Discussion of article "Using Discriminant Analysis to Develop Trading Systems" - page 2
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
In general, I approximately got a similar result - there is no way to reject the null hypothesis that the considered coefficients are equal to zero.....
Yes, that's the main point.
Discriminant analysis is a method of mathematical statistics. Its sister econometrics does not use this analysis for some reason. Anyway, I can't remember why? The result of the article is interesting.
In general, I got approximately a similar result - there is no way to reject the null hypothesis that the considered coefficients are equal to zero....
may I ask a few questions?
1. How many bars, what currency pair and what timeframe were used for analysis?
2. How do the coefficients behave if we take a completely different piece of history? How do the coefficients behave if we increase the number of analysed bars by 10 times?
3. How many bars should be minimally taken for YES to reject or confirm the null hypothesis? Is there such a number of bars in the history?
4. How true is the assumption that the sought coefficients are eternal constants and do not change with time?
5. How long did the YES take to run on your computer?
can I ask you a few questions?
1. How many bars, which currency pair and which timeframe were used for analysis?
2. How do the coefficients behave if we take a completely different piece of history? How do the coefficients behave if we increase the number of analysed bars by 10 times?
3. How many bars should be minimally taken for YES to reject or confirm the null hypothesis? Is there such a number of bars in the history?
4. How true is the assumption that the sought coefficients are eternal constants and do not change with time?
5. How long did the YES take on your computer?
1. EURUSD. The article states that: data was collected using a strategy tester from 1 August 2011 to 1 October 2011 on the H1 period.
2. In the example of the article, 70% of the sample was used for training and 30% for testing. In the test plot the coefficients retained their significance. As for other plots - try it yourself.
3. The number of bars should be at least 5 times more than the number of indicators tested to build the model. The more bars, the more reliable the model.
4. You may be able to find such indicators and coefficients for them :-) This was not the goal of the article.
5. We are talking about milliseconds.
Reflections in the comments about the null hypothesis are not related to discriminant analysis. Faa1947 writes about regression analysis, which is not the same thing.
Just another way of subtle tinkering.
Agreed. Discriminant analysis, neural networks, self-learning programs, Kohonen maps, etc. are just tools of analysis, but we need a forecast. A market model is needed for forecasting. Until there is no market model, it is inefficient to use analysis tools.
Thus, in the discussed article, the market model was sewed into several indicators taken for analysis. This model describes the market poorly, which manifested itself in the high probability of coefficients equality to zero.
There is no sense to propose and discuss tools without a market model.
Agreed. Discriminant analysis, neural networks, self-learning programs, Kohonen maps, etc. are just analysis tools, but we need a forecast. A market model is needed for forecasting. Until there is no market model, it is inefficient to use analysis tools.
Thus, in the discussed article, the market model was sewed into several indicators taken for analysis. This model describes the market poorly, which manifested itself in the high probability of coefficients equality to zero.
There is no sense to propose and discuss instruments without a market model.
The probability of zero equality mentioned in the comments has no relation to this model :-)
In general, this is a holy war argument - whether technical analysis works or not. Someone successfully trades using only technical analysis, others prefer fundamental analysis, and some prefer both.
According to the systems theory: it is not necessary to know how a system works in order to predict its behaviour. It is doubtful that a market model for forex is even possible. The appearance of any model will immediately be taken into account in the price according to technical analysis :-)
The probability of zero mentioned in the comments has nothing to do with this model :-)
In general, this is a holy war argument - whether technical analysis works or not. Someone successfully trades using only technical analysis, others prefer fundamental analysis, and some prefer both.
According to the systems theory: it is not necessary to know how a system works in order to predict its behaviour. It is doubtful that a market model for forex is even possible. The appearance of any model will immediately be factored into the price according to technical analysis :-)
There are many forex market models. Here are some of them: 1. The price wanders randomly. 2. Price is a smooth function with added noise. 3. There are sometimes trends in price movement. 4. A lot of models from economics and fundamental analysis.
To predict the behaviour of a system, you don't even need to know its structure. It is possible, for example, to transfer the history of the system to the future. But the transfer rule will be the model of the system.
Hi,
Thank you very much for producing this article! I am ever so grateful. I was wondering where to find a software like this! So what if I like a certain setup already but I want to improve my win rate? How do I use the DA Statistica for this?
I have used some neural network and Genetic Programming software. They are able to build strategies towards a user's goals - win rate, Net Profit, Frequency of trades, E(r), Profit Factor, etc. Is it possible to do that with this software? Can you show how if you can?
StatSoft Statistica was used for the DA analysis.
I believe, other goals can be reached too. However you should think out how they can be included in your DA model. Just as an idea, if you trade on H1 time frame, you may try to forecast your win rate by H4 or D1 indicator data.
The DA for H4 timeframe is carried out in the same way as for H1.