AI Powered Sports Betting
Why ask about sports when there are algorithms that work in financial markets and all sorts of other predictions, that go back farther than sports betting and have larger amounts of dollars at stake? Value, momentum, carry, quality, overreaction; these are near-universal guides. There are a number of reasons why these techniques are not more widely recognized. One is that they deliver only thin edges that take discipline to exploit in useful fashion.
And they can go years without working, which is why you need a portfolio of them for consistent profits. The first is that you can beat the odds in sports betting, investment, business, politics, war or anything else by intuition and commonsense without careful quantitative analysis. As an example of the latter, when Ed Thorp published a winning strategy for Blackjack in , many mathematicians—mathematicians! As examples of the former, talk to anyone or read any opinion column. When people look back on in the distant future, what is the biggest thing they will be unable to understand about our society?
I think the greatest mystery will be that we allocated power accorded to claimed ability to manage risk—either by predicting outcomes or by dealing with events as they arose—and we never, ever thought of checking those claims. People run for office and win by making all kinds of confident claims about what will happen, what should be done, have have worse than random track records if you check their predictions.
People run companies because they have loud confident voices and are never uncertain. The world is not run by subtle thinkers who weigh evidence carefully and make tentative-better-than-random decisions. And the people in charge like it that way and do whatever they can to discourage the idea that anyone can do better than they can. The one crime that always leads to death is trying to predict the future in any way but though God telling you.
Like models in the financial markets, models in the betting markets are a sort of arms race. A lot of sports betting markets not major football or US sports but a lot of sports that are larger than one would think start out with lines produced from very simple models, with further line movement not coming from models but instead injury news or public action.
The first person to come in with a respectable model in such a sport will make windfall profits. They will often wager a large percentage of the games on offer, find bets with double digit edges, and because the line is so far off they can wager relatively large, as even paying extra vig or re-betting they still have a large edge.
What always happens is this type of play - wagering big, moving the line heavily, and winning, causes other players as well as the opening line-makers to take notice of what kinds of games tend to generate these line moves and the odds will shorten ahead of time accordingly. However, this is a slow process. The modeler will lose a little value to these defensive mesaures, but they will still be way ahead of the game. The real problem comes when other modelers build models similar to what the first modeler built and begin taking value out of the market.
The original modeler will get less value on their plays. Furthermore, the closing line of the game will move closer to efficiency, since there is now additional action on the game, both in the right direction. This tends to flow back to opening line-makers who then begin to set odds more correctly The only way to get back to profits from that cycle is to build a better model than what is out there, but there are limits to how strong a model can be.
Then I'll assume that the higher rated team should win don't know by how much and if my pick to win is the underdog on the spread and there aren't any major injuries on the team, then I'll play that game. This is a simple system that should only have me playing a few games a week, but it should win a lot theoretically. This is a simple system which weights different aspects. Here is the results with conditional formatting applied: Now, we need to keep track of the picks our system makes.
For my system, I'm only going to play a game if there are no major injuries and the underdog has a higher 'Total Rating' on my spreadsheet than the favored team. This should allow me to take the better team and a positive spread, giving me a high chance to win. This part is important. For instance, tonight the Heat played the Celtics. My ratings say the Heat have a slightly higher total score.
Now, the spread is Boston Therefore we will take the Heat and the points theoretically and we would win the bet as the Heat won by 8 tonight. This system is very simple, but it should work in the long run.
Now, make sure that before you start putting money on your system, you run it for a few weeks or so and make sure it is profitable in the long run. Sure you may have bad days, but overall you should be winning. The forecast function is very useful to forecast how accurate your next prediction should be.
It also really helps to be knowledgeable in the basic use of Excel and formulas in Excel. You might want to read up on that. I know this guide is very simple, it is intended for those that are new to data analysis and data gathering. I kept adding to the example NBA model, here it is with the area where my predictions are and outcomes: I agree these are very essential. I never used to want to use them but now I'm not looking back. Well, this should help you then.
