Possibly: Reverse In Sample and OOS periods.

I'm uncertain about the effects of this, but...

Many people are testing a hypothesis that if we generate many EAs over a very short time period, these EAs can take advantage of temporary market bias. I don't know if this is true or not (I hope that it is!), but if it is:

It seems that many people don't want to include an OOS period...and I think of two reasons for that: 1. It takes more work (and slightly longer generation time?), and perhaps this extra work is unnecessary; and 2. They are worried that the OOS period "uses up" the most recent and important data.

If we are going to include an OOS period, it seems to me that it would be better if this OOS is at the beginning of the data series. Why?

Because in that way, the EA would both be optimized with the very latest data AND still be adjacent to the OOS period. In this way, if the market is recently changing in a way relevant to the strategy, the OOS period will still look relatively the same (quality-wise) and it won't lessen the efficiency of the generated strategy due to being optimized over older data.

Using smaller data periods, I don't see a drawback to this approach, only benefits.

Maybe you have knowledge about this one way or the other?

Wish: In report session would be good if we can see the Backtest output for OOS too

suggestion in green.

I've found good strategies that relies only Entry rules + Risk Management (Take Profit + Stop Loss). The exit by indicator never happens because the exit part is an very unsual trigger combination like RSI(32) > 99 or very high volume.

So here is my wish1:

• Exit rule indicator with option "Don't use any indicator".

So we can generate strategies that relies only on Entry rules + Risk Management (Take Profit + Stop Loss).

----

I know that in the end of indicators list we have the option to "Enable all - Disable all" for checkboxes and this will turn all checkboxes. But can you consider a click event in the title? Exemple 1 click in the title "Entry Rule" will turn only check box true

So here is my wish2: Check / Uncheck  checkboxes when click the title "Entry Rule--- Exit Rule"

I hope what I have requested is possible.
Thank you guys!

Would it be possible to see the variables in each of the Monte Carlo tests?  Click on a line and it would show up what that particular scenario was? It would be useful to see where there are outliers which variables are likely to be causing the outlier.

Hi,
Would be like to have the option to be able to import Renko Bars Data into EAS

Thanks

If this were to be implemented, I would request this be optional and not the default.  I sort of have an aversion to old data:

(a) Indicator parameters are computed using a data set.  I would prefer these computations be performed using the latest and greatest data.  When you include old data, then it skews the results using data that is no longer relevant.  I can't say where the cut-off line is between old and new data, just that I don't think more data is necessarily better.  I'd rather the focus be on newer data.

(b) The larger the data set, the longer the computation will take.  It may not be noticeable because Popov's algorithms are so fast, but if you double the size of the data set then you will halve the number of strategies that will be generated / optimized within a fixed period of time.  I'd prefer more strategies because then there is a greater probability of discovering some winners.

Everything is a trade-off -- I'm just suggesting if this feature were added that there be an option whether or not to use it.

Hi Steve, thanks for your comments. Please notice we can always use "data horizon start time" to limit the maximum number of bars we use to build our strategies. On the contrary it is not possible to have more data (for example m15 bars) than the bars that the broker allows us to download. This data (the data from the broker) is changed continuously /cycled, so the broker doesn't allow us to download more than a number (fixed) of bars. Being able to store more bars (instead o losing them) is something good. We could always use them all or only a few...

Yep -- you're right -- sorry...  I forgot about being able to use the data horizon start and end dates to control how much and which data is used.   

Popov just now mentioned an "automated data uploading service" -- I think that will be very, very cool.

One other thing I might add about using a data set to train / optimize that is different than the one used for back testing...

People often worry about "over-optimization" or "curve-fitting".  If the data set you use for training / optimization is different than the one used when back testing -- and if the back testing statistics continue to look good -- then it is highly unlikely that the parameters computed during training are curve-fitted.   Does this make sense?

Another thing, which should be very trivial to implement, is to be able to sort Collection and Portfolio by strategy ID. This makes it useful when multiple variants of a strategy are added to the collection from the optimisation process. This would allow quick identification of all variants in the sorted list.