.
.
.
ATTENTION: THIS THREAD IS OUTDATED
Please go to http://datfra.com/ in order to find my current research [ Strategy Builder, EA Collection, Papertrading ]!
WHAT?
The most in-depth analysis of technical indicators and expert advisors you will ever see (based on 1-10 million "backtests")
This analysis essays will show you what parameters and market conditions work best for the analysed systems - instead of relying on the default parameters for indicators, for example, you will be able to use the empirically best ones.
WHY?
To alpha-test the algorithmic trading framework I am working on!
I will analyse public Indicators and free EAs with it on a regular basic (every 1-3 days) and write down everything in a blog:
http://darwins-fx-research.com/
So, to be updated, subscribe to me or keep an eye on my blog!
HOW?
Because backtests are unreliable and therefore mostly useless, I am using a "Walk Forward Analysis"-like testing approach, and to circumvent the shortcomings of it I am using parameterspace analysis.
That means, I do analyse the indicators/EAs based on ~1-10 million "backtests" and do not analyse performance 'on some history data', but how performance during optimisation (past) and forward trading (future) is related, just like in real trading!
"Please analyse ***"
My framework builds upon Metatrader4, so I can analyse most indicators and Expert Advisors for it. If you want some particular things on some particular market to be analysed, add me on Skype: darwin-fx.
--------------------------------------------------------------------------------------------------------------------------------------
tl;dr
The "averaging method" and "pricefield" parameters are almost completly unimportant. The D line period is also, however, there is a small optimum around 15-45.
But the slowing parameter should be >= 6 and the K-Parameter between 13-25 to have the highest chance of valuable signals for this indicator!
How this analysis is done
I created a very simple Expert Advisor for this essay: It is always in the markets, and changes position when a Stochastic Crossover happens. Thats it! That way, no money- or riskmanagement biases the analysis.
Here you can find the documentation: http://docs.mql4.com/indicators/istochastic
This analysis is based on 2.600.000 in-sample&out-of-sample windows (or "5.000.000 backtests") on EURUSD/H4 (01.01.2006 - 01.01.2014)
You can find the Data I used for this analysis here, and the program I used here (free & without registration)
For the professionals among us: This is based on an in-sample&out-of-sample method, and I did first remove all datapoints with in-sample profit < 0$, and then analyse out-of-sample profit. In-sample timespan was 6 months, out-of-sample timespan was 14 days.
Note: The X-Axis is the parameter value, the Y-Axis is the profit. As we are dealing with so many datapoints, I did bucket them together in bars: In the table on the right, you can read what parameter-values the single bars stand for (first row=bar index, second row=parameter)
------------------------------------------------------------------------------------------
The pricefield indicator can be 0 or 1. Look in the table at the right, you see that both values have a profit of 11.* => COMPLETLY UNIMPORTANT
http://darwins-fx-research.com/image...PRICEFIELD.png
------------------------------------------------------------------------------------------
The averaging method can be 0-3. They are all quite the same, but there is a small optimum in the 3rd bar (the one you cant see due to gnuplot fucking scaling up) => 2 ("Smoothed averaging") is best, but is unimportant
http://darwins-fx-research.com/image...sd_h4/TYPE.png
------------------------------------------------------------------------------------------
D line period was evaluated from 1-100. You can see that there is an optimum (well at least its not so bad there) around the 7th - 19th bar => 15-45 works best, but is unimportant, as long as its not below 15
http://darwins-fx-research.com/image...urusd_h4/D.png
------------------------------------------------------------------------------------------
Slowing was evaluated from 1-6. Here it gets a bit more interesting, as we can see a clear trend in the last 5 bars. And a slowing of 6 even shows a positive expectancy (i should have evaluated the ones above 6, also. Sorry) => A slowing >= 6 is recommended
http://darwins-fx-research.com/image...h4/SLOWING.png
------------------------------------------------------------------------------------------
K line period was evaluated from 1-100. This is clearly the most important parameter for stochastic indicator! And if we choose it between 13-25, we have a good positive expectancy for the indicators signals! => 13-25 is the optimum. Most important parameter!
http://darwins-fx-research.com/image...urusd_h4/K.png
------------------------------------------------------------------------------------------
How do the 2 most important ones, K and slowing, play together?
You can see that a higher "slowing" value (Y-Axis) does not neccessarily increase expectancy of the signals a lot, but it makes a broader K-Parameter-range 'profitable', which increases the robustness of the signals
http://darwins-fx-research.com/image..._h4/3dmapx.png
------------------------------------------------------------------------------------------
.
