Exploring the Factors Influencing the Impact of the COVID-19 Pandemic on Global Shipping: A Case Study of the Baltic Dry Index

Round 1

Reviewer 1 Report

1) At line 11: The BDI is a metric that reflects the worldwide shipping costs and directs related to supply and demand conditions, making it an indicator of economic production." 

"an indicator of economic production." should be "a leading early economic indicator for global economic production"

 

2) Afterwards, "....13 independent variables, including raw materials, energy, stock market indexes, global 14

port calls, and confirmed COVID-19 cases to investigate how to influent the BDI" should be detailed for: raw material prices, energy prices? 

Further, international scrap steel prices, iron ore prices,  global port calls and the Commodity Research Bureau Index are found to significanyly affect BDI. Here above, authors should have mentioned them beforehand.

 

3) Robustness is important. The results should be checked for robustness. 

 

4) Could you revise Figure 1 so that the key dates are visible on it? First official case, first lockdown policy dates, announcement of world health organization decleration etc? 

 

5) BDI is generally considered as a predictor of economic activity, therefore, it is generally assumed as an explanatory variable, a predictive regressor, not the opposite, i.e., generally, BDI is not taken as the dependent variable as it is in this study. Discuss the two appproaches by refering to research that assume BDI as dependent and as dependent. Reformulate the literature section at least a parahraph with this respect. I suggest addition of 

Examination of the predict-ability of BDI and VIX: a threshold approach, M Bildirici, IŞ Onat, Ö Ersin

Forecasting BDI Sea Freight Shipment Cost, VIX Investor Sentiment and MSCI Global Stock Market Indicator Indices: LSTAR-GARCH and LSTAR-APGARCH Models M Bildirici, I Şahin Onat, ÖÖ Ersin Mathematics 11 (5), 1242

 

6) Section 3.1., in variable explanations here, references should be added for sentences such as "...perceived as a leading economic indicator." and " SP500...it reflects a broader range of market changes...in the stock market.", "VIX is...a popular measure of the stock market's expectation of volatility", "The Global Port Calls...composed of 82 international ports worldwide, covering more than 60% of global port trade"

 

7) In table 1, Buker Index should be Bunker Index

 

8) data frequency should also be stated in the data section before table 1. 

 

9) Correlation table takes too much space possibly due to font and spacing. Revise if possible. Also add * ** for showing significance after t tests. Interpret them in terms of significance. 

 

10) Check the paper for Grammar and typos. Example: at line 276, verb, "export", should be "estimated"   

 

11) How robust are the results to heteroskedasticity and ARCH effects? The findings should be checked with diagnostic tests. Not reported in Table 5, except VIF. Is multicollinearity the main concern in an econometric application with daily data? Daily time series and especially BDI and especially variables such as stock market index and VIX are known to be subject to heteroskedastic, leptokurtic distribution in addition to nonlinearity. These should be integrated to the analysis. I suggest extending the empirics by suggesting a two step process, at the first step, take your results from stepwise regression, afterwards, at the second, add arch garch structure and reestimate the model by taking heteroskedasticity into consideration. Then, compare if any deviations occured before and after in a table and in a relevant textual explanation sections. Therefore, Table 5 should be revised with this respect. Because, if the method is proposed to be robust to heteroskedastity by some estimator or robustness method, it is secondary importance and not very realistic after the existent literature on financial markets.  This is the foremost and major concern I direct to this study and it should be done with important attendance. 

 

12) This is time series data and unit root tests should be added as a table and for robustness, 2 or more tests are required. Also, diagnostics table is missing. They should be added to the data section as Table 2 and 1 (in order). I also suggest ARCH-LM tests for variables under investigation.  

13) Any transformation to data such as logarithms and first differencing should be conducted and noted at the data section following critique 12. 

14) Following corrections and necessary varaible revisions due to tests I directed above, revision of empirics. It is suspected that variables are in levels and nonstationary. AIC values are very high. The above-mentioned corrections will also deal the high AIC's.  

15) Conclusion 

16) Future directions

 

Minor issues, typos, minor grammar errors. 

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

I consider that the topic is actual and scientifically interesting. The manuscript is clearly structured and organized, it is easy to follow and the terminology is appropriate to the subject. The content of the paper is succinctly described and contextualized in relation to the presented theoretical background, also tables and figures are used effectively and support the text. To increase the scientific soundness, I present the following comments and recommendations:

1. I would recommend to emphasize more in the abstract the relevance, originality and quality of the research, persuasively suggesting to the potential reader the items of interest that the work proposes.

2. I recommend that you present the research method used much more clear and in detail, especially Section 3.2: Variables selection method—Stepwise Regression, providing the necessary elements for the reproduction of research by any other research group that uses it exactly (the repetitive and reproducible nature of science).

3. I believe that the authors should reflect the extent to which the results answered the questions mentioned in the introductory part: What is the research topic and what is your paper's contribution/ innovation for the research?

4. The concluding elements of the paper are represented by strong statements based on scientific arguments that are presented clearly and concisely. However, I believe that the authors should reflect the extent to which the results answered the questions mentioned in the introductory part: What is your paper's contribution/ innovation for the research?

