Over fitting is a bettors worst enemy!
Because of this I think that it is important that you have a good understanding of what it is and why it is so dangerous.
Every bettor has a selection process. This can be anything from choosing a horse with your favourite name to an advanced analysis. It doesn’t matter whether your process is good or bad at this point, what is important is that everything is a process.
A lot of “old-school” successful bettors will tell you that they don’t have a system. This may be true but what they do have is a process. Often the punter themselves will not be able to tell you what it is, just that they take into account the form, going, trends and other information. However if we went in and analysed what they were doing we would be able to right their process down.
Beginning bettors often start by looking for a winning system. This is because there is a lot of information available in horse racing and you need a way to filter out what is not useful and a fixed set of rules can do this quickly and effectively.
So why then do these systems almost never work?
Because they are over fitted. What this means is that they are based on events that have happened in the past. All races in 2010, for example, have been taken and the rules have been created to produce the most profit from them. Now this all looks great on paper but in reality there is no saying that what happened in 2010 is going to happen in 2011!
When you take the system that has been created and start using it you find that you consistently lose money even though it made huge profits in 2010. This is because you have over fitted!
The system has been built around the information from 2010 so carefully that there is no way this is ever going to work again unless the exact same results happen, which is of course impossible.
So how do you build a system?
Good question! You have to use historical data to build it because that is all we have to go on to figure out what is going to happen. The trick is to put into place some simple techniques that prevent you from being able to over fit.
Today I am going to share one of these secrets with you!
The technical term for this method is called bootstrapping.
What we do is we take all our 2010 data and then choose a random 60% from it. We now look to see what happened and put the first rule in place and if this improves our results we keep it.
Before doing anything else we then go back to our data and choose another random 60%. We then create our second rule on this new data.
Rinse and repeat.
By doing this you are changing the historical information slightly every single time which allows for the differences that we will experience tomorrow, next week and next year.
This prevents that huge over fitting that most betting systems contain and helps to find selection methods that are going to work in the long term!
Hello
I hardly think that one years of data is worth diddly squat, by the time one has drilled down one would be lucky to have 100 bets and little or nothing strike rate wise to show, let alone any profit.
Can’t help thinking that this type of foray into the betting market would require an extremely large bank for a very small return.
Bootstrapping in my opinion is far too simple and is highly inaccurate and more hopeful than confident.
Cheers
Hi Patrick, the amount of bets over 1 year of data is very dependent on what you are doing with it, you could certainly end up with just 100 bets but you could also end up with significantly more. It is more the principle I was trying to show than the specifics but you would need more than 100 bets, I usually say that I want 100 winners as an absolute minimum but ideally 300 or 400 winners.
I would have to disagree with you on bootstrapping, it has shown its effectiveness time and again in my models. It adds enough variation into the data to assist in the prevention of over fitting if used in the correct way.