Question for those who trade what they test...

[LeviF]

How does your backtested system performance compare to real results? What ratios of returns and drawdowns have you experienced? IE - real results of 50% of test CAGR and 2 x test DD.

[LeapFrog]

...real results of 50% of test CAGR and 2 x test DD.

Golly I hope not - that would be a shocking result and show very poor backtesting work, IMHO. I beat my systems, by a decent margin when they are in drawdown and about equal when they are wining. Each and everyday I compare my actual trades to system generated trades and chase up any and all discrepancies. Negative discrepancies means something has gone wrong and should be corrected immediately.

Having said that, I don't trade my systems like a hedge fund might. I take much higher risks than a hedge fund, expect and do get better returns than a hedge fund (i'm generalizing here), and DON'T trade my systems for 30 years straight without a break, so it depends how you measure things.

[7432]

I can't remember where, but I have read comments about returns being 50% of test results, DD bigger, but not 100% bigger.
I assume its because they were testing in a closed environment, like little or no slippage, no taxes or expenses taken out, no adjustment for volume, continous trading, that kind of stuff.
I make that assumption because I can run a 10 or 20 year test that shows me making gazillions of dollars and having a MAR of 5.
but when I add slippage of 20%, pay taxes and myself, and adjust volume so my test isn't trading 10,000 lots of euros, the results don't have me owning the world anymore.
just a little piece of it. haha.

[svquant]

Perhaps there are two questions here?

In general the difference between your real equity curve (account) and the theoretical curve (TBB) should be very small. Slippage differences, TBill rate differences, perhaps a missed or late trade due to vacation etc.

If the difference is large then there is a problem in the backtest - like using too low a figure on slippage or a poor model of the fills one is getting in higher frequency systems.

If you are asking about the back tested returns vs out-of-sample real time returns that can be quite different. The nature of markets and individual instruments change over time so past results do not equal future results. This is easy to see by starting a back-test say in 1985 and run it to 2000 using a basket of commodities with one of the built in systems. Then run the same system and portfolio from 2000-2008. Look at CAGR and MaxDD difference for a feel of what it could have been like real-time to deploy a system in 2000.

In terms of rule of thumbs I have heard expect or plan for a MaxDD that is 1.5x the back-tested version in real-time. So if back testing is showing 20% MaxDD real-time will likely hit 30% sooner or later.

[IRVLLC]

I get about 105% of my stock systems and 88% on my future system. The difference being I take a bigger beating in slippage on the futures. I use 20% of ATR for slippage in my simulations and actual commish costs.