Hi guys, i’ve been writing my own EA’s for MT4 and backtesting them, i’m aware of the dangers of curve fitting / over-optimisation etc and have read a few books on automated trading so should be able to use a valid approach (i hope!) to come up with something potentially profitable.
My problem is the software & data i’m using, seems very cumbersome and slow, there must be better ways. See my set up below;
- I’ve been using FX dreema to create the EA’s and export mq4 & ex4 files, i’ve paid for the full subscription so would like to keep using this; it is really good and i’ve invested time learning it now.
- I’ve been backtesting on the MT4 strategy tester; only allows 4gig FXT files (about 5 years of data) and takes probably 3mins to run a 1year test on the daily timeframe.
- i’ve been using free data from dukascopy by downloading through Tickstory Lite and saving in MT4 folders.
How do you guys test an automated system over say a few decades? What software is best? Where do you get the data? Im not bothered if i have to pay as long as it isn’t a crazy amount!
Thanks in advance.
Your current development environment seems fine, it could get more efficient for sure but besides a boost in speed you won’t get much else. I’d say stick with what you currently have rather than spend that time constantly upgrading when you could be using it to develop your strategies.
As for backtesting over several decades, it’s just a matter of getting the data which can cost quite a bit if you want accuracy. But at the same time, what’s the point of backtesting for that long of a period? The market 20 years ago is quite different than what it is today, it wouldn’t make any sense to base your modern day trading decisions based on the performance 20 years ago.
Ah thanks for that makes sense and sure is cheaper / easier. i just get conflicting info, the books i’ve read suggest decades of data. If you don’t mind what data do you recommend and how far back do you go for successful results?
Cheers.
Yeah lots of conflicting information out there. My advice is not to go word for word but rather take it as different opinions and then determine what you believe would work for your goals. I can see times where decades of data would be required; they are rare and situational.
I have been collected my own tick data for several years now so I have quite the collection. I would suggest starting to collect your own data as well. As for testing now, I’d just use what you have been working with already (Tickstory), ideally you want to be testing with tick data as opposed to interpolated bars.
The time frame for your backtests depends on the specific strategy. If your trading strategy is trading on the 5 minute time frame, perhaps a few months is all you need. But if you have a long-term trend following strategy on the daily time frame, then you might need a year or two. This is just factoring in the time frame your strategy is running on. Another variable would be the frequency of entry signals. If your strategy is taking multiple trades a week then 6 months might be sufficient. Whereas if you have a strategy that generates 1 trade a week, then you may be 1+ years worth. There is no objective rule, although a rule of thumb is to generate 150-250 trades to get an accurate representation of your strategy. This just comes down to personal experience and what works for you. Once you are familiar with how it all works, then you’ll be able to pump out strategies like an assembly line! 
I personally find the speed of “Every tick” backtesting annoying, so when I want to try something for fun, I’m using “Open prices model”. Well, I must follow some rules, I can’t use trailing stop and small stops, otherwise the results will be fake. But some once-per-bar strategies can be backtested much faster in this way.
In the last weeks I was playing with a Demo account, manually, with some success!. Normally I don’t trade at all. But I realize that markets are different. EURUSD has strong movements once or twice a day, it also follows trendlines with mind blowing accuracy. Other markets looks calm and boring. Other markets looks crazy and it’s hard to find some pattern. If EURUSD looks like this now, do you think that it looked the same way in 2003?