Discussion of article "Regression Analysis of the Influence of Macroeconomic Data on Currency Prices Fluctuation"
It is not clear on what basis you think you can trust the results obtained?
After all, it is known that regression analysis has such significant limitations that practically exclude the possibility of its use in financial markets.
Therefore, it is necessary to prove that the obtained results can be trusted.
The result of any analysis is the creation of a probabilistic picture of the situation development. Of course, the result obtained from the equation is not a 100% guaranteed forecast in such a picture, because, for example, it lacks a mechanism to take into account the influence of politicians' speeches, force majeure, results of meetings, etc. But I think it can give a probabilistic assessment of changes in the price of a currency for a small period.
Secondly, the regression equation is the same technical indicator that facilitates the trader's life, and whether to believe its readings or not depends on the trader himself.
Thirdly, you are right to question the fact that the results can be trusted, requires proof, but in my article I have outlined the algorithm of actions in pictures, attached a script for generating tables to export the news feed in the form of a table for processing, use, check the article in practice, collectively and decide whether this is applicable in the financial markets or requires revision or not applicable.
The result of any analysis is the creation of a probabilistic picture of the situation development. Of course, the result obtained from the equation is not a 100% guaranteed forecast in such a picture, because, for example, it lacks a mechanism to take into account the influence of politicians' speeches, force majeure, the results of meetings, etc. But it can give a probabilistic assessment of the change in the price of a currency for a small period, I think, it can.
Secondly, the regression equation is the same technical indicator that facilitates the trader's life, and whether to believe its readings or not depends on the trader himself.
Thirdly, you are right to question the fact that the results can be trusted, requires proof, but in my article I have outlined the algorithm of actions in pictures, attached a script for generating tables to export the news feed in the form of a table for processing, use it, check the article in practice, collectively and decide whether it is applicable in the financial markets or requires revision or not applicable.
The problem you have raised is a cornerstone of data mining. This problem is most elaborated in the CORELearn package. Here is a link to the documentation. The most widespread, in the sense of in many packages, is the Gini index. The most promising index is Relief with its many modifications.
Do not leave your research, as you have raised a basic trading question.
Good luck.
...
Thank you.
That's funny.
First you built a model based on the analysis of a certain period, and then you proved that the market reaction to an event from the same period demonstrates the expected behaviour. Well, it is understandable, the model already takes this event into account. It's a classic fit. What is even more amusing is that the reaction to news releases is rarely measured even in hours, let alone days.
Perhaps if you do the same for short-term intervals, the result will be more close to life and it will have a chance to pass forward tests successfully
That's funny.
First you built a model based on the analysis of a certain period, and then you proved that the market reaction to an event from the same period demonstrates the expected behaviour. Well, it is understandable, the model already takes this event into account. It's a classic fit. What is even more amusing is that the reaction to news releases is rarely measured even in hours, let alone days.
Perhaps if you do the same for short-term intervals, the result will be more life-like and it will have a chance to succeed in forward tests
The article is titled "Analysis" and it does not talk about forecasting, hence the forward test.
If we talk about your remark, then without substantiating the applicability of forward tests, making them is an empty occupation and very dangerous, because you can accidentally get a satisfactory result of a forward test and believe your model, and it will safely lose in real trading.
The article is titled "Analysis" and it does not speak about forecasting, hence about forward test.
If we are talking about your remark, then without substantiating the applicability of forward tests, making them is an empty occupation and very dangerous, because you may accidentally get a satisfactory result of a forward test and believe your model, and it will safely lose in real trading.
I can't judge the article by the title alone. The topic is interesting to me and I read it in full. And here is what the author writes in the conclusion:
However, I would like to draw your attention to the fact that the forecast is not a 100% guarantee that the currency will go exactly in the predicted direction. The result of the forecast is a probabilistic event, the occurrence of which depends on many factors. Besides, it is recommended to periodically check the regression equation when new data comes in.
Good luck with your predictions.
Just juggling data is no fun. The author builds a model of the market described by regression equations, and then proves that the model sort of works, while using validation on the data used to build it. This is exactly what I pointed out as some sort of fudging.
