OZmium Sports Betting and Horse Racing Forums - View Single Post

Quote:

Originally Posted by Vortech

Michal - To follow up on this this good thread can I seek some explanation from you on the following

I've been testing a new particular system and have the following outcomes.

Test Period: 2009 to 2011
5251 Bets - Profit $353.00 on NSW Tab

2nd Phrase of testing for 2012 - 13
2012 Shows a $400 loss
2013 Shows a $150 Profit ... back on track

Is there anything else system developers should be using to test the long-term profitable of a particular system. In this case... would you sack or back.

What other variables should be under consideration apart from sample size

Regards

Hi Vortech,

Its a tuff call. For a system to implode so badly in a single year, there may be issues at play that you may be not aware. One of them is the dynamic of the form or ratings that make up that system. Has there been a change in the form or ratings algorithm ? Might your data be compromised in that year ? Missing results? Is the number of selections smaller?

Does your result rely on a few long shots? While these are OK to rely on (if you can handle the run of outs) there is no guarantee you will get another one. Remember longshot winners are a market mistake, the market is getting smarter so it makes fewer mistakes and while Fluke horses win at great odds all the time and will continue to do so, the ones that do so based on some sort of predictable form/rating pattern have their odds reducing.

Look at only the losers in the 2 samples, it may be that the same patterns lose and it is only that there are more of these in that year and eliminating them improves your result across the board.

It may be that there are too many unintentional bets, that is bets that fall outside of your intended filter range of the original sample, as explained in the article on our website.

The decision to keep or sack is not one for me to make, there are other good threads here that look at other things that may help you make that decision.

The best guide is logic, is your system logical? Then large sample size, however it must be in relation to the test period. 2000 selections in 12 years is the same as 166 bets a year. Please don't make a mistake thinking that in such a situation 2000 selection means anything. Your numbers look OK as I believe that if you can find 2-3 bets a day with a system and make profit then you might be well onto an actual system.

Looking at the dividends distribution is also important. Where is your profit coming from and can you handle that? Waiting for a 'system saving' longshot may take you outside your comfort zone.

Understand that a system will have an AVERAGE result dynamic that is a long term indication of the system performance however the daily results of any system is almost random. It is only long term that the system returns to its true performance. Best illustrated with the coin toss, 50-50 longterm and anything from between 0 -100 when you take a random sample size. The complexity of horse racing however means that the randomness is far more prevalent then we all would like. The above why we always give our strike rate results for longterm (all our) data sample. This is the ONLY true and honest measure that cannot be manipulated.

Hope this helps