Analyzing Crude Oil Prices under the Impact of COVID-19 by Using LSTARGARCHLSTM

Round 1

Reviewer 1 Report

1. Citation styles, typographic errors, spelling, capitalizing, etc. must be checked again and corrected.
2. I don’t think “and/or” needs to be used. Usually “or” requires just one of statements to be correct, so it included “and” case. If Authors want to stress some kind of peculiarities, then maybe different sentence construction would be better than typing “and/or” which especially repeated in subsequent lines seems a bit strange.
3. Introduction should be much shorter and more concise. Clearly it must be stated without going into details: what, how and why was done, and why is it novel, important and different from previous researches. Of course, some necessary details can be presented in this Section, but currently the Introduction is too vague and long, and it is hard to quickly grasp the main ideas from it.
4. Most of detailed references and reviews should be moved to Literature review section. This Sections is too short now, and taking under consideration my Remark 3, it would be then longer. Except that, the peculiar numerous “strange” names of the models is overwhelming and decreases readability of the paper, at the same time not introducing any new learning for the reader. I suggest to change some names if it is not against the common rules in the field. Secondly, instead of typing a lot of papers and names of models, I suggest to write more what these researches have proposed, explored. Stating that, for instance, “employed neural networks-GARCH” does not mean much, as it would mean to describe in one or two sentence (not-technically) what are these kind of models. Just typing a lot of specialistic terminology does not make a good Literature review, nor is it beneficial for anyone. Besides, abbreviations should be used when necessary, i.e., later can be used to express the same thing with less words (letters). If the term will be not used later, then I suggest to heavily reduce this abundance of several capitalized names. I also suggest to rename, just for the purpose of this paper, the main model, and to reformulate sentences in a way that this long capitalized symbol will not be repeated so often.
5. Line 165: “STAR-STGARCH method that permit non-linearity in both conditional variance and mean” Can you write in the paper what you mean by non0linearity in conditional variance and mean?
6. All symbols used in Equations must be defined and described. For example, what is \gamma in Eq. (1)? Generally, from the current presentation of the models and equations I cannot learn anything new; unfortunately. The presentation of Methodology must be improved. Maybe some graph-sketch should be beneficial.
7. In Data section, the description of time-series must be present, source of data, etc.
8. It would be beneficial to shortly add information what Tsay test and Hsieh test check, as well as, what Shanon entropy and Lyapunov exponent measure.
9. Line 241: What do you mean by “efficient model”?
10. Please make full name for JB test, then abbreviate it. Similarly, BDS test.
11. What is KSS test? Do you mean KPSS? Kwiatkowski–Phillips–Schmidt–Shin?
12. The software that you use must be cited, otherwise there is not possible to replicate your computations, especially, that many test require some initial parameters which different softwares differently initialize.
13. You must decide if “,” or “.” is the decimal separator.
14. Line 271: What does it mean that “system must be withing ~66 days”? Where did you define the concept of a system being in x days?
15. Tables as well as typing is not nice. It must be cleared and typed neatly.
16. Generally, I don’t see great novelty in your research. Of course, it is a novel one. But: You slightly modify the very well known methods, which were explored in a lot of fields. In such a case the paper at least should be neatly written and provide some additional usefulness from “didactic” point of view and be readable for general audience. The current form is untidy written and it is impossible to seriously learn new concepts from your paper.
17. Line 332: What is “epoch”?
18. You should decide and present MSE, or RMSE. Not mix it. Comment that reader can convert are a bit trivial for scientific paper in impacted journal. Moreover, it just makes more mess and I don’t see a purpose to once use this, once that measure in the same paper.
19. Table 6: What is training rho? (Again, undefined symbols…)
20. What is in-sample and out-of-sample periods?
21. Line 364: ???
22. Wilcoxon and DM test are well known enough that you do not have to multiply the number of formulas… If so, why not write details for other less popular tests?
23. Line 378: DM test does not apply to “models” but to “forecasts” obtained from the models. This is a subtle but important difference.
24. The Authors didn’t convince me that the paper is really novel, or has some extraordinary and valuable didactic presentation. In the current form the papers looks as more a research report. Except all my remarks above, in order to make the paper publishable in the impacted journal, I would suggest to improve the discussion of the results: What are the consequences of the obtained results.
25. Secondly, I suggest to repeat the computations with different time-series of oil price, for instance, WTI, Brent and Dubai; in order to check if the results are robust to different measures of oil price.
26. Well, actually, form the current form of the paper I even don’t know which oil price time-series was taken to perform this research in the current form…

Author Response

Firstly, we thank the detailed review and the comments provided.

