GeneralPosted by Tobias Bramhed Sun, August 21, 2016 20:49:09
From now on, this blog is located at:www.machinebetting.com
Follow me there!
Betting resultPosted by Tobias Bramhed Tue, July 12, 2016 22:02:37
Finally the Euro Championship is over - The Swedish team did not deliver to good. I hoped for more but did not really expect anything. It is now time for players like Zlatan Ibrahimovic, Kim Källström and Andreas Isaksson to leave the national team and the process starts to build a new and younger team. It feels good to get a new start...
The results are now updated with June data:
It was a good month, although the low number of bets (486) the yield landed at 3.4% which gave me 6266 SEK into my account (about 661 Euro).
As I mentioned in last post I made some adjustments on the 12:th of may, and they seem to deliver.
The results, by bet type, with data from 12 of May to 10 of July looks like:
I am especially glad that backing Away and Under now is in profit - Something I have struggled with for a long time. I am still using a smaller bet size for those types, I will upsize the bets gradually over the nearest half year IF the good results lasts.
Betting resultPosted by Tobias Bramhed Tue, June 14, 2016 21:22:04
Finally the Euro Championship 2016 has started, yesterday Sweden managed to get 1-1 against Ireland BY PURE LUCK. A big Disappointment, I really hope that the Swedish team can get it together for the match against Italy on Friday.
Now for the results updated with May:
984 bets executed, and the model turned the bad trend from March and April into a small profit of 1012 SEK (equivalent to 0.3 % ROI).
Breaking it down in back type:
The away and under betting is still in loosing figures, but when examining the effects of changes made in 12:th of may:
We see positive ROI on all types, even if the sample size is quite small I embrace the positive vibes :)
AnalysisPosted by Tobias Bramhed Thu, May 26, 2016 22:43:01
Did a little data exercise to see how good the pregame market has been to nail the correct probability for home/draws/away wins. The data consist of 91515 soccer matches played in 2014,2015 and 2016.
I calculated the implied probabilities from the pregame odds (1/odds), and distributed the over-round evenly over the three scenarios. Then I summed them together with the actual wins.
This first table shows that the market hade the most problems with the Away predictions. Market predicted the Away teams to win corresponding to 29027 times, but they actually only won 28192 times (which is a diff of 2.9%).
I dug a little deeper to see how the situation is when dividing by pregame favorite:
Now I start to identify some areas which the market seems to have bigger problems with... such as the probability for an Away win when the pregame favorite is a Home win.
This could be a starting point when designing a new strategy. Find a nische where the market has problems, and try to be the best in that nische :) Now I will continue to mine my data, and see if I can find a new nische area.
AnalysisPosted by Tobias Bramhed Tue, May 17, 2016 22:15:41
In my models I use different type of classifications for both teams and leagues. One interesting way to look at leagues are how interesting they seem to be to Betfair costumers. One way to check this is by calculating the average matched sum per event and match.
The average is calculated from 2015 and 2016 data, and the top 25 leagues/events are:
The average matched is in SEK. The Premier League stands out as the number one preferred betting markets with an average matched 44.5 MSEK per match on the Match Odds market. That is more than twice the amount in Primera Division and four times Serie A!
The Community Shield is the match between Arsenal and Chelsea held on 2.august 2015 (won by Arsenal with 1-0 by the way).
Especially the Scandinavian markets are interesting for me, and the top3 there are:
Betting resultPosted by Tobias Bramhed Tue, May 17, 2016 21:08:48
And these are the betting results with april included:
There was a massive raise in number of bets (1351), unfortunately the ROI was -0.9% which ended up with a loss if 4850 SEK.
Two months with negative results made me make some bigger adjustments in my model, implemented from 16 may. I hope to see some improvements during the next months :)
When breaking down the result on back type:
Based on those figures I have left Home and Draw unchanged, and focused on improving Away and Under.
A last note, if you compare number of bets with previous shown results you will see that they don't match. This is due to an error counting number of unique bets.
Betting resultPosted by Tobias Bramhed Mon, April 11, 2016 09:49:16
And here are the results for March :
Oh no! March ended up with a loss of 1460 SEK ... So summing up the first quarter of 2016 we see:
1. total turn of 1 264 598 SEK, that's approximately 136 406 Euros.
2. 2 751 unike markets were bet on, out of 55 499 followed. That is a hitrate of 5 %.
3. ROI of 0.8 %, and normalised ROI at 1.5%.
The turnover is inline with my expectations, but the ROI of 0.8 % is too low (my target is 1.5%).
breaking it down on back type:
Its especially when backing away teams that the model struggles... So for the 2:nd quarter I have made som adjustments to that model.
I have also med a general modification to the betting rules, I now demand the market to be a little bit more effective than I did earlier. There seem to be some bad bets originating from (probably) slow score updating. I think this adjustment will partly handle that problem.
The difference between ROI and normalised ROI is something that I will need to do some more thinking about :) There are two issues when following the ROI:
1. How much of my requested amount is actually matched?
If I request 1000 SEK and only get matched for 100 SEK - That will skew up the evaluation of the model.
2. How much do I request?
Depending on market type I request either 2% or 3% of my portfolio. That means that a later loss (given that the model is long term profitable) induces a higher blow to the ROI than a loss in the start.
So evaluating model in terms of normalised ROI its doing fine....
Just trying to put some data together (I will follow up on this subject later on!):
Betting resultPosted by Tobias Bramhed Tue, March 15, 2016 21:03:43
And here are the results from betting in Februar:
This month generated approximately the same amount of bets, but as my portfolio is bigger I had an increase in average bet size and therefor also the total amount bet (435' vs 322' in Januar).
I managed to make a profit for the second month in a row in 2016 (!). The ROI ended up as 0.4 %, so total for the year is a ROI of 1.5% (from my previous post we can draw the conclusion that the strategy now is performing according to its expectation regarding ROI.... ).
Breaking this years result on type of bets:
We see that backing away teams has had a hard time... i will eventually start to look for some way to twerk that strategy into profit.