Discussing the article: "Forecasting exchange rates using classic machine learning methods: Logit and Probit models"

 

Check out the new article: Forecasting exchange rates using classic machine learning methods: Logit and Probit models.

In the article, an attempt is made to build a trading EA for predicting exchange rate quotes. The algorithm is based on classical classification models - logistic and probit regression. The likelihood ratio criterion is used as a filter for trading signals.

Financial market researchers will always face the difficult task of choosing a mathematical model to predict the future behavior of trading instruments. To date, a huge number of such models have been developed. So the question arises: how not to drown in all this variety of methods and approaches, where to start and which models are best to focus on, especially if you are just starting to forecast using machine learning models? If we try to reduce the forecasting task to a simple answer to the question - "will the tomorrow's closing price be higher than the ones of today?", then the logical choice would be binary classification models. Some of the simplest and most widely used are logit and probit regression. These models belong to the most common form of machine learning, so-called supervised learning.

The task of supervised learning, in turn, is to teach our model to display a set of inputs {x} (predictors or features) into a set of outputs {y} (targets or labels). Here we will predict only two market conditions - the rise or fall of the currency pair price. Therefore, we will have only two classes of labels y∊ {1,0}. Price patterns, namely standardized price increments with a certain lag, will act as predictors. This data will form our {x, y} training set to be used to estimate the parameters of our models. The predictive model based on trained classifiers is implemented as LogitExpert EA.


Author: Evgeniy Chernish

 
Where is the prediction, in the sense of getting a better effect than random trading?
 
Stanislav Korotky #:
Where is the forecasting, in the sense of getting a better effect than random trading?
All questions to his majesty forex market and efficient market hypothesis.
 

Thanks, good interesting article.

Imho, you can already try to use fundamental data on daytrips. This is not in the sense of criticising the article, but as a way of thinking. I wonder how macroeconomic data can be adequately "mixed" with price data. The problem is their rare change, for example. Probably, macroeconomics can also be used somehow in price preprocessing - transition from nominal to real exchange rates, for example.

 
Aleksey Nikolayev #:

Imho, you can already try to use fundamental data on daytrips. This is not in the sense of criticising the article, but as a way of thinking. I wonder how macroeconomic data can be adequately "mixed" with price data. The problem is their rare change, for example. Probably, macroeconomics can also be used somehow in price preprocessing - transition from nominal to real exchange rates, for example.

There is an inbuilt news calendar with macroeconomics - mix its data into predictors.

 
Evgeniy Chernish #:
All questions to his majesty the forex market and the efficient market hypothesis.

The title is then misleading.

 
Aleksey Nikolayev #:

Thank you, good interesting article.

Imho, you can already try to use fundamental data on daytrips. This is not in the sense of criticising the article, but as a way of thinking. I wonder how macroeconomic data can be adequately "mixed" with price data. The problem is their rare change, for example. Probably, macroeconomics can also be used somehow in price preprocessing - transition from nominal to real exchange rates, for example.

Thanks Alexey ! Frankly speaking I have never been interested in fundamentals and not because it can not give additional information, but simply because it is impossible to cover the vastness. That's why I don't even look in this direction yet.
 
Stanislav Korotky #:

The title is then misleading.

Why ? It uses a classification predictive model that makes predictions. It counts correctly what's put into the model. What's wrong then? That the model can't beat a naive prediction ? I didn't promise that.)
 
Evgeniy Chernish #:
Why ? It uses a classification predictive model that makes predictions. It counts correctly what's put into the model. What's wrong, then? That the model can't beat a naive prediction ? I didn't promise that )

"The impossibility of predicting exchange rates using classical methods..."

 
Stanislav Korotky #:

"The impossibility of forecasting exchange rates using classical methods..."

It didn't even occur to me that it was impossible. I just made a prediction and checked with the python library to check for errors. Maybe someone will add some filters, their own features, maybe someone else will do better. And you immediately impossibility.