decomez6 wrote: ↑Thu Jul 02, 2020 4:38 am
‘Back fitting the first subset? You mean repeating the same strategy in the same sample will give Out skewed results?
No. What I mean is, let's say you have data for 2018 and 2019.
You would look for a strategy that worked in the 2018 data, and then test to see if you got the same results in the 2019 data.
If you just put 2018 and 2019 together in one test, you have a higher risk of backfitting, ie finding that winning combo of parameters in any data that give you winners, but won't necessarily work going forward.
decomez6 wrote: ↑Thu Jul 02, 2020 4:38 am
The only problem is the quality of any data collected. True and valid results depend on it.
True
decomez6 wrote: ↑Thu Jul 02, 2020 4:38 am
Betfair will not allow enough data for anyone to make a LADDER SIMULATOR.
Not true. What makes you say that? You can save the entire API data stream if you have the means to do it, many do. Befair licensing restrictions prevent that feature from being avaiable in commercially available software.
You can kickstart your data by downloading Betfair data here ->
https://historicdata.betfair.com/#/home
Or if you wanted to look at the API then you can do that here ->
https://docs.developer.betfair.com/disp ... omq5qye0ni
Or you can search the forum for threads about using BA to collect data.
decomez6 wrote: ↑Thu Jul 02, 2020 4:38 am
This begs the question how will a backtested strategy behave in an organic , living and reactive market ,When all it knows are static data points.
Static data is enough to give you the foundations of a strategy, you don't need millisecond data because 1s data will average out enough for some basic analysis. Then some people do move on to a simulation, but the most common way is to then do a very small stakes live test.
You really need to design a strategy around your capabilities, it's perfectly possible to find a working strategy in quite modest data for use in pre-race markets, tennis, football etc etc, but if you think you're going to create some sort of high speed in-running strategy, you either need to seriously improve your data gathering and analysis or be prepared to try things live and refine them, with real money.
Automation will require you to have a reasonable R&D budget you can burn on trials, it's not easy or quick and most people who head that way have also developed some basic ideas first by manually trading for months or sometimes years. Coming up with a plan out of nowhere and then expecting it to just make money is setting yourself up for disappointment. Enthusiasm is great but being realistic is much more useful. It's an ultra competitive world, 000s of people involved and many at levels of sophistication you wouldn't believe, and that's why so few people, iro 5%, who try trading can make it work for them.
Try your idea manually first, if you can't make it work using the tv pics and your live brain it's going to be harder still with a primative bot that's got its eyes closed.