5. We have found that bibliographic references (in total of 25) are described accurately, sincerely and deontologically by the authors.

 

6. Minor corrections: Renumber the Table 1, subsection 4.1 (line 254); Renumber the Table 2, subsection 4.2.2 (line 307).

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

This study investigates the impact of the pandemic on global shipping 10 by analyzing the Baltic Dry Index (BDI) using stepwise regression to select variables and build models before and after the pandemic. The mathematical model is properly formulated with the numerical experiments as case study.

 

Based on the results, please discuss the theoretical and practical implications for practitioners and policy makers.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Dear Authors, 

Thank you for providing the detailed response file which included a table presentation regarding the comments, how they were met in the revision and the transformed text after correction. 

 

When I evaluate the second round version of the paper, I see that the comments regarding editing has been corrected.

The major problem is for empirics. The comments I made after comment 3 was on empirical section. Heteroskedasticity is important concept in Econometrics. Engle got a nobel prize for its 1982 contribution. This paper is 4 decades ago. You cannot avoid basic OLS assumptions, importance of Gauss Markov theorem, you cannot avoid how heteroskedasticity leads to biased estimates and estimators. 

Another issue with this respect (GM theorem and OLS) is the stationarity tests. Authors also declined conducting them and revising the empirics accordingly. Working with non stationary variables leads to spurious regression results.

I do not understand why the basic steps followed in econometrics for time series data are not followed and the corrections are made. 

As a reviewer, this puts the reviewer into the decision of reject. However, I consider that authors have not gone through the comments carefully with these respects to grasp the seriousness of the issue. I strongly suggest from the authors to please go over the empirical section. 

Not even the basic descriptive statistics table I asked for is not presented. 

In econometric analysis with time series data and in a paper that conducts this method, the econometric steps should be presented as they should be:

Descriptive statistics table, stationarity unit root tests table, then differencing data as much as it is necessary, estimating model, diagnostics tests, if problems exist such as heteroskedasticity, you should deal with it either by adding ARCH GARCH terms or following a second best but less efective method, obtaining robust to heteroskedasticity results. Therefore, shifting empirics to ARCH GARCH could be avoided which is not the best way but with applying robust techniques to heteroskedasticity. These are not done and especially the unit root testing and not doing UR tests is a serious issue. 

 

Further issues: 

a) if outliers and structural changes exist, you should control them by adding dummy variables, which is expected if looked at figures presented for the variables. 

b) Figure is corrected but again it has a typo. 2012 on the right hand side is typo since horizontal axis starts from 2019, follows 2020, then 2012? it should it be 2021? 

Dear Authors, I regret to inform you that empirics with this serious problems should lead to rejection on my side. However, I assume that you might have considered that I suggested revising the paper into a GARCH paper which is not my intention. Your models are parametric and the parameters are subject to serious biased results unless you don't prove that model became consistent in terms of parameters and efficient in terms of OLS by controlling nonstationarity, heteroskedasticity, structural breaks, outliers, not to mention autocorrelation in the model. Otherwise, sorry to inform that my decision will be reject since there are serious reliability concerns for the empirical findings of this paper.

Due to  potential of the paper and due to the careful work that authors had done so far except for the empirics - which is most important part that is not corrected and waiting to be corrected - I change my reject decision to major revision for another round.   

Minor revision for typos.. 

Author Response

  

Author Response File: Author Response.docx

Round 3

Reviewer 1 Report

Dear Authors,

First of all, thank you for taking the comments and integrating them to the article. I am happy to see that you added the requested unit root tests and added a section on stationarity check.

Also, I see that requested heteroskedasticity testing is added to the paper. It is not very understandable to see daily data including BDI data leads to homoskedasticity. However,  the results of the authors for this respect pointed at no heteroskedasticity, authors assumed that residuals of the models are homoskedastic and therefore, no need for taking necessary steps regarding heteroskedasticity. 

I noted that unit root tests led to the conclusion that series are nonstationary for their levels. They become stationary if they are first differenced. This means that regression results would be spurious regression results if the authors do not continue the analysis with first differenced series. After examining this last version of the paper and by comparing it to the previous version of the paper, I see that the necessary steps after unit root testing is not done. Estimation results are the same. AIC results are the same. Interpretations are the same since estimated model is the same model. However, this is a huge mistake to continue the analysis without the first differenced data.

As I noted in the previous round, authors should follow necessary steps in econometrics. This is not followed for the empirics. This is a very serious issue. It is impossible to my end to accept the paper unless estimation resutls are not corrected. And I don't understand why authors did not do it in the last round after noticing that their data is nonstationary. At the following link for Parker's econometrics book chapter 4, please see Table 4-1 for Granger and Newbold's spurious regression result. 

https://www.reed.edu/economics/parker/312/tschapters/S13_Ch_4.pdf

As you see, if series are nonstationary, it is possible to obtain spurious results which is represented by R squares being around 0.27 or so, though the variables have no association in reality as given for first differenced results with R squares being zