Forward Testing
Many traders are familiar with the concept of back testing their trading strategy, but often less thought is given to forward testing.
Back testing is straightforward. You take historical tables or charts, pretend that you had been trading at that time using the strategy you are testing, and you see if the results turn out as a profit. Usually this is done with a computer program, in order to deal with the mass of data, and many charting programs include this facility. You can use back testing to refine any variable parameters in your strategy, such as the number of days in a simple moving average, to see what gives you the best result.
This technique seems sensible if not obvious. There can be issues with it however, and using forward testing will help overcome those issues.
One of the main problems is that you can over refine the variables, tweaking them for the absolute maximum, and unknowingly create a strategy that is very specific for the actual historic data, and which fails to accommodate the variability of future price movements. By concentrating on one particular sample of data, you risk making your strategy too limited in its applicability.
This is where forward testing, sometimes called walk-forward testing, can restore some realism to your strategy. The technique is easy to understand. When you set up your back testing, you only use a part of the historical data available to you, and reserve another part for the forward test. The data you have used for back testing is referred to as in-sample data, and the reserved section is out-of-sample data.
Once you have done whatever you wish to your strategy, back testing and optimizing it with the in-sample data, it is time for you to do forward testing. You take your optimized strategy, and in a similar way to back testing you set your computer software to run the strategy on the out-of-sample data. The purpose of this is to validate the strategy works with different market values.
Depending on the results, you may choose to adopt this tested strategy in your trading, or realize that you have over optimized and go back to basic development. Whatever you do, you should not try and tweak the variables on the basis of the forward test, because you would simply be optimizing again on a different set of figures, with all the associated consequential problems.
Some computer software is configured to make forward testing simple. It allows you to choose what proportion of your data you will apply to the back test, and reserves the rest for your subsequent test. Some software will even allow you to step through the years, starting at different dates and optimizing the strategy for a couple of years, with the third year applied to forward testing. Performing this task successively is one that is only possible because of computers, and it gives a thorough analysis of the performance of your strategy.
There is another type of forward testing which you may be advised to do before going live with a real account. That is to test truly forward in time, opening a paper or demonstration account and applying your strategy over the coming weeks or months to check on its performance. Many traders choose to forgo this additional check, having spent a significant amount of time developing the strategy in the first place and wanting to get into real trading. It is important that you incorporate some type of forward testing to verify your strategy before going live, and splitting historic data into in-sample and out-of-sample batches is a good way to do this.
Join the discussion