The real thing that is super useful is the automatic data pulling. Create a plain text file, and enter the URL generated into the body of text, followed by 2 spaces. Sorry this wasn't clearer - enter the url generated as the body of the txt file followed by 2 spaces, you can name the file what ever you want.
Also, my Text Edit was defaulted to saving as rich text file which didn't work when I went to use it as the query - you have to change the default format to plain text. I was able to get this to work on Mac Excel but I had to add the additional lines for formatting below:. Where would you recommend someone looks to next to further develop an understanding of models and how to make them more advanced?
I mean R is very useful, but shouldn't be necessary unless your sport has a TON of teams in it. I would also suggest you look into Matlab and maybe databases. With large data sets you're really getting into an area where coding is necessary. Statistics and Excel functionality. There are so many useful built in functions in Excel and having a basic understanding of what they do and how to use them is very useful.
Google is your friend. Math knowledge helps too. Coming up with a formula that wins is the most important part. You could also incorporate VBA coding and databasing into your spreadsheet, taking out the need for the raw data thrown over on the right side of your sheet. If you hit the export to excel button before you import it on your website it will create a table that includes team names.
I literally made it last night as I typed this out. It has only been run on the Heat and Lakers games last night record. A lot of what you said is what I already knew, but I wish I had this blueprint when I started to make a spreadsheet.
Do you feel like there should be more factors added into the system rather than Oliver's four factors? Do you account for HCA? You mentioned you account for fatigue. How I use this to account for fatigue and HCA. So personally I think it may be a myth. Do what works for you - you will need to experiment. What I did was run R-Square analysis on factors versus points, and I took the top results.
Go read about R-square if you're not familiar with it. This is the most difficult part of my system. Typically I won't bet on a team that lost it's all-star level player for 5 games or so.
I do not factor in HCA at all. Not saying that's correct; it's what has worked for me though. It helps find what variables correlate with each other. Now, there is some inherent problems with using it.
One of the highest correlations with points is FGM. Well of course it does! But this is not useful as a predictor - the correlation is because the points results from made field goals.
It really depends on your lines though. Yes, I would definitely add more factors to the formula, this was mostly a for instance and I made the spreadsheet as I was typing this up. I am going to keep working on it and adding to it. The best way to account for injuries is to first off ask if they really matter that much.
For NBA, injuries are much more important to a team than say NFL because individual players have a much bigger impact. I would really just look at the teams your system says are a play and check for injuries on the team it picked.
Well, in all honesty, this is more of an approximation. With my adding factors into it, I wasn't specific enough. I should have said it like this: You know, the weights thing is really just adjust until you get a good predictive outcome, but the ones I used are the original ones proposed by the Four Factors creator. I would actually probably throw in a Pythagorean ratio in there too. It is a concept initially developed for MLB to predict how many wins a team should have based on their points for and against.
You might want to use different exponents though. Games where you can score more than one point at a time definitely need a different exponent for the ratio. I might use the total ratings from the four factors with the pythagorean to come to a total rating. This is just me personally though, you can honestly just ask yourself what things affect a basketball game's outcome?
You can make a formula based on what you deem important and weight the factors yourself. We kind of did 2 months ago or 2 XMAS ago. We do have tools in the wiki hwere that doesn't get a lot of attention. I swore I was going to stay out of this convo since I didn't want to take away from the post. You invest some money in a position, and at some future time you get a payoff which may be zero. Prices of the positions or odds on bets change as new information….
The odds of that happening are close to ,, to 1. You have to bet on their overall record this season 16 games without knowing any more information than this. Where do you put your money? Extrapolating three wins and a loss over a game season gives us 12 wins, four….
For example, when two events happen in sequence, we have a tendency to believe that the first one caused the second. Tweet For just a moment, let go of the idea that gambling is one day going to make you rich. The point is to appreciate gambling, to….