backstesting question

[buddyanton]

how long shud you backtest for?
a couple months? a couple weeks?
once you have determined time in which you have backtest, what shud your project profit/loss be.

say in 12 months 20-40% growth is that good, conservative or below average?

your help is appreciated

[david wood]

a couple of months is not enough. go for a couple of years.

Also it's not enough to check what was the profit, yo need to look also at the maximal drawdown. for example if you had a 20% profit and 20% drawdown, it sounds bad. 20% profit with 5% drawdown is much more desirable.

[pipso facto]

If you can take out say about a third of your data and not get a significant difference in your results, you should have enough data.

I use daily charts and use about 15-20 years data for backtesting sometimes 30 when I can get it. Probably a bit of overkill, but I like to see how my systems respond to as many different trading conditions as possible.

As far as expected profits, I second David on this and agree that they are irrelevant without your max drawdown. A system that returns 20% year and suffers a max drawdown of 10% including commission would be pretty good indeed.

[rhodytrader]

Originally Posted by buddyanton

how long shud you backtest for?

It depends on your trading timeframe and the frequency of trades. With a high frequency system then the test period can be shorter. The basic rule of thumb is that you should make sure your test set includes all market conditions you're likely to see: trending, non-trending, high volatility, low volatility, etc.

[buddyanton]

hey boys,

thanks for the help. so taking max drawdown into account. at the end of 12 months if you are still showing 30% with max draw down 4%. is this fair?

[ellave]

What exactly do you mean by drawdown, please can you guys elaborate?

[rhodytrader]

A drawdown is the distance from an equity high point to the equity low point which follows it. So if your account balance increased to 1500 due to your gains, then fell back to 1200 from some losses, the drawdown would be 300 or 20%.

[david wood]

Originally Posted by pipso facto

I use daily charts and use about 15-20 years data for backtesting sometimes 30 when I can get it. Probably a bit of overkill, but I like to see how my systems respond to as many different trading conditions as possible.

30 years is not a "bit of overkill" , it is more of a mistake.
The Forex market was so different 30 years ago then today... even 10 years ago
that I bet it can never return to that same condition again.
So you are testing systems on un-realistic conditions and getting meaningless results. I would not test for more than 2-3 years.

[pipso facto]

Originally Posted by david wood

30 years is not a "bit of overkill" , it is more of a mistake.
The Forex market was so different 30 years ago then today... even 10 years ago
that I bet it can never return to that same condition again.
So you are testing systems on un-realistic conditions and getting meaningless results. I would not test for more than 2-3 years.

Firstly, The point of backtesting is not to recreate any "conditions" that may have existed. Doing so would be futile since one never knows what will happen tomorrow. Your "conditions" from 2-3 years ago may not exist then either.

Secondly, just because I start my backtest 20 years ago does not mean that I don't have the results from 2-3 years ago. I don't see how the two are mutually exclusive. I start the test 20 years ago and run it until as recently as my data allows, which includes the last 2-3 years.

I should mention that I use daily data; if you're using hourly data going back 2 years, you may still use more data than me. Still, I think the more data the better; I don't see how it hurts.

So thanks for the advice, but I think I'll pass.