Methods of carrying out a rolling forward - page 6
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
I think the best thing is to do a WF analysis using third party tools, then show MQ and ask them to build it into the tester.
But something tells me it's going to be difficult. I, for example, without a DB, have not made up my mind. A simple calculation:
1 pass of optimization yields 10000+ strings
+ 1 pass of forwarding - another 10000+ lines...
I have an answer, there is no need to collect anything.
If an in-house tester doing back optimisation for a walk step uses some criterion/fitness function , then we save one set with its maximum value. It comes to the forefront to build a good criterion - YES, which can also look at all trades from the frame. The result of the run with a high criterion value is calculated - we store it. I.e. after optimization on the step back we have only 1 winning set on which the step forward will be run.
I.e. for 12 steps, 12 sets will be matched
I have my answer, there is no need to collect anything.
If an in-house tester doing back optimisation for a walk step uses some criterion/fitness function , then we keep one set with its maximum value. It comes to the forefront to build a good criterion - YES, which can also look at all trades from the frame. The result of the run with a high criterion value is calculated - we store it. I.e. after optimization on the step back we have only 1 winning set on which the step forward will be run.
I.e. for 12 steps, 12 sets will be matched
What if your fitness function brings out the best option in the top?
I, for example, have decided for myself to analyse all 10000+ so that by changing the selection criteria I can arrive at the one that will produce consistent results over the entire WF period. In my previous experiment, with <20% drawdown on a yearly period I got 2 months in which drawdowns and drains occur.
Now I want to tighten up unified selection criteria and try to have a drawdown <15% <10%. Add other parameters to the selection criterion - number of trades, recovery, Sharpe, etc. But since we have only 12 files stored, we will have to reoptimize all the 12 months + all the forwards. Every time I have to re-optimize every time I change a selection criterion - this is the only thing I have to do)) That's why I decided to store all data and then reoptimise them.
What if your fitness feature brings out the top not the best option?
I mean forward testing built into the terminal tester. Maybe it should be included to complete the picture? I can only see a few optimization results manually and the tester will calculate them all... but I'm not sure there's a sense in wasting time on it.
Maybe, when seeing all the forwards, we may choose something else, not a drawdown of <20%, as a single selection criterion?
I do this:
1. On TF D1, I select all available history (for euro/dollar I took it from the beginning of 1973 to the present time);
2. I optimize the entire range (over 10 000 bars) of the Expert Advisor to determine the largest value of the recovery factor (RR) as the ratio of the net profit (NPL) to the maximum drawdown (MP) - RR=58935/4657=12.66; Number of trades = 10730; Expected payoff (EPC)=58935/10730=5.49 points; Expected loss (ELO) =4657/10730=0.434 points; Strategy performance criterion (SEC)=EPC/ELO=5.49/0.437=12.66;
3. Then I run the Expert Advisor from any period of history with constant parameters - in this case from the beginning of 1974, 75, ....., 2012 and determine the current values (KEST) = expected payoff current (EPC)/expected payoff final (EPC)= expected payoff current (EPC)/0.434, which shows the stability or instability of the TS over time. This criterion indicates how many times the probability of winning exceeds the probability of losing.
Here is what we got for 41 years, from 1973 to 2013:
I do this:
2. I optimise the entire range (
I've figured out how to implement walk-forward in pure MQL using the native MT5 optimizer in one optimization with full period.
I will give the details later.
I've figured out how to implement walk-forward in pure MQL using the native MT5 optimizer in one optimization with full period.
I will give the details later.
I've figured out how to implement walk-forward in pure MQL using the native MT5 optimizer in one optimization with full period.
I will give the details later.
It's just that there are so many nuances, not only with sampling. This is kind of just the tip of the iceberg.
Question number one, what will volkin forward give?, for those who want to test their systems and so can check it - with whatever conditions your heart desires. The reason is that the strategy may be predetermined.
Although development is probably a good thing.
This is not a volking-forward. You first optimise the whole plot, and then take some measurements on the same plot. During the volking-forward process, we optimize the areas older than the check ones, and then check the more recent ones that are not optimized, then shift them by the size of the check one and repeat the process.
Let me disagree.