Backtesting/Optimization - page 42

New comment
 
belea:
...And right now i hope its pretty obviously that what its on chart its not the same like in journal.. Thx

It is not the same as you have price value on the chart and MACD and Stoch values are in journal.

You shound compare value of MACD and Stoch in separate window with the journal but as I see it may be the same. Can not say exactly as you wrote hotizontal line on the main indow instead of separated window.

 

MT4 optimization models

I would like to start a discussion on the optimization options that are currently available in MT4 (not the actual implementation of the GA used for optimization, which should be a topic for MT5).

In its current version MT4 has 3 “models”:

1) Every tick (claims to be the most precise optimization method)

2) Control points (described as a very crude method)

3) Open prices only (if expert uses opening prices only)

To start with, option 2 is totally useless and should be disregarded for backtesting.

I would like to challenge the claim that the “every tick (the most precise method)” is actually the most precise when it comes to backtesting of trading strategies. First of all

I will make an assumption that the historical data that the backtester uses is correct, otherwise further discussion is pointless. Since the finest time frame MT4 can use for backtesting is one minute that is by definition the most precise actual data available. The “every tick” algorithm simulates the tick data within the finest time frame available (i.e. 1 minute) which can be extremely incorrect during large moves after news announcements. Since the tick distribution within 1 minute bars is MT4-algorithm dependent you are likely to see changes from one software build to another.

In my opinion the most precise method to back test is to put your EA on 1 minute time frame (using opening prices only) while programming the EA to use user-defined time frame for indicators (15, 60 min, etc.).

The next question I propose to discuss is the usefulness of time frames longer than one hour (up to a day or so). The reason for it is that the bars longer than one hour are broker-dependent and GMT+1 broker will have 4-hour bars quite different from GMT+2 broker which makes an EA that depends on such a time frame very time-zone dependent.

 

Great article on proper way to optimize an e.a.

Expert Advisors Based on Popular Trading Systems and Alchemy of Trading Robot Optimization (Cont.) - MQL4 Articles

Learn and Earn!

Peace,

F.F.L.

 

Thanks. Really good article.

 
newdigital:
Thanks. Really good article.

Yes, thanks FFL

FerruFx

 

Backtesting Sample

When backtesting an indicator or system, what would be the ideal sample size to correctly gauge the market.? Yes, yes i know, how long is a piece of string....! Most indicators have a default setting of either 14 or 21 so this implies the last 14 or 21 days is a good enough sample for forecasting future market movement. It could be an interesting discussion and i would like to kick it off here. Is there a predictable pattern for each pair or is it a function of fibonacci sequence...?

My idea is to trade a specific EA but then keep optimising once a week to plug in the most recent numbers..but how far back should i go. I don't beleieve you should go back as far as possible as this takes too much time and does not place the correct weighting on the most recent market movements. Markets change all the time and it would be nice to use a well thought out model then just to use the standard default settings given on most indi's

 

2 things to check

1. it has to survive all market conditions = go testing as far as possible

2. current fluctuations short term = try to go with the flow, adjust your indicators like it is a frequency.

ok?

daet:
When backtesting an indicator or system, what would be the ideal sample size to correctly gauge the market.? Yes, yes i know, how long is a piece of string....! Most indicators have a default setting of either 14 or 21 so this implies the last 14 or 21 days is a good enough sample for forecasting future market movement. It could be an interesting discussion and i would like to kick it off here. Is there a predictable pattern for each pair or is it a function of fibonacci sequence...? My idea is to trade a specific EA but then keep optimising once a week to plug in the most recent numbers..but how far back should i go. I don't beleieve you should go back as far as possible as this takes too much time and does not place the correct weighting on the most recent market movements. Markets change all the time and it would be nice to use a well thought out model then just to use the standard default settings given on most indi's
 

which mt4 data feed is the valid one

hi guyzz,

i just had a case with odl, i had 2 live account which i set for an EA to running in the same tf and everything is the same

but the different is the account number.and guesswhat? the mfi indicator value is different one to another and that makes the ea is opening in one account but not in another.. and i just ask the customer service in odl and they don't know what happened,

what i want to ask is, which data feed and indicator that barely near valid , is it only reuters chart can tell the valid indicator value?

please help me as you guys know even in many mt4 has differrent indi values compare one to another.thx a lot

 

Optimization - Genetic Algorithms?

The optimization status is 172/10496 (1030301)

genetic algorithm is ON

It has been 30 hours 32 minutes so far (after this it says 1822.45.07)

How much longer is this likely to take?

I haven't used genetic algorithm before, I thought it was meant to speed up as it progresses?

1...353637383940414243444546474849...95
New comment
Reason: