Your trading system needs to be robust enough to stand up to different market conditions and events but at the same time it needs to be focused enough to deliver decent profits.
It’s important to decide upon a metric that you can use to measure the system performance objectively. Relying on the CAR (compound annual return) figure is not always a good idea because this metric does not take into account the risk that was involved in producing those gains. It’s a better idea to look at CAR/ MDD or the risk/reward ratio itself.
Guest post by FXTM
4 – In-sample testing
The optimising of the system needs to be run on in-sample data which means you need to freeze a certain amount of data first before you begin testing.
It’s a good idea to set aside at least a couple of years of data for in-sample testing, although if you’re looking for an intraday system you can certainly get away with less.
Whatever the amount of data at your disposal it’s important to separate it out into in-sample and out-of-sample. The in-sample will be used to do all your optimizing and to find a system that produces good profits. The out-of-sample will be kept aside for later on.
5 – Out-of-sample testing
Once the in-sample testing reveals a trading system that produces good profits over a number of different settings and date ranges, it’s time to test it on the more recent out-of-sample data. This data is clean and has been set aside to objectively determine whether the system will work in the future or not.
A system must be run on out-of-sample data only once. After the first run the data becomes contaminated. That system, with any minor adjustments can never be run on the same data again as the result will be statistically worthless.
If you run the system on out-of-sample data and it performs similarly to your in-sample tests, it’s a good sign that the system is robust. However it doesn’t perform as well, you’ll have to throw it away and start again.
6 – Further tests
If you have got to step 5 and found a system that performed nearly as well on out-of-sample data as in-sample then you have found a good system but the work doesn’t stop there. You can test it over different markets and perform walk-forward tests. You can also run statistical tests on the results such as Monte Carlo analysis and evaluate the metric results of the systems. The individual metric results from the back-tests such as CAR, risk/reward, payoff ratio, average win, expectancy etc can often reveal some important insights into the system itself.