Our paper was corrected and revised.  Citation styles, typographic errors, spelling, capitalizing, etc.  have been checked again and corrected.

Section 2  named  as the volatilities of Oil Price was added and the volatility of oil price was explained. The results were detaily discussed and explained. Introduction section, and section 3 and 4  were revised   and emphasized the novelty and contributions. Methods and models were explained. The results were detaily discussed. 

In order to demostrate the forecasting effiency of  proposed method we have included results on two other data sets, namely Brent and  Dubai  crude oil  price data sets.

Corrections on the article in accordance with the referee requests are as follows:

 

Review  1

1. Citation styles, typographic errors, spelling, capitalizing, etc. must be checked again and corrected.

Citation styles, typographic errors, spelling, capitalizing, etc.  have been checked again and corrected.

2. I don’t think “and/or” needs to be used. Usually “or” requires just one of statements to be correct, so it included “and” case. If Authors want to stress some kind of peculiarities, then maybe different sentence construction would be better than typing “and/or” which especially repeated in subsequent lines seems a bit strange.

We have replaced ‘and/or’ with ‘or’ in once sentence since it carries same meaning as the reviewer suggested.

3. Introduction should be much shorter and more concise. Clearly it must be stated without going into details: what, how and why was done, and why is it novel, important and different from previous researches. Of course, some necessary details can be presented in this Section, but currently the Introduction is too vague and long, and it is hard to quickly grasp the main ideas from it.

 

We have shortened introduction and emphasized the novelty and contributions. WE have also moved some of the references in this section to the section on related works.

4. Most of detailed references and reviews should be moved to Literature review section. This Sections is too short now, and taking under consideration my Remark 3, it would be then longer. Except that, the peculiar numerous “strange” names of the models is overwhelming and decreases readability of the paper, at the same time not introducing any new learning for the reader. I suggest to change some names if it is not against the common rules in the field. Secondly, instead of typing a lot of papers and names of models, I suggest to write more what these researches have proposed, explored. Stating that, for instance, “employed neural networks-GARCH” does not mean much, as it would mean to describe in one or two sentence (not-technically) what are these kind of models. Just typing a lot of specialistic terminology does not make a good Literature review, nor is it beneficial for anyone. Besides, abbreviations should be used when necessary, i.e., later can be used to express the same thing with less words (letters). If the term will be not used later, then I suggest to heavily reduce this abundance of several capitalized names. I also suggest to rename, just for the purpose of this paper, the main model, and to reformulate sentences in a way that this long capitalized symbol will not be repeated so often.

We have moved the detailed references and reviews to the Literature review section. İnstad of referring to previous works by using abreviations we have elaborated on the significance of each of these works for the field. We have also eliminated most of the abbreviations and especially those that are not used by the current work.

5. All symbols used in Equations must be defined and described. For example, what is \gamma in Eq. (1)? Generally, from the current presentation of the models and equations I cannot learn anything new; unfortunately. The presentation of Methodology must be improved. Maybe some graph-sketch should be beneficial.

\gamma has been defined as the transition parameter that governs the rate of transition between the two regimes. The presentation of Eq. 1-4 has been changed to improve their readability.

6. In Data section, the description of time-series must be present, source of data, etc.

We had used  the WTI oil price data set in our original paper.  In order to demostrate the forecasting effiency of  proposed method we have included results on two other data sets, namely Brent and  Dubai  crude oil  price data sets.

7. “It would be beneficial to shortly add information what Tsay test and Hsieh test check, as well as, what Shanon entropy and Lyapunov exponent measure.”

We have added information on these test under the results section.

 

8. Line 241: What do you mean by “efficient model”?

We have used the term “efficient model “ to imply that the model has  a forecasting error less than the errors of the compaired models.

