MSGformer: A Hybrid Multi-Scale Graph–Transformer Architecture for Unified Short- and Long-Term Financial Time Series Forecasting

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. The bibliography needs to be improved, because it is not made in accordance with the journal requirements

2. The citations included in the article are slightly outdated, there are few references to the latest literature

3. There is no justification for using the presented method globally - this was a study conducted on indices and not on individual shares - will the study results be repeated if we take into account individual securities (did the author verify this)

4. There is no justification for using the presented method on other markets - will the specificity of the capital market not cause this method to work very well on one market and not at all on the other (did the author verify this)

5. There is no justification for using the presented method in longer time intervals - will this method be applied on daily or monthly data? will give the same results (did the author check that)

Author Response

The modified attachment has been submitted

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

I have readed your work where you combine the modelling of short and long term influences to apply to financial time series.

When padding the time series, you complete them adding ceros. Could it be possible to use classical imputation techniques?

In formula 12, the denominator in obtaining R2, seems to include a non-needed i subscript on the mean.

In table 2, and in the following figures, several goodness of fit measures are provided; what is the forecasting horizon in each model? Have you tried to decompose the MSE in its mean, variance and covariance components?

Also, for some periods, it seems that there is a systematic error, usually when there is a downward trend. What is the cause? These could have been assessed with non-symetric forecasting measures.

Have you tried to forecast individual stock's prices? In the manuscript, there were index time series.

In the references at the end, it would have been desirable to include all authors instead of using 'et al.', and the DOI's to be able to find them directly.

Author Response

The modified attachment has been submitted

Author Response File: Author Response.pdf