Who want to test good EURUSD M5 Strategy

こんにちは、,

I have a good EURUSD M5.

The strategy runs on EURUSD M5 and M15

Who want to test it ?

Well, took up your question. 

I am not sure what you wanted to find out, but I did the simplest thing and downloaded you mq4 file and run it on the last 5 months of M5 eurusd data. The attached image is my result: it lost.

The start/stop dates of my data are; 2013.10.07 and 2014.01.30.  

This from the the MT4 backtester:

Total net Profit: -526.00, Profit factor: 0.71, Total trades 79 (28 short, 51 long).

Now I don’t know if my data overlaps your data or not, but I couldn’t show this was profitable.

How does this compare to your test period and do I have a good sets?

Of course when one sees a strategy trending in one direction or the other with sizable pip gains/losses, (this) one is often interested. Namely, for a losing strategy, I think about switching the trade directions around to make it profitable. I didn’t try this.

The next questions I ask are; 

How robust is this to price data (can be tested with SQ) and what are the sensitivities of the parameters to profits/losses (can’t be addressed by SQ).

Can you offer and more thoughts on what I see versus what you see?

ジム 

こんにちは、,

I generated this strategy for two years with fhdb data

I made a backtest of 12 years with fhdb data (test with unseen data)

It is very interesting. The equity curve it good for two years in the generation process.

The equity is good for the unseen data for 10 years.

The big question is, is this strategy is curvefitted?

トーマス

皆様、こんにちは,

Basically I am having this concern too.

Is the following true?

Even though we include ”OOS data” for strategy generation, it is actually similar to generating a strategy based on a full data comprised of IS and OOS?

OR is it that,

Generating strategy on a 2-3 years data and use it for the unknown 10 years data (not part of strategy generation) is more logical? If it works, there is a even higher chance of working out.

TFL

I don’t have the answer. But I do know that when I create a strategy on a small period of 2 years and then test it on a larger unseen data range, the strategy falls a part almost every time.

While this does not answer your question directly, my experience does let us know that this method will unveil at least some (almost all in my case) of the bad strategies.

ステアルノ

I think when you use OOS sample for filtering or testing then OOS data becomes contaminated and is not OOS any more. For robustness check I think it’s more wise to set apart a part of the OOS data for a final check¦.