9. Please make full name for JB test, then abbreviate it. Similarly, BDS test.

BDS is Brock- Dechert-Scheinkman test and  JB is the Jarque-Bera test statistic for residual normality. The open forms of abbreviation of these terms  have been added to the paper

10. What is KSS test? Do you mean KPSS? Kwiatkowski–Phillips–Schmidt–Shin?

KSS is Kapetanios, Shin, and Snell unit root test. The open forms of abbreviation of KSS  has been added to the paper

11. The software that you use must be cited, otherwise there is not possible to replicate your computations, especially, that many test require some initial parameters which different softwares differently initialize.

R software DChaos package for Largest Lyapunov Exponent (LLE), and  R Software Entropy package for calculating Shannon Entropy (SE),  Chao-Shen Entropy Assessing code are employed.

LSTM has been implemented on the Keras framework.

12. “You must decide if “,” or “.” is the decimal separator.”

It has been corrected.

 

13. Tables as well as typing is not nice. It must be cleared and typed neatly.

All the tables in the paper have been redesigned.

14. Generally, I don’t see great novelty in your research. Of course, it is a novel one. But: You slightly modify the very well known methods, which were explored in a lot of fields. In such a case the paper at least should be neatly written and provide some additional usefulness from “didactic” point of view and be readable for general audience. The current form is untidy written and it is impossible to seriously learn new concepts from your paper.

It was written and provided  some additional usefulness from “didactic” point of view and be readable for general audience

 

15. Line 332: What is “epoch”?

One pass over the entire set of samples in training in a iterative training systems. In other words each iteractions is an epoch

16. You should decide and present MSE, or RMSE. Not mix it. Comment that reader can convert are a bit trivial for scientific paper in impacted journal. Moreover, it just makes more mess and I don’t see a purpose to once use this, once that measure in the same paper.

In the current form of the paper MSE is no longer used.

17. Table 6: What is training rho? (Again, undefined symbols…)

rho is the spearman correlation coefficient.

18. What is in-sample and out-of-sample periods?

In-sample forecast is the process of formally evaluating the predictive capabilities of the models developed using observed data to see how effective the algorithms are in reproducing data. It is kind of similar to a training set in a machine learning algorithm and the out-of-sample is similar to the test set.

19. Line 364: ???

It was corrected

 

20. Wilcoxon and DM test are well known enough that you do not have to multiply the number of formulas… If so, why not write details for other less popular tests?

The formulas for these two tests have been removed from the text.

21. Line 378: DM test does not apply to “models” but to “forecasts” obtained from the models. This is a subtle but important difference.

It was applied to forecasts but not models. Table was corrected  

22. The Authors didn’t convince me that the paper is really novel, or has some extraordinary and valuable didactic presentation. In the current form the papers looks as more a research report. Except all my remarks above, in order to make the paper publishable in the impacted journal, I would suggest to improve the discussion of the results: What are the consequences of the obtained results.

 

Introduction section was revised   and emphasized the novelty and contributions. And the results were  expanded and discussed

23. Secondly, I suggest to repeat the computations with different time-series of oil price, for instance, WTI, Brent and Dubai; in order to check if the results are robust to different measures of oil price.

In addition to the results on  the WTI oil price data set that we reported  in our original submision,  we have included new results on the brent and dubai  oil price  data sets in the current submision

24. Well, actually, form the current form of the paper I even don’t know which oil price time-series was taken to perform this research in the current form

We had used  the WTI oil price data set in our original paper.

Author Response File: Author Response.docx

Reviewer 2 Report

1. Results: Recommend to be Rejected

This paper proposes a hybrid model (namely LSTARGARCHLSTM) for analyzing oil price volatility from 29th May 2006 to 31th March 2020, due to many economic and non-economic factors such as Covid-19. The GARCH, LSTARGARCH methods and their extensions GARCHLSTM and LSTARGARCHLSTM based on deep learning were comparatively evaluated. Among these models, the LSTARGARCHLSTM model exhibited significantly better prediction quality in terms of out of sample accuracy than the others in the face of chaos, outliers and nonlinear behavior.

It is with very minor merits for Energies. The only decision is rejection.

Firstly, the abstract is disorganized, i.e., poor writing. It should be refined to precisely illustrate what authors have done in this paper within 200 words.

Secondly, for Sections 1 and 2, authors should provide the comments of the cited papers after introducing each relevant work. What readers require is, by convinced literature review, to understand the clear thinking/consideration why the proposed approach can reach more convinced results. This is the very contribution from authors. In addition, authors also should provide more sufficient critical literature review to indicate the drawbacks of existed approaches, then, well define the main stream of research direction, how did those previous studies perform? Employ which methodologies? Which problem still requires to be solved? Why is the proposed approach suitable to be used to solve the critical problem? We need more convinced literature reviews to indicate clearly the state-of-the-art development.

By the way, some cited papers are not listed on the reference list, such as Barone-AdesiBourgoin and Giannopoulos (1998), Adrangi and Chatrath(2001), Lahmiri (2017), Komijani et.al (2014), and He (2011), please check these papers carefully.

For Sections 3 and 4, they are suggested to be re-written and re-organized. In addition, authors should introduce their proposed research model/framework more effective, i.e., some essential brief explanation vis-à-vis the text with a total research flowchart or framework diagram (Figures 1 and 2) for each proposed algorithm to indicate how these employed models are working to receive the experimental results. It is difficult to understand how the proposed approaches are working.

For Section 5, authors should also conduct some statistical test to ensure the superiority of the proposed approach, i.e., how could authors ensure that their results are superior to others only based on Tables 7 and 8? In addition, authors also have to provide some insight discussion of the results with Tables 7 and 8.

Author Response

Firstly, we thank to  the detailed review and the comments provided.

Our paper was corrected and revised.  Citation styles, typographic errors, spelling, capitalizing, etc.  have been checked again and corrected.

Section 2  named  as the volatilities of Oil Price was added and the volatility of oil price was explained. The results were detaily discussed and explained. Introduction section, and section 3 and 4  were revised   and emphasized the novelty and contributions. Methods and models were explained. The results were detaily discussed. 

In order to demostrate the forecasting effiency of  proposed method we have included results on two other data sets, namely Brent and  Dubai  crude oil  price data sets.

Corrections on the article in accordance with the referee requests are as follows:

 

Review  2

1. Firstly, the abstract is disorganized, i.e., poor writing. It should be refined to precisely illustrate what authors have done in this paper within 200 words.

We have  refined and rewritten the abstract by emphasizing the main points made in the paper.

2. Secondly, for Sections 1 and 2, authors should provide the comments of the cited papers after introducing each relevant work. What readers require is, by convinced literature review, to understand the clear thinking/consideration why the proposed approach can reach more convinced results. This is the very contribution from authors. In addition, authors also should provide more sufficient critical literature review to indicate the drawbacks of existed approaches, then, well define the main stream of research direction, how did those previous studies perform? Employ which methodologies? Which problem still requires to be solved? Why is the proposed approach suitable to be used to solve the critical problem? We need more convinced literature reviews to indicate clearly the state-of-the-art development.

We have shortened introduction and emphasized the novelty and contributions. We have moved the detailed references and reviews to the Literature review section. İnstad of referring to previous works by using abreviations we have elaborated on the significance of each of these works for the field. We have also eliminated most of the abbreviations and especially those that are not used by the current work.

 In the literature , Garch model has been trained with lstm or other methods to achieve forecasting performance. However, Garch model may not achieve high forecasting performance under the presence of nonlinear or chaotic movement since it considers the data period as a whole. Since the data period is considered as a whole, the structure in different regimes is not analyzed.

On the other hand, the regime switching models partition the data period into two regimes. Conducting analyses for two almost stable regimes rather than the whole period gets rid of the forecasting problems arising from chaotic structure. When LSTAR Garch LSTM is used for modelling, , the determination of the arch and garch coefficients  for each regime facilitates the positive relation of the volatility to its past to be uncovered. Not only policy suggestions can be made accordingly, but also accurate predictions of future oil prices can be made on the basis of arch and garch coefficients discovered within each regime.

