How can the backtester be fixed? - page 2
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
http://eareview.net/tick-data
, calculate some time in to download the data,
Indeed :)
I've studied the results from a trivial EA that enters and exits based only on the values of indicators on previous 4 hour bars. Nothing else. The results should be the same when backtested on any time frame up to 240 min using "open only", as the open of the next bar after the end of a 4 hour bar does not depend on the time frame. But the results vary.
Firstly on the H4 backtest (about 4 years total): plausible mediocre results.
On 1 hour it gets a supernaturally good spell over a period of more than a year, but recent results look plausible.
And on all shorter time frames (1 min to 30 min) the results for all recent months (more than a year) are ridiculously good.
When I look at the data at a random time just after the results on the shorter term backtests have started turning up sharply, I find these ranges:
1min: 1.22410-1.22510
5min: 1.22410-1.22510
15 min: 1.22410-1.22600
30 min: 1.22410-1.22640
60 min: 1.22410-1.22780
240 min: 1.22778-1.23259 - NOT CONSISTENT
(timestamp 2010-6-3 0000 in all cases)
So it looks like the 240 minute chart is not consistent with the shorter ones.
This is highly consistent with a mismatch between broker data and downloaded data, but of a nature not consistent with the actual prices being not the same (the differences are way too big - more than a couple of pips difference is very unusual in my experience, and certainly not for a whole hour). Also interesting is that the period of strange results is about 2000 H4 bars (about 14 months), while when doing the H1 backtest the period over which performance returns to normal is well over 2000 H1 bars (over 6 months). Perhaps the strange results are when the backtest timeframe and the H4 chart have different sources of data (both downloaded, or both from server give normal results, one from each does not). But on the 30 min chart there is no period of normality at the end of the backtest (the graph remains implausibly steep).
Anyhow, there is still the question of why the data is not consistent. My first guess was that the broker data is time-shifted somehow (I originally thought it was an error dealing with Locales. Another theory was that there was a confusion between bar end and bar start timestamping).
Still need to solve the mystery of why the data mismatches, but getting closer ...
I had all those kinds of problems with strategy tester too, I tried removing all the history and reloading it and then for ages I had no history at all lol I remember posting on this forum about a year ago regarding all that, mt4 wouldnt reload the history and there were big gaps in it when it did, I never found a solution that worked, in the end I uninstalled mt4, reinstalled and never messed with strategy tester again, I decided it is too unreliable, too many unknowns and unexplainable errors to the point where even if I had tester results of high profitablity I would not trust that result enough to let it trade my account so really there is no point in using it when I feel I cant trust the results.
Since then I only test my EA ideas on the live price feed by making indicators place objects on the chart at trade signal entry and exit points then leave it to run on several time frames for some period of time. I have found this to be a far more revealing method of testing, for me one of the most important things revealed by this method is how indicator lines reacting to live prices reveal alot obout their repositioning characteristics when you see where the objects were placed in relation to where the indicator ended up.