The Magic of Filtration

Introduction

Most of the developers of automated trading systems (ATS), one way or another, use some form of signals filtering. Although it isn't the only way to improve the system characteristics, it is considered most effective. Beginner "grail-builders" often fall to the magic of the filters. It's very simple to take some trading strategy, hang a dozen of filters on it, and here it is, - a profitable Expert Advisor.

However, there are opponents of the use of filters. Filters significantly (sometimes 2-3 times) reduce the number of deals, and there is no guarantee that in the future they will be as effective as in the past. Certainly, there are also some other compelling reasons.

So let us take a further look and consider all of this one step at a time.

The hypothesis of the meaninglessness of filtering

If the Expert Advisor is unprofitable ("drainer"), it is unlikely that some type of filtration will help,- the magic of filtering is powerless here.

Therefore, in this article, these Expert Advisors will not be considered. Although, we must know that there are studies of quantum-frequency filters, capable of turning virtually any "drainer" into a pseudo-Grail.

Hypothesis of the dangers of filtering

If the Expert Advisor in its characteristics is approaching an ideal Automated Trading System, then filtering will only worsen it.

We should clarify what is meant by an ideal automated trading system. By this is meant such a trading strategy, which generates only profitable transaction, ie does not bring any losses. In such a system, the number of unprofitable trades = 0.

What is a filter?

In its simplest form, a trading signal filter is a logical restriction such as: if A is not less than B (A> = B), then the signal is skipped, and if it is smaller (A

An example.There may be a correlation between the results of ATS trading and atmospheric pressure in the village of Kukushkino, ie You can create an appropriate filter and improve the profitability of the Expert Advisor, which will take into account the weather in this small Russian town. However, it is unlikely that someone would appreciate such an innovative approach to filtering, in spite of the fact that it could raise the profitablity of the system.

Classification of filters

Although there is a great variety of filters used in ATS, they can still be divided into two main classes:

• bandpass filter (P-filter) - transmits a band of signals;
  
• discrete filter (D-filter) - selective transmission of signals by the mask (template).

Where to start?

Let's consider the filtering mechanism by looking at an example. let's use the DC2008_revers Expert Advisor (see attached file), which was specially developed for this article, and explore its features (without using any filters). Here is the report obtained during testing.

Report

Symbol		EURUSD (Euro vs US Dollar)
Period		1 Minute (M1) 2009.09.07 00:00 - 2010.01.26 23:59 (2009.09.07 - 2010.01.27)
Model		All ticks (the most accurate method based on all of the least available timeframes)
Parameters		Lots = 0.1; Symb.magic = 9001; nORD.Buy = 5; nORD.Sell = 5;
Bars in the history	133967	Modeled ticks	900848	Quality of Modeling	25.00%
Mismatched traffic errors	0
Initial deposit	10000.00
Net profit	4408.91	Gross profit	32827.16	Gross loss	-28418.25
Profitability	1.16	Expected payoff	1.34
Absolute drawdown	273.17	Maximum drawdown	3001.74 (20.36%)	Relative drawdown	20.36% (3001.74)
Total transactions	3284	Short positions (won%)	1699 (64.98%)	Long positions (won%)	1585 (65.68%)
		Profitable trades (% of total)	2145 (65.32%)	Losing trades (% of total)	1139 (34.68%)
Largest		profitable trade	82.00	losing trade	-211.00
Average		profitable trade	15.30	losing trade	-24.95
Maximum Number		consecutive wins (profit)	29 (679.00)	consecutive losses (loss)	16 (-290.34)
Maximum		consecutive profit (count of wins)	679.00 (29)	consecutive loss (count of losses)	-1011.00 (10)
Average		consecutive win	5	consecutive loss