Profit factor (pf) vs Avg.Trade Net Profit

[paglos]

Hi all,

I am backtesting an automated trading system, I have two different results:

Parameter A : High pf Low Avg.Trade Net Profit ( Because more trades so high pf )
Parameter B : Low pf High Avg.Trade Net Profit ( Because less trades high Avg.Trade Net Profit)

If other performance/risk factors are the similar, which system you would choose ? I trade forex with Interactive brokers, as far as I know I trade against other IB clients and Market makers, so liquidity might be a problem ?? Put less money to each trade and trade more or put more money and trade less?

 

[Handle123]

First question, how far back did you back test?
Many never back test far enough and don't have a clear idea of what is min sample size.
Also, a good method should work over other timeframes, might not give best results, but should be profitable.
What is drawdown of each?

 

[nonlinear5]

Hi all,

I am backtesting an automated trading system, I have two different results:

Parameter A : High pf Low Avg.Trade Net Profit ( Because more trades so high pf )
Parameter B : Low pf High Avg.Trade Net Profit ( Because less trades high Avg.Trade Net Profit)

If other performance/risk factors are the similar, which system you would choose ? I trade forex with Interactive brokers, as far as I know I trade against other IB clients and Market makers, so liquidity might be a problem ?? Put less money to each trade and trade more or put more money and trade less?

I would not use either Profit Factor or Average Trade for performance evaluation. Consider the (extreme) demonstration below.

Systems A and B have the same profit factor, 2.0. Clearly, a lot of information about the system performance was lost in that "2.0" figure. Virtually all risk-adjusted performance metrics would rate system A much higher than system B. But profit factor fails to distinguish between A and B, as it's totally oblivious to the shape of the equity curve.

 

[i am nobody]

I would not use either Profit Factor or Average Trade for performance evaluation. Consider the (extreme) demonstration below.

Systems A and B have the same profit factor, 2.0. Clearly, a lot of information about the system performance was lost in that "2.0" figure. Virtually all risk-adjusted performance metrics would rate system A much higher than system B. But profit factor fails to distinguish between A and B, as it's totally oblivious to the shape of the equity curve.

The chart is not very realistic. How can you make 1 big profit and then 12 consecutive losses? That system sucks and should first be improved before trading it.

I think you should first watch how much the market offers you as potential profits.

Next you should try to take as much profit as possible, combined with the lowest number of trades possible.

Each trade has a risk. With big profits per trade the risk/reward ratio is much more favorable.

Try to find the optimal point between maximum total profits and minimum number of trades.

 

[paglos]

First question, how far back did you back test?
Many never back test far enough and don't have a clear idea of what is min sample size.
Also, a good method should work over other timeframes, might not give best results, but should be profitable.
What is drawdown of each?

Hi Handle123
my backtesting period is 15-24 months, daily forex trading. I have only that much historical data.

 

[Handle123]

You need to buy much more data, this is what losing traders have for data, if dailies a min is 10 years and more. You want sample sizes of 3,000 then you have a good enough amount to get many answers. Plus you go through different cycles of economies. It is better to use your funds in back testing than do what I did was lose money in the markets then buy data.

 

[paglos]

You need to buy much more data, this is what losing traders have for data, if dailies a min is 10 years and more. You want sample sizes of 3,000 then you have a good enough amount to get many answers. Plus you go through different cycles of economies. It is better to use your funds in back testing than do what I did was lose money in the markets then buy data.

Thanks that's very sound advice. But I was "told" to test data from environment I trade which is IB. I am afraid data from other sources will not reflect what will happen in IB forex platform.

 

[Handle123]

Thanks that's very sound advice. But I was "told" to test data from environment I trade which is IB. I am afraid data from other sources will not reflect what will happen in IB forex platform.

Well then you need a different program to do backtesting, I don't see how one software be any different than another if you can transfer studies or whatever you are using. Can't you find data that can be loaded into IB? I know data that IB uses is very spotty, so anything would be better.

 

[Xela]

I am afraid data from other sources will not reflect what will happen in IB forex platform.

It may be true that it won't completely accurately reflect what happens on IB, but this is unlikely to be hugely significant, and extremely unlikely to make the difference between overall profit and overall loss - and it certainly doesn't in itself make it "not worth doing".

 

[paglos]

thanks all, any recommendation about intraday fx history vendors on the cheap side ?
I found CSI but it needs some work to download and feed into my program.