New RT test

9 replies

Patrick

Customer, bbp_participant, community, 424 replies.

Visit profile

5 years ago #114969

Does it make sense to have variable bar option (50min, 70 min candles for H1 and 25min and 35 min candles for M30 strategies) to see how robust is the strategy?

0

Patrick

Customer, bbp_participant, community, 424 replies.

Visit profile

5 years ago #136200

*timescale analysis

0

tomas262

Administrator, sq-ultimate, 684 replies.

Visit profile

5 years ago #136201

Hi Patrick, that sounds interesting. Will notice Mark about this to consider

0

tnickel

Customer, bbp_participant, community, sq-ultimate, 457 replies.

Visit profile

5 years ago #136213

Hi, I think it made sense.

I know some other people who do this.

 

thomas

https://monitortool.jimdofree.com/

0

geektrader

Customer, bbp_participant, community, 542 replies.

Visit profile

5 years ago #136214

Personally I am doing this too, H1 tested on M30, H4, etc. Should at least give a positive slope. What I am looking for though is shifted timeframe test. E.g. build the H1 bars for the backtest with a shift of +5 +10 or +15 etc minutes. As long as we have 1 minute data, that would be absolutely easy to do for SQ. E.g. don´t start the 1H bar at 13:00, but at 13:05 or 13:06 and keep that 6 minutes shift form there on – this will result in completely new H1 bars for a stability test, yet they are still build exactly from your 1M data that you used to create the main strategy without time-shifted bars.

0

Mark Fric

Administrator, sq-ultimate, 3 replies.

Visit profile

5 years ago #136232

yes, it is possible to do this manually as geektrader describes it, and I think I recommended it in some article. H1 strategy should work on M30 and H4 as well at least to some extent – it could be losing, but not a freefall.

 

But it is a good idea to add this to robustness tests as well.

Mark
StrategyQuant architect

0

Patrick

Customer, bbp_participant, community, 424 replies.

Visit profile

5 years ago #136236

Whereas Monte Carlo analysis is a well-known tool for anyone who has had experience
with the evaluation of risks, timescale analysis is a new method which we have not seen
in any publications so far. We find it a useful tool which can be easily performed with
any standard software package that allows the bar compression of the used price data to
be changed.
 
Emilio Tomasini, Urban Jaekle – Trading Systems. A new approach to system development and portfolio optimization
 
____________________________________________________________________________________________
 
i already use it (test H1 str on H4 and M30)  but from my experience H4 but its too far (60 min vs 240 min) too say strategy is good x bad
 
so i mean, not to test H1 on H4 and M30, but chechk H1 on 55min, 50 min, 45 min, 40 min,  35 min and 30 min for example and then 75 min, 90 min, 100 min or whatever similiar.  nothing too far from 60 min candle.

0

geektrader

Customer, bbp_participant, community, 542 replies.

Visit profile

5 years ago #136240

The shifted H1 candles will do wonders for stability testing (e.g. from 13:06 to 14:06 instead of 13:00 to 14:00). I am currently generating these bars with a own MQ4 script and import them into SQ, but would be cool if SQ would have it out of the box.

0

Patrick

Customer, bbp_participant, community, 424 replies.

Visit profile

5 years ago #136242

The shifted H1 candles will do wonders for stability testing (e.g. from 13:06 to 14:06 instead of 13:00 to 14:00). I am currently generating these bars with a own MQ4 script and import them into SQ, but would be cool if SQ would have it out of the box.

It would be nice to have both test, one with shifted open of the bar (from 13:06 – to 14:06) and another one for changing candles from 60  min to 70 min

0

geektrader

Customer, bbp_participant, community, 542 replies.

Visit profile

5 years ago #136249

Yes, and another really nice option would be:

 

“Reverse data”. That´s simply nothing more than using the data in reverse for a backtest. E.g., if original data is from 1.1.2000 to 1.1.2016, SQ simply builds the bars the other way around, 1.1.2016 becomes 1.1.2000 and 1.1.2000 becomes 1.1.2016, all data in-between is reversed too of course. Which is another nice way to test the stability.

0

Viewing 10 posts - 1 through 10 (of 10 total)