when i was immersed in BF, i gravitated twds totally mechanical strategy approaches. a typical example might be looking for large deviations in price to counter matches on volume vs price across all runners etc. I had to gather a lot of data to backtest the premise of the ideas as there is just no way that you can pin a tail on a donkey blindly enough times to make it profitable.MemphisFlash wrote: ↑Mon Jan 22, 2024 10:35 pmyes that is correct, finally someone gets it.The Silk Run wrote: ↑Mon Jan 22, 2024 6:19 pmI couldn't agree more Tom. Keep it simple, basic fundamentals. Each animal has a 1:6 chance of winning ....
so, in my experience, you DO need historic data to validate and refine ideas but in the mechanical world, those are divorced from the nuances of Form/Track/Time of day etc and are focussed on looking at repeated patterns/deviations/reversals etc with a view to applying those across a range of markets. like sniffer says (I was involved in his greys test but he was 100% the brains behind it), you definitively NEED to split out your data into two sets, one for in sample testing (the 80%) and the other remaining 20% to be tested out of sample across your mechanical rule/strategy.
In my mechanical strategies, I had to use the BF API to do this as BA (despite many repeated requests from myself and others), was unable to plumb in any features to allow for backtesting. So to recap, imho, using historical data to assess potential outcomes can be useful as long as you approach it with a mentality of in sample vs out of sample data, otherwise it definitely is a pure backfit and will most likely deliver a negative outcome in live markets.
