Evaluating the effectiveness of filters in the construction of an ATC

New comment
 

I propose to discuss a difficult topic - evaluating the effectiveness of filters when building an ATS.

There is a global idea in building ATSs - a strategy, for example - to buy when the market is falling, or vice versa - to open positions in the direction of market movement. Suppose the strategy includes basic entry points - let us assume the same MA MAs - in the first case a rebound and in the second case a breakout. Everything is simple and clear, but then we reach an understanding that in such conditions it is better not to trade, or reduce the risk, change the exit point - tactical decisions appear.

I am interested to know who evaluates the efficiency of the filter for their ATS. I am looking for the right approach.

Right now I do the following:

1. Collect statistical data during optimization without filtering

2. Collect data with filters - let's say filter has a number of variables

3. The item is similar to the second one since there are a lot of filtering ideas and it is not right to mix everything in one pile

4. After collecting the data, I estimate the results internally within a filter type - choose the best variant

5. I compare better variants with the benchmark (data without filtering); if there are improvements, I proceed to step 6.

6. Perform optimization with combined filters to find effective variant

7. I review the results within the optimization and select the basic settings for each currency pair.

I do all these steps as a whole for 13 currency pairs - i.e. I evaluate efficiency of the filter by taking into account data from 13 instruments. The analysis is done in terms of buying and selling - separately, in addition, there is a splitting within the selection by a changeable base parameter of ATS itself (taking into account the different volatility of currency pairs within the time range).

 
I forgot to mention that to evaluate the data, after point 2 and 3, the optimisation results are collapsed, i.e. averages, best values and totals are used (oh I don't know how to spell it right, but I like the word).
 
-Aleks-:

I propose to discuss a difficult topic - evaluating the effectiveness of filters when building an ATS.

There is a global idea in building ATSs - a strategy, for example - to buy when the market is falling, or vice versa - to open positions in the direction of market movement. Suppose the strategy includes basic entry points - let us assume the same MA MAs - in the first case a rebound and in the second case a breakout. Everything is simple and clear, but then we reach an understanding that in such conditions it is better not to trade, or reduce the risk, change the exit point - tactical decisions appear.

I am interested to know who evaluates the efficiency of the filter for their ATS. I am looking for the right approach.

Right now I do the following:

1. Collect statistical data during optimization without filtering

2. Collect data with filters - let's say filter has a number of variables

3. The item is similar to the second one since there are a lot of filtering ideas and it is not right to mix everything in one pile

4. After collecting the data, I estimate the results internally within a filter type - choose the best variant

5. I compare better variants with the benchmark (data without filtering); if there are improvements, I proceed to step 6.

6. Perform optimization with combined filters to find effective variant

7. I review the results within the optimization and select the basic settings for each currency pair.

I do all these steps as a whole for 13 currency pairs - i.e. I evaluate efficiency of the filter by taking into account data from 13 instruments. The analysis is done in terms of buying and selling - separately, in addition, there is a splitting within the selection by a variable underlying parameter of ATS itself (taking into account the different volatility of currency pairs within the time range).

I apply filters as a part of theory. I.e., first I develop a theory according to which a regularity appears and I will try to trade with this regularity. Then filters are developed on the basis of this theory to help remove erroneous entry points as they look like a pattern but for some reason it does not work there. You need to assume the conditions under which the pattern does not work and do not trade when the conditions arise. Then I run tests to see how the pattern works and which filters are effective, and where I went wrong. That's about it. I will describe it with a simple example:
I want to open a position on the option and close on the clause. Obviously, to assess risks and observe the profit factor we need the candles to be not smaller than a certain size, otherwise we may lose everything on the spread, even if the pattern is working. The filter here is simple. We take ATR and use it to calculate the probability of entering the market and sufficient volatility. Then a question arises: what period should the indicator have? Then another filter is designed - an algorithm for adjusting indicator period to this regularity. This is how I briefly develop filters for TS.
 