 

 

3. By the way, some cited papers are not listed on the reference list, such as Barone-AdesiBourgoin and Giannopoulos (1998), Adrangi and Chatrath(2001), Lahmiri (2017), Komijani al(2014), and He (2011), please check these papers carefully.

The references indicated by  the reviwer have been added to the list of references.

4. For Sections 3 and 4, they are suggested to be re-written and re-organized. In addition, authors should introduce their proposed research model/framework more effective, i.e., some essential brief explanation vis-à-vis the text with a total research flowchart or framework diagram (Figures 1 and 2) for each proposed algorithm to indicate how these employed models are working to receive the experimental results. It is difficult to understand how the proposed approaches are working.

Introduction section, and section 3 and 4  were revised   and emphasized the novelty and contributions. Methods and models were explained. The results were detaily discussed. 

5. For Section 5, authors should also conduct some statistical test to ensure the superiority of the proposed approach, i.e., how could authors ensure that their results are superior to others only based on Tables 7 and 8? In addition, authors also have to provide some insight discussion of the results with Tables 7 and 8.

In Table 7 and 8, it was added some statistical tests and the results were discussed.

 

 

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

1) The abbreviation LSTARGARCH should be expanded in the Abstract.
2) The frequency of data should be stated in the Abstract.
3) RMSE and MAE should be expanded in the Abstract.
4) The advantage of LSTARGARCH should be shortly given in the Abstract
5) Negative price? It was a futures prices, not spot prices. Please, clarify this.
6) I would really like to see longer Literature review, and shorter Introduction. And Introduction to be less general about the fact that oil prices are volatile, especially recently, because this can be written as for Introduction in namely a few sentences. More is appropriate for Literature review. On the other hand, things like "two regimes" should be precised. For example, "two regimes (rising prices and falling prices; recession, expansion; or whatever Authors mean and study herein). More clear explanation why the LSTAR model is used: advantages, disdavantages. "we employ a hybrid modeling": What do you merge in the model? Why? Of course technical terms are necessary but in short, like "Shanon Entropy and Lyapunov exponent tests", but add more general and convincing explanation like: "they are able to capture the problem of...", "they are usueful in modelling". And again not to define by definition like they "are able to capture chaotic behaviour", but describe as for non-specialist what it is in particular, for practitioners, and so on. And do it in a short, convincing, witty way. Not as a multiplication of technical terms. These technicalities should be kept for other parts of a scientific article.
7) Sec. 2 and 3 can be joined into just one Literatrue review
8) What is s_t in Eq. (1)? What role does it play? How \gamma "governs the rate of transition". What range can this variable take? What economically means small value and large? The Methodology section is still not elegant enough for me. This must be a balance between details (not too much), but everything which is too complicated to explore in formulas for article, should have some comment good enough to give the reader at least some imagination what is going on.
9) Despite stating it in Response to reviewers, there is still not nice explanation of abbreviations. They should be explained in this place, in which the first occour. They are explained, but in quite random places in the text. What I mean: the paper still is not "easy to read"
10) Of course, I know what is in-sample and out-of-sample! I expect that the exact periods will be stated in the text.
11) Le is one used with lower subscript another time without. Again, once full names is used another time abbreviation, another time both. I dont see any sens in this. Again, the paper is not written in a neat way for me. It is as a result hard to read, and many things remain hard to graps.
12) Table 1: What are lbop, ldop, etc.? Volatilities. Ok, but how computed, where is the exact formula?
13) Please, as it is really hard to go on over Methods of this paper. Write the paper in rigourous way. Something like this. Let y_t be the volatility of oil price, i.e., y_t = ... . Volatility computed in such a way will be denoted by:... And so on.
14) Descriptions in Table 2 are duplicated.

