Backtesting Methodology (what's a better way?)

[numbertea]

I've been letting my automated system trade on its own with no intervention for about 2 months. 99 trades for me to research and improve the system. I have all the trades imported into excel. I have for each trade: date, time, which specific algo hit, the prices traded at, the resultant P/L. I am sorting according to P/L lowest at top and now running those days data through my system with a visual of all the chart and indicators I have been using. I check out why this was such a loss and postulate what I might be able to do to fix the issue without negatively affected the good trades that I may have in the future.

I then check through all trades similarly and look for improvements to the system.

I also of course sort the trades by algorithm type and see what the risk profile is for each trade type and whether or not the trade should still be part of my system or if it needs improvement to stay.

After all work on this I will backtest on all data and verify that I have made good choices in my efforts for improvements. Then the system will be back online Monday morning.

Is there better or less time consuming ways to go about this R&D? After I let it run another 2 months I'll be doing this again.

I do keep an eye on P/L each day to be sure that I didn't make a typo or something which didn't rear it's head while testing but other than something horribly wrong I will just let it run itself.

Cheers

 

[Solas0077]

The way you do it is best because you do it, not some software.

 

[FXColes]

Far too many EAs out there are designed to make the developer rich off deception and a commonly ploy is to impress a newbie 'holy grail' searcher with a an impressive looking backtest. All you do is a over-optimize some parameters in an EA for for 6-12 months in such a way that it makes a dreamy equity return. This is one of the main reasons I don't like to use exclusively use EA's and am pursue manual trading. Also, the more parameters you have the more you can optimise it to make it look super profitable but also the more likely it will be to fail in the future.

I don't personally believe you can have a fully automated system that can make consistent, long-term gains, but I do think backtesting can be used to quickly identify the EA's which clearly aren't going to work.

 

[NVP]

I've been letting my automated system trade on its own with no intervention for about 2 months. 99 trades for me to research and improve the system. I have all the trades imported into excel. I have for each trade: date, time, which specific algo hit, the prices traded at, the resultant P/L. I am sorting according to P/L lowest at top and now running those days data through my system with a visual of all the chart and indicators I have been using. I check out why this was such a loss and postulate what I might be able to do to fix the issue without negatively affected the good trades that I may have in the future.

I then check through all trades similarly and look for improvements to the system.

I also of course sort the trades by algorithm type and see what the risk profile is for each trade type and whether or not the trade should still be part of my system or if it needs improvement to stay.

After all work on this I will backtest on all data and verify that I have made good choices in my efforts for improvements. Then the system will be back online Monday morning.

Is there better or less time consuming ways to go about this R&D? After I let it run another 2 months I'll be doing this again.

I do keep an eye on P/L each day to be sure that I didn't make a typo or something which didn't rear it's head while testing but other than something horribly wrong I will just let it run itself.

Cheers

99 trades ?........is that all ?

N

 

[NVP]

Far too many EAs out there are designed to make the developer rich off deception and a commonly ploy is to impress a newbie 'holy grail' searcher with a an impressive looking backtest. All you do is a over-optimize some parameters in an EA for for 6-12 months in such a way that it makes a dreamy equity return. This is one of the main reasons I don't like to use exclusively use EA's and am pursue manual trading. Also, the more parameters you have the more you can optimise it to make it look super profitable but also the more likely it will be to fail in the future.

I don't personally believe you can have a fully automated system that can make consistent, long-term gains, but I do think backtesting can be used to quickly identify the EA's which clearly aren't going to work.

your Sample period is the key dude ...........also in forex I would suggest you just pick a very very difficult period to trade and find the system that keeps you least in the red ......

trading is about not losing ........thats a totally different Concept to winning

its not difficult to find the bad times historically
N :smart:

 

[numbertea]

Appreciate the opinion but I am way past that. I'm just trying to figure out the best way to analyze the data from past trades. I'm not talking about optimizing my system so much as picking out the best ways to insure letting the winners run and cutting the losers as quick as possible. My system uses very general trading strategies and I backtest on so much data that optimizing really isn't my issue here. What I'm trying to know is stuff like: what is better, 70% good trades with 5x profit and 100 trades or 4x profit with 75% good trades and 60 of them over same time period. What will be better going forward? I'm going to run multiple systems to get answer later but I can only actually trade 1 system at a time.

Cheers

 

[NVP]

Appreciate the opinion but I am way past that. I'm just trying to figure out the best way to analyze the data from past trades. I'm not talking about optimizing my system so much as picking out the best ways to insure letting the winners run and cutting the losers as quick as possible. My system uses very general trading strategies and I backtest on so much data that optimizing really isn't my issue here. What I'm trying to know is stuff like: what is better, 70% good trades with 5x profit and 100 trades or 4x profit with 75% good trades and 60 of them over same time period. What will be better going forward? I'm going to run multiple systems to get answer later but I can only actually trade 1 system at a time.

Cheers

umm ...are you sure you are asking the correct questions ?

and what good do the answers prove ?...the market will do what it wants in the future regardiless of these answers ........

See signal - trade signal - Exit Signal ........the rest is detail

good luck in your studies though.....we all have different paths to walk in trading at different times

N

 

[meanreversion]

Appreciate the opinion but I am way past that. I'm just trying to figure out the best way to analyze the data from past trades. I'm not talking about optimizing my system so much as picking out the best ways to insure letting the winners run and cutting the losers as quick as possible. My system uses very general trading strategies and I backtest on so much data that optimizing really isn't my issue here. What I'm trying to know is stuff like: what is better, 70% good trades with 5x profit and 100 trades or 4x profit with 75% good trades and 60 of them over same time period. What will be better going forward? I'm going to run multiple systems to get answer later but I can only actually trade 1 system at a time.

Cheers

Work out your overall return in %/year, calculate your maximum drawdown and then look at the ratio between the two. If it's more than 1, you may have a decent system. Similarly, you could calculate your Sharpe ratio.. take the annualised volatility of returns, divide by the annualised standard deviation of the same set of returns.. if the resulting number is greater than 1, you're on the right track. Is that the sort of thing you were looking for?

(PS don't overanalyse by drilling down to individual trades, you need to just consider the entire portfolio returns)