Determine the future operability of the vehicle.
And you can see that on OOS - the TC's stats are much better.
| continuous winnings (profit) | 6 (465.32) | continuous losses (loss) | 8 (-199.84) |
not very good either.
that on the optimisation check is simple:
Wind back a year and test one month, then optimize for the previous month, run again compare the results, then optimize for that month and run the next one and so on. this will show a working algorithm or the optimization is just a fit with history.
As for the OutOFSample test, I would make it longer, so it would have at least 30-40 trades. And not only for last month but several out of samples (one for each year, the time can be up to the mark, but of course it would be better in different market conditions).
You see the thing is, the higher the OOS, the less it will live on the real world. Here, too, I think it is necessary to reduce the OOS, so that the TS has a longer life. But concerning the tests of previous years - my opinion it does not play a role because the market was different and testing is not relevant. Moreover, it was written here that after autumn 2006 our brokerage companies started to filter quotes more strongly.
I have noticed (from my expert) that good "inertia" work lasts up to two tenths of the "run-up", i.e. the optimisation period. In this case, the optimization period was 8 months, so if the profit does not start to fall after two months, then you are a hero. I wish you more.
You wind up one year ago and test one month, then optimize for the previous month, run it again and compare the results, then optimize for this month and run the next one, etc. So you can see if the algorithm works or if the optimization is just a fitting for the history.
It seems to me that this is too cumbersome process for an ordinary TS that will live better than half a year with these optimized parameters. There has to be some simpler method.
Six months without over-optimisation in profit is a very good result in my opinion.
I absolutely agree with you. So now do the math, if you have 1 month of OOS you have 5 months of real, 2 months of OOS you have 4 months of real. And so on. So OOS should be reduced if possible, not increased.
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
I thought I'd bring up a very hot topic, in my opinion. I've been struggling with it for a long time and haven't found an answer yet. Here's the TS. Two tests with constant lot 0.1. One on the optimization period, the other OOS - data that TC has not seen. What are the criteria to determine - will it work or not? And for how long?
Period of optimization -
Out-of-Sample -