Introduction And Season Recap‏ 1,276 views


this post will be both an introduction and a whole season recap for me! i joined this site a little more than a month ago but I only bet on regular season nhl at the moment so my season has now officially ended.

First a little bit about me and my betting method. I am a statistician and I did my master's thesis on predicting stock volatility, then moved on to a job where my main task was building prediction models. Based on that experience I have now been building a statistical prediction model for nhl for about 2 years now and I had been both following and betting on nhl before that too. Most handicappers obviously don't want to share too much about their methods so I really have no idea how many successful handicappers use an actual prediction model, how many just derive some estimates from power rankings and such, and how many use completely different systems, but I decided to share how I come to make a betting decision. I am very good at modelling and data mining but a not a good handicapper and this season I found out that most of the manual corrections I make on the model recommended matches end up costing me money. So at this moment I am completely model reliant and I'd say about 90% of all my betting decisions are based completely on the model. Oh sure, there is a long long list of manual checks I do for each team but they just don't add enough value to my bets. This is also why you will very rarely get an extensive match analysis from me as the answer for my betting decision will almost always be the same, ie. according to model predictions and historical data mining this and this particular team at these odds look profitable to me.

So how did my model do in the first full season I used it. Presenting the results isn't as clear cut as one might think because I obviously learned much from actually making betting decisions based on the model, so my approach and methods changed during the season. There's only so much you can prepare for with just backtesting previous seasons, but that is still how I started. I had made several indicators before the start of the season while backtesting previous seasons. In the beginning I mostly followed those indicators and the results at first were ridiculously good. From the beginning of the season to March the 20th the model had recommended 125 bets with 82 wins and 24,4 ROI (had I put 1 unit on each bet).

I was thrilled about the results and just kept following the model but at that time I hit a rough few weeks. I still believed in the indicators because they had worked so well previously and figured it was mostly just bad luck. But when the bad results continued I started digging in deeper and found that the indicators I had made really weren't that good to begin with and that I could easily make them much tighter and better. I really have no explanation on why they worked so tremendously at the beginning of the season but from March 21st to the end of the season the pre-made indicators recommended 155 bets with 79 wins (51%) and -5,1% ROI. For the pre-made indicators this means that the results for the whole season were as follows: 280 recommendations, 161 wins (57,5%), 8,06% ROI

But because the indicators started to fail, I made the whole system much tighter and basically abandoned the idea of a predetermined set of indicators and started using a completely different approach which I have been using it for the last month on pickmonitor with some nice results. Combined with the early good results the season as a whole proved to be a rather profitable one.

So that's my season recap! Hopefully it will be interesting to someone and maybe encourage people to follow me next season when there are a lot more regular season games to bet on. I'm really looking forward to it because even with very suboptimal indicators my model provided really nice results and as I said, lately I have been able to make the model even better. Naturally i dont expect to keep having 17% ROI but I do have high hopes for at least moderate profitability. I will also try to add something to the nhl discussions here come next season as this site looks like a really good place for a handicapper compared to what a lot of the other sites offer.

Oh, and it's also worth mentioning that I'm from Europe so I very rarely get to watch any of the games (although I'd want to) and this also means that I wont have any information that gets out just before the matches and I also usually can't bet on closing lines and such

Edited 4/28/13 at 2:13AM by lunatus - No reason listed.


Hi Lunatus,

Welcome to the forum and congrats on a solid season 61.32% is very nice. One thing I think that might have hurt your system is that with a condenced season 48 vs. 82 games and 4/5 nights 3/4 nights with travel the system players don't really factor that into play and where your slide might have happened could have been due to the human aspect that computers and numbers don't and really can't factor.

In those 4 games in 5 nights thier is fatigue, nagging injury quote unquote ( healthly scrathes) , jet lag and lack of effort. If you on your 4 game in 5 nights on the laod your goalie lets in a softy but you have say 2-0-1 so 5 points in those 3 previous gmaes mentally teams could mail it in and then a 1-0 games turns into a 5-1 game. The is also alot of resting for the better teams at the end of the season. Perfect example was Chicago they clinched really early so you never knew 100% who would get alot of ice time. I usually layoff the 1st and last month of the NBA season for these very reasons of rest and motivation 2 things that cannot be measured but digits.

I look foward to future discussions.




thanks for the reply!

you are correct with the shortened season. I would have much preferred to use the model on a normal season as that is how all the parameters were estimated. I usually use the newest seasons as my population for estimating the parameters, but now I will have to do some research before adding this shortened season to the mix as I don't know how well it compares with the average normal-length season.

That is actually a good topic for another discussion because the quality of the model is so dependent on the base population one uses but I'll leave that for later since its midnight already.

I was also supposed to stop betting a while ago because of the second reason you mentioned but just couldn't help myself and kept betting because things were going so well. Though overall I have found out that the indicators I use seem to work about as well at the very end of the season as they do before, but this is one of the things I'm going to research more during the summer as I'm making better indicators anyway. Of course the very last few days or so may very well be an exception and for example I had no way of predicting the Chi vs ST. Louis game as Chicago sat so many of their best players.

At the start of the season I don't do any bets before all teams have played more than 10 games and preferably even about 20 since the variables I use in the model need some time to adjust.


Please Login or Register to reply.

Ready to start winning?

Yes, Sign me up now No, show me the handicappers first