.
ATTENTION: THIS THREAD IS OUTDATED
Please go to http://datfra.com/ in order to find my current research [ Strategy Builder, EA Collection, Papertrading ]!
WHAT?
The most in-depth analysis of technical indicators and expert advisors you will ever see (based on 1-10 million "backtests")
This analysis essays will show you what parameters and market conditions work best for the analysed systems - instead of relying on the default parameters for indicators, for example, you will be able to use the empirically best ones.
WHY?
To alpha-test the algorithmic trading framework I am working on!
I will analyse public Indicators and free EAs with it on a regular basic (every 1-3 days) and write down everything in a blog:
http://darwins-fx-research.com/
So, to be updated, subscribe to me or keep an eye on my blog!
HOW?
Because backtests are unreliable and therefore mostly useless, I am using a "Walk Forward Analysis"-like testing approach, and to circumvent the shortcomings of it I am using parameterspace analysis.
That means, I do analyse the indicators/EAs based on ~1-10 million "backtests" and do not analyse performance 'on some history data', but how performance during optimisation (past) and forward trading (future) is related, just like in real trading!
"Please analyse ***"
My framework builds upon Metatrader4, so I can analyse most indicators and Expert Advisors for it. If you want some particular things on some particular market to be analysed, add me on Skype: darwin-fx.
--------------------------------------------------------------------------------------------------------------------------------------
tl;dr
The "averaging method" and "pricefield" parameters are almost completly unimportant. The D line period is also, however, there is a small optimum around 15-45.
But the slowing parameter should be >= 6 and the K-Parameter between 13-25 to have the highest chance of valuable signals for this indicator!
How this analysis is done
I created a very simple Expert Advisor for this essay: It is always in the markets, and changes position when a Stochastic Crossover happens. Thats it! That way, no money- or riskmanagement biases the analysis.
Here you can find the documentation: http://docs.mql4.com/indicators/istochastic
This analysis is based on 2.600.000 in-sample&out-of-sample windows (or "5.000.000 backtests") on EURUSD/H4 (01.01.2006 - 01.01.2014)
You can find the Data I used for this analysis here, and the program I used here (free & without registration)
For the professionals among us: This is based on an in-sample&out-of-sample method, and I did first remove all datapoints with in-sample profit < 0$, and then analyse out-of-sample profit. In-sample timespan was 6 months, out-of-sample timespan was 14 days.
Note: The X-Axis is the parameter value, the Y-Axis is the profit. As we are dealing with so many datapoints, I did bucket them together in bars: In the table on the right, you can read what parameter-values the single bars stand for (first row=bar index, second row=parameter)
------------------------------------------------------------------------------------------
The pricefield indicator can be 0 or 1. Look in the table at the right, you see that both values have a profit of 11.* => COMPLETLY UNIMPORTANT
http://darwins-fx-research.com/image...PRICEFIELD.png
------------------------------------------------------------------------------------------
The averaging method can be 0-3. They are all quite the same, but there is a small optimum in the 3rd bar (the one you cant see due to gnuplot fucking scaling up) => 2 ("Smoothed averaging") is best, but is unimportant
http://darwins-fx-research.com/image...sd_h4/TYPE.png
------------------------------------------------------------------------------------------
D line period was evaluated from 1-100. You can see that there is an optimum (well at least its not so bad there) around the 7th - 19th bar => 15-45 works best, but is unimportant, as long as its not below 15
http://darwins-fx-research.com/image...urusd_h4/D.png
------------------------------------------------------------------------------------------
Slowing was evaluated from 1-6. Here it gets a bit more interesting, as we can see a clear trend in the last 5 bars. And a slowing of 6 even shows a positive expectancy (i should have evaluated the ones above 6, also. Sorry) => A slowing >= 6 is recommended
http://darwins-fx-research.com/image...h4/SLOWING.png
------------------------------------------------------------------------------------------
K line period was evaluated from 1-100. This is clearly the most important parameter for stochastic indicator! And if we choose it between 13-25, we have a good positive expectancy for the indicators signals! => 13-25 is the optimum. Most important parameter!
http://darwins-fx-research.com/image...urusd_h4/K.png
------------------------------------------------------------------------------------------
How do the 2 most important ones, K and slowing, play together?
You can see that a higher "slowing" value (Y-Axis) does not neccessarily increase expectancy of the signals a lot, but it makes a broader K-Parameter-range 'profitable', which increases the robustness of the signals
http://darwins-fx-research.com/image..._h4/3dmapx.png
------------------------------------------------------------------------------------------