- Random Flow Theory and FOREX
- Requests & Ideas
- Experiments ...
1) The use of linear regression analysis only makes sense with a statistically significant datapool. Hedge funds like to use it in all kinds of financial markets, but only as a corroborative indicator and not as a be-all-end-all trading tool.
2) Well, since you're attempting to model the relationship between two variables, with a given collection of data values, in order to find the "best fit" between the two variables, you'd somehow need to make sure that your data isn't skewed by statistical outliers. Work with median values and avoid averages.
3) What you're looking for is manual backtesting software, but I've already written in another forum post why that is usually out of the question for retail traders. https://www.mql5.com/en/forum/433683#comment_42379223
Even coding it as an EA and testing it via the inbuilt strategy tester of Metatrader will give you misleading data because you're working with In-Sample datasets. <Deleted<
4) It's hard to give a correct answer to that. You need to elaborate on the exact method of your strategy and how you want to implement it. Another problem that stands in your way of receiving a good answer to this problem (either from members here or your own intellect) is how you define the beginning and end of a trend. It's easy to claim when a trend started and ended AFTER the fact, but for this strategy to reliably work, you'd need to be able to see the signs before the whole thing actually starts to happen.
What I'm trying to say is, that this matter is extremely complicated and that you need a large enough sample size to improve the statistical significance of your metrics.
- 2022.09.29
- www.mql5.com
1) The use of linear regression analysis only makes sense with a statistically significant datapool. Hedge funds like to use it in all kinds of financial markets, but only as a corroborative indicator and not as a be-all-end-all trading tool.
2) Well, since you're attempting to model the relationship between two variables, with a given collection of data values, in order to find the "best fit" between the two variables, you'd somehow need to make sure that your data isn't skewed by statistical outliers. Work with median values and avoid averages.
3) What you're looking for is manual backtesting software, but I've already written in another forum post why that is usually out of the question for retail traders. https://www.mql5.com/en/forum/433683#comment_42379223
Even coding it as an EA and testing it via the inbuilt strategy tester of Metatrader will give you misleading data because you're working with In-Sample datasets. I have also written a blog post about this issue.
4) It's hard to give a correct answer to that. You need to elaborate on the exact method of your strategy and how you want to implement it. Another problem that stands in your way of receiving a good answer to this problem (either from members here or your own intellect) is how you define the beginning and end of a trend. It's easy to claim when a trend started and ended AFTER the fact, but for this strategy to reliably work, you'd need to be able to see the signs before the whole thing actually starts to happen.
What I'm trying to say is, that this matter is extremely complicated and that you need a large enough sample size to improve the statistical significance of your metrics.
Linear regression has been used before, for example in "Least Squares Moving Average (LSMA)" or "Endpoint Moving Average (EPMA)", but maybe not in the way you envision it.
So, if you say that your approach of repeated regression has not been attempted before, then probably no one is going to be able to answer your queries and you will have to experiment for yourself to find out.
Yeah I completely misunderstood you. I thought you were talking about regression analysis as in "regression channel" and thought your strategy is based on that. Sorry for wasting your time. Perhaps a member with proficiency in robust statistics can answer your question.
Linear Regression is the mathematical model. You can calculate the variables you are looking for in different ways. One of the standard ways is OLS (ordinary least square), because you need no assumptions for this. If you use Linear Regression with OLS (which - i think - most of regression channel etc. indicators do), than you can repeat the calculation and you will get the same numbers as before.
Repeating the calculation and getting better variables only works by a filter and/or smother-approach. I think you could transfer the Shumway-and-Stoffer-Smother-Approach and the Kalman-Filter for the Ornstein-Uhlenbeck-Process to the Linear-Regression-Model. But I think you would waste your time. Using a smother and filter for the OU-Process, especially the trending OU-Process, should be more efficient.
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use