What historical data source do you actually trust for MT5 backtesting?

New comment
 
Been running intodissapointed backtest results lately and traced 
it back to the data source rather than the strategy itself.

Tested the same EA on EURUSD M1 across three different data sources 
and got completely different equity curves. Same parameters, same 
everything. The spread between best and worst result was about 16% 
in drawdown.

Main issues I found:
- missing bars around high impact news (NFP, FOMC)
- DST handling inconsistencies between sources
- silent forward-fill on gaps in MT5 history center

Curious what sources people here are actually using for serious 
backtesting. Free or paid doesn't matter, just looking for something 
clean and consistent.

Dukascopy? HistData? Something else?
 
Bart Sabajo:
Been running intodissapointed backtest results lately and traced 
it back to the data source rather than the strategy itself.

Tested the same EA on EURUSD M1 across three different data sources 
and got completely different equity curves. Same parameters, same 
everything. The spread between best and worst result was about 16% 
in drawdown.

Main issues I found:
- missing bars around high impact news (NFP, FOMC)
- DST handling inconsistencies between sources
- silent forward-fill on gaps in MT5 history center

Curious what sources people here are actually using for serious 
backtesting. Free or paid doesn't matter, just looking for something 
clean and consistent.

Dukascopy? HistData? Something else?
If you just want to clean up or add to sparse historic data for years ago, I find that free HistData is fine for that purpose. Just know that the smaller the timeframe/chart structure that you use, the less that 3rd party data will simulate your broker-dealer's data─where you'll actually be trading. I certainly wouldn't rely on raw ticks from any 3rd party for backtesting on every tick. Again, building OHLC H4 bars or 100 pip Renko bricks is a different story.
 
Ryan L Johnson #:
If you just want to clean up or add to sparse historic data for years ago, I find that free HistData is fine for that purpose. Just know that the smaller the timeframe/chart structure that you use, the less that 3rd party data will simulate your broker-dealer's data─where you'll actually be trading. I certainly wouldn't rely on raw ticks from any 3rd party for backtesting on every tick. Again, building OHLC H4 bars or 100 pip Renko bricks is a different story.
I've had good results with HistData for forex — 
cleaner than most broker exports. For futures 
specifically it's harder, most free sources have 
gaps around rollover dates. What instruments are 
you looking for exactly?
 
Bart Sabajo #:
I've had good results with HistData for forex — 
cleaner than most broker exports. For futures 
specifically it's harder, most free sources have 
gaps around rollover dates. What instruments are 
you looking for exactly?

I've used free HistData for FX only.

I would never use 3rd party historic data for say, CME futures. Of course there are gaps between contracts of the same instrument. The front month contract always overlaps with the next contract in time, and there are many ways/methods/algo's used to rollover. I have no idea how a 3rd party data provider could know "how" I want to rollover.

EDIT: To be clear, my EA's trade based on custom generated Renko charts (open and close prices) so historic price differences among FX data providers must be rather substantial to create problems for me.
 
Ryan L Johnson #:

I've used free HistData for FX only.

I would never use 3rd party historic data for say, CME futures. Of course there are gaps between contracts of the same instrument. The front month contract always overlaps with the next contract in time, and there are many ways/methods/algo's used to rollover. I have no idea how a 3rd party data provider could know "how" I want to rollover.

yes, that's right.

Because if someone found good datafeed (external or any), but I have to use his EA on my broker's datafeed so ... sorry ... I will get the results which is related to the datafeed of my broker ...

==============

For information: there are the coders and traders; but there is some other specialization - testers. I remember Sadaloma from Indonesia and Brono from France ... they were the testers, means: they tested EA on almost all the brokers' datafeed by backtesting and real trading (on demo and live) to find about the following: which broker is most suitable for this EA, and the settings (of EA) for that.

Yes, they did it (mostly) for free in the forum threads (former tsd forum) ...

-------------

But if someone told here that he found some very good extenal datafeed to backtest ... so I want to ask: what is my broker about sorry?

 
Bart Sabajo:
Been running intodissapointed backtest results lately and traced 
it back to the data source rather than the strategy itself.

Tested the same EA on EURUSD M1 across three different data sources 
and got completely different equity curves. Same parameters, same 
everything. The spread between best and worst result was about 16% 
in drawdown.

Main issues I found:
- missing bars around high impact news (NFP, FOMC)
- DST handling inconsistencies between sources
- silent forward-fill on gaps in MT5 history center

Curious what sources people here are actually using for serious 
backtesting. Free or paid doesn't matter, just looking for something 
clean and consistent.

Dukascopy? HistData? Something else?
Always use what your broker makes available to you - even if there's not enough of it - or change broker
New comment