In my time I had a lot of trouble with fundamental analysis and tried to automate it for a long time. The proposed FA method facilitates it quite a lot. Analysis of macroeconomic variables helps to identify those that could influence price changes. But such analysis does not give a 100% guarantee that they will also be relevant in the future. Remember: past trading results cannot be a guarantee that it will be the same in the future.
Checking the equation on future data can be done by means of the programme. You can also do it yourself. In the process of data preparation, limit the extreme date using the button "Select Cases" (see Fig. 13 of the article), in the opened window enable the checkbox "Enable Selection Conditions", below select "Specific, selected by:" and in the box write down the interval of lines to take into account "v0>0 and v0<999". Rows further than 999 will not be considered. After the analysis and selection of significant variables, go back to this window and change, move the interval forward starting from row 1000 and check the results again. The % accuracy will be shown in the matrix in the programme.
This equation algorithm should be a part of a trading robot, an Expert Advisor. Unfortunately, everything is not fully automated, it is difficult to collect data from the sites that post calendars, the same indicator can be written differently, a word can make a mistake, etc. because the periods are not short-term.
I do not insist that this method is manna from heaven, I just offer you an additional tool that can make your work easier and save you time.
In my time I had a lot of trouble with fundamental analysis and tried to automate it for a long time. The proposed FA method facilitates it quite a lot. Analysis of macroeconomic variables helps to identify those that could influence price changes. But such analysis does not give a 100% guarantee that they will also be relevant in the future. Remember: past trading results cannot be a guarantee that it will be the same in the future.
Checking the equation on future data can be done by means of the programme. You can also do it yourself. In the process of data preparation, limit the extreme date using the button "Select Cases" (see Fig. 13 of the article), in the opened window enable the checkbox "Enable Selection Conditions", below select "Specific, selected by:" and in the box write down the interval of lines to take into account "v0>0 and v0<999". Rows further than 999 will not be considered. After the analysis and selection of significant variables, go back to this window and change, move the interval forward starting from row 1000 and check the results again. The % accuracy will be shown in the matrix in the programme.
This equation algorithm should be a part of a trading robot, an Expert Advisor. Unfortunately, everything is not fully automated, it is difficult to collect data from the sites that post calendars, the same indicator can be written differently, a word can make a mistake, etc. because the periods are not short-term.
I do not insist that this method is manna from heaven, I just offer you an additional tool that can make your work easier and save your time.
I cannot agree with you in principle.
Your model does not provide any information at all - it misleads people. And here's why.
1. Regression models are applicable to stationary time series. That is why there are ARIMA, ARCH and a lot of other models, which before building a model try to transform the original time series into some other one, but which will somehow resemble a stationary series (mo and variance are equal to a constant).
2. After converting the original series to resemble stationarity, a regression model is built for the sole purpose of looking at the difference between the fit from the model and the real data. If this difference (residual - returns) is stationary, then the next step can be done.
3. If the results of the first two steps are positive, what Vladix wrote about - forward tests. If the results obtained on the model training dataset are close to the results on the test and validation datasets (these are three different pieces of the time series), then only then we can start to talk about trusting the obtained results. In the opposite case - in no case, it will be a very dangerous self-deception, blind faith in numbers.
The whole problem with the first two steps is that they cannot be fulfilled. The main and most nasty obstacle is the correlation between the dependent variables - multicollinearity. There are other problems as well. This explains that the problem you have raised - determining the influence of dependent variables on the independent variable - is attempted to be solved with special indicators, e.g. Gini, Relief.
With your article you have touched upon the basic problem of building trading systems - selection of initial data for models. In terms of labour intensity, it is at least half of the time, if not all 75%. Even in the way you have done it, it is very important for understanding the basic problems of trading.
...I can't agree with you on principle. ...
All right, don't agree. Until other approaches are found, I propose to use this one ))))
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
New article Regression Analysis of the Influence of Macroeconomic Data on Currency Prices Fluctuation has been published:
Fundamental analysis is deemed incomprehensible by many. It is unclear how to carry it out, which parameters to take into account and which not to. Finding out the impact of the accounted parameters and the length of time it is to be considered for is not simple either.
In 2011 I came across the article Multiple Regression Analysis. Strategy Generator and Tester in One and found the method described there interesting. I have conducted research on the application of this method to the fundamental analysis and describe the results in this article.
Author: Salavat Bulyakarov