Maxim Romanov:
I apply filters in theory. That is, first I develop a theory according to which a pattern appears. Then filters are developed on the basis of this theory to help remove wrong entry points as they look like a pattern but for some reason it is not working there. You need to assume the conditions under which the pattern does not work and do not trade when the conditions arise. Then I run tests to see how the pattern works and which filters are effective, and where I went wrong. That's about it. I will describe it with a simple example:
I want to open a position on the open and close on the close. Obviously, in order to assess risks and observe the profit factor, the candles need to be at least a certain size, otherwise everything can go down the spread, even if the pattern is working. The filter here is simple. We take ATR and use it to calculate the probability of entering the market and sufficient volatility. Then a question arises: what period should the indicator have? Then another filter is designed - an algorithm for adjusting indicator period to this regularity. This is how I briefly develop filters for TS.

The theory is clear, I think many have the same idea, the question is how to estimate the optimal ATR period, what indicators to use.

Recently I have designed a filter that theoretically should be suitable for ATR, but it turned out that a fixed value works better for 13 currency pairs - such a paradox.

 
-Aleks-:

The theory is clear, I think many have such a way of thinking, the question is how to assess which ATR period is optimal, what indicators to use.

Recently I was making a filter, in theory ATR should be good for it, but it turned out that the fixed value works better for 13 currency pairs - such a paradox.

"Fixed value" is equivalent to ATR with a larger period and a correction factor.
 
-Aleks-:

The theory is clear, I think many have such a way of thinking, the question is how to assess which ATR period is optimal, what indicators to use.

Recently I was making a filter for which theoretically the ATR should be good, but it turned out that the fixed value works better at 13 currency pairs - such a paradox.

From experience: it seems to me that ATR and other methods of volatility determination do not work for determining the trend sections of currencies.

Currencies themselves have a weak trend component, unlike futures or a better stock market.

 
forexman77:

From experience: it seems to me that ATR and other volatility methods for determining the trending areas of currencies do not work.

Currencies themselves have a weak trend component, unlike futures or a better stock market.

How does it not work? It simply measures the size of the candle and averages it out. If I have to measure the size of a candle, I'll just take it and measure it. Trend and flat are all contrivances and have little to do with candle size.
 
Maxim Romanov:
How is it wrong? It just measures the size of the candle and averages it out. If I need to measure the size of a candle, I just take it and measure it. Trend and flat are all contrivances and have little to do with candlestick size.
If I understand correctly, the idea is considered here that if the ATR passes a certain threshold, the trade can be opened? That is, if the ATR is at low values, the price is in a range at the moment.
 
George Merts:
"Fixed value" is equivalent to ATR with a larger period and a correction factor.
Not exactly, I took ATR for days 3 and 100 (I tried 50% and 61.8% from them) - 100 showed better of course, which says more about static deviation, but this ATR(100) will be different for different pairs, and the fixed value of 90 points for all pairs turned out to be more effective, which surprised me.
 
forexman77:

From experience: it seems to me that ATR and other volatility methods for determining the trending areas of currencies do not work.

Currencies themselves have a weak trend component, unlike futures or a better stock market.

ATR is not bad at showing likely limits (meaningful levels) - I trade on M15 and ATR use daily to make tactical decisions intraday - Fibonacci coefficients are used.

 
forexman77:
If I understand correctly the idea touched upon here is that if ATR passes a certain threshold, then a trade can be opened? That is, if the ATR is on the low side, then at the moment the price is in a range.
This was a simple filter example, as part of some abstract strategy. I haven't said anything about trends and bands because these concepts are just as abstract without any description of a specific strategy. There are no trends and floats. Each trend is a part of a flat on a larger horizon and vice versa. My point is that the development of a filter requires a complex approach and it has to be created based on the tasks that the system solves. I did not say that I have thrown the averages, then I threw the oscillators, looked which ones do not work and threw the ones that do not work.
12345678
New comment