In one of the previous posts Ive introduced the statistical Forex system definition and marked up the important problems that should be solved in the process of its creation. Here I will try to explain more about the problem of the statistics gathering.
Choosing the timeframe for the statistics that will be gathered for your system consists of two parts — choosing the chart timeframe discreteness and choosing the length of the period over which the statistics will be gathered.
Choosing the right chart timeframe is a matter of balance between the uninformative short-term data with many samples and the small quantity of the more specific information. If you choose a tick or 1-minute timeframe discreteness youll have to face a large amount of data and a lot of CPU power used up on gathering, calculating and finding this data; finding some patterns in these vast arrays of information wouldnt be an easy task and the further strategy building will very complicated in this case. On the other hand, using daily or weekly periods will give you too little information. For example, one year of the market analysis of D1 charts will give you just a little more than 200 data samples. In my opinion, considering the 24 hours a day and 5 days a week nature of the Forex market, the best choices here are M30, H1 or H4 timeframes as they give you a fair amount of samples with a decent informativity, because such samples will have a greater variation. Alternatively you can use multi-timeframe statistics, but that would lead to a really complex system, which, of course, will have a better potential.
Sampling periods length is an important parameter of the statistics gathering. Using a small period will allow you to recognize the most up-to-date rate patterns and your strategy will probably benefit from them in the short to medium term. Unfortunately, short sampling period can contain too little of these patterns and if the market changes they will probably fail to help with the recognition of the changed price dependencies. Long sampling period will give a very wide array of patterns which can be used in comparing. But the difference between the market today and the market several years ago can bee too large, so those patterns can lead your system to a high inaccuracy ratio. Getting statistics over the past 2-3 years is a balanced decision here. You catch more than one long-term trend and you get a lot of the medium- and short-term trends caught into your statistics with such period, while really outdated data isnt spoiling your statistics.
Of course, these decisions should also depend on your system, the nature of the data you will be collecting and the timeframe that it will use in the actual trading. But dont forget the negative and positive sides of the different data timeframe and the sampling periods — try to avoid the extreme values that could possibly ruin your strategy.
