in the 5000 games that you had refered (WTA), how many had been value bets? ( considering the value betting theory)b) By the way, in soccer mathes do you have any «Klaassen formula»:)) to achive the win-probability percentage of a team?Regards from PortugalLuis

        • Lewis

          Jim, sorry to bother you again:)) Howver, now I confused with the following:

          You said about the laying system:

          «This time the betting system was profitable due to laying at short odds».

          However, we know that usually we find value bets in higher odds, because the bookies generally have poor prices for the favourites. 

          Note: Actually is easy to understand that lay at short odds it coud be a valuable strategy.
          The confusion is that you said that you can find value bets, and the value bets, generaly have higher odds.


          • Lewis

            Ok, Jim, we had talking about diferent things. Now, I supose I understood your idea:

            «If we laid at bookmakers’ average odds we would win but would break even at a bit higher odds. If we backed at bookmakers’ average odds we would lose, but again break even at a bit higher odds»

            Anyway, and talking about my idea, I supose that the major of your value bets are in the dogs..!I think that in arbitrage betting you can experience that betting in higher odds is always a 
            good business in the long term!

            In the other hand, I have the perception, ( I don´t test it) that laying  the favorites in tennis and trading them,could be a good strategie as well, if you define the racio about 1/4.

          • Jim Makos

            Betting on underdogs have been considered a good strategy in general. When I was testing Poisson distribution, most of the bets were on the underdogs – although it was still a losing strategy. As far as laying the tennis favorites concerns, I can’t comment without testing it first! 🙂

          • Jim Makos

            If the win prob is 95% and the odds are 1.1, that is still a value bet, although the odds are surely short. However, backing those selections will result to a loss, so I inverted the system and transformed the backing bets into laying bets, leading to lay the initial “value bets”.

        • Jim Makos

          a) 3893 value bets out of 4903 b) I haven’t found one.

          • Luis

            Many thanks Jim. Invaluable information to all that want make some tests.

            So, 79% of value bets. Any ideia in this sample how many value bets between this odds:  1.15 and 1.50?

            Not to bother in the next weeks:))!


        • Master G

          Here is a rating system for soccer: But I have not tested it yet.

  • Tony

    hi jim
    i’m a daily follower of your newsletter,great stuff!
    nowadays we can lay almost evry tennis fav at bookies odds or lower in-play due to the overreacting traders,sometimes after 1 or 2 points scored!

  • anonymous


    Great post 🙂

    There are quite a few opportunities in sports when the bookmaker odds for a moneyline selection are equal to the betfair lay odds. Would you recommend laying at this price on betfair?


  • varunsriram1

    Quick question – why do you always include a 5% commission? Many sites, such as Bovada, Bookmaker, and Pinnacle charge no commission for tennis bets. Great read btw 🙂

    • At the time of writing, I was betting exclusively on Betfair, a betting exchange that charges 5% commission on the net result of each betting market instead of adding vigorish. Thus, I needed to take that into account in my calculations. Thank you for commenting.

  • Suppi1234

    Hey Jim,
    how did you do the backtesting? I would like to test it with the 2015/16 data but am not able to create an excel spreadsheet with the needed calculations and the graphs. I already downloaded your excel spreadsheet provided and did some manual testing myself. Would love to know how you did it.