And this applies to all my abve questions. I dont expect answers to them in Response to reviewers. The answers to this questions must be inserted into the text of the article, so a reader will not have to ask them when reading the paper! I appreciate you corrected the paper comparing to previous version, but I am not feeling ok to recomment its publishing in impacted factor journal in the current form. I am also not able to go over line by line over this paper and its next versions, because the work required to make it acceptable would make me in my opinion a coauthor of this paper. Please, spend enough time on this paper and seriously order this paper. It requires quite much work to make it publishable. You have ok results (as I guess now), but the presentation is still unfortuantely poor.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

1. Results: Recommend to be Major revisions

This paper has been revised and can be considered for publication in Energies. However, before acceptance, some critical issue should be improved. Firstly, there are two Section 2, please revise it carefully. In Section 5, the accuracy comparison of WS and DM, authors only roughly cite Francis and Roberto (1995) but without adding it into the reference list, please check it carefully. In addition, the revision authors have done is insufficient, please provide the total test details to demonstrate their professionals in this field. Authors can refer the following references for their writing improvements.

Zhang, Z.-C., Hong, W.-C., Li, J. Electric load forecasting by hybrid self-recurrent support vector regression model with variational mode decomposition and improved cuckoo search algorithm. IEEE Access 2020, 8, 14642-14658.

Pant, T., Han, C., Wang, H. Examination of errors of table integration in flamelet/progress variable modeling of a turbulent non-premixed jet flame. Applied Mathematical Modelling, 2019, 72, 369-384.

Kundra, H., Sadawarti, H. Hybrid algorithm of cuckoo search and particle swarm optimization for natural terrain feature extraction. Research Journal of Information Technology 2015, 7, 58-69.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 3

Reviewer 1 Report

1) I am still not satisfied how the long and eyestriking abbreviation "LSTARGARCHLSTM" is multipled throughout the text. Phrases like "In the proposed LSTARGARCHLSTM model" can be really reduced to "In the proposed model". Then it will be much more readable.

2) Methodology section is still too hard for non-specialist to easily get what is really done there. In other words, it is hard for the reader to understand what is really done in the paper.

3) What is the difference between lo_p variables and lbop_t, lwop_t, ldop_t? From the defining formulas they seem to be the same. If so why introduce additionally notation lo_p?

4) There are many footnotes, which maybe better to move directly into the text.

5) The whole paper still seems a bit chaotic and exposition is not lead in a neat, easy to grasp way, and so on. I suggest making notations, indices, and so on more neat. Improve tables, try to put p-values or so in the same line - I mean improve graphical and typing presentation of the text. The paper depends heavily on methodology, formulas, tables, and methods which require many parameters. If so, all of these should be clearly and nicely present.

In the current form the paper still does not look elegant enough for me to be published in impacted journal. I see improvement comparing to previous versions, but the presentation in my opinion must still be improved.

Author Response

Any revisions were clearly highlighted. 

 

Firstly, we thank the reviewers for the detailed review and the comments provided.

Corrections on the article in accordance with the referee requests are as follows.

 

1) I am still not satisfied how the long and eyestriking abbreviation "LSTARGARCHLSTM" is multipled throughout the text. Phrases like "In the proposed LSTARGARCHLSTM model" can be really reduced to "In the proposed model". Then it will be much more readable.

Comment about  abbreviation “LSTARGARCHLSTM have been taken and the suggestions corrected .

2) Methodology section is still too hard for non-specialist to easily get what is really done there. In other words, it is hard for the reader to understand what is really done in the paper.

Taking into account the useful comments of the referee the Methodology section was revised and reorganized.

3) What is the difference between lo_p variables and lbop_t, lwop_t, ldop_t? From the they seem to be the same. If so why introduce additionally notation lo_p?

Defining lo_p variables formulas  have been removed.

4) There are many footnotes, which maybe better to move directly into the text.

Some footnotes were moved directly into the text.

5) The whole paper still seems a bit chaotic and exposition is not lead in a neat, easy to grasp way, and so on. I suggest making notations, indices, and so on more neat. Improve tables, try to put p-values or so in the same line - I mean improve graphical and typing presentation of the text. The paper depends heavily on methodology, formulas, tables, and methods which require many parameters. If so, all of these should be clearly and nicely present.

In the current form the paper still does not look elegant enough for me to be published in impacted journal. I see improvement comparing to previous versions, but the presentation in my opinion must still be improved.

p-values and some explanations were added in the tables. Methodology, formulas, tables, and methods were revised and some sections are reorganized.

Best Regards,

Authors

Reviewer 2 Report

Authors have completely addressed all my concerns.

Author Response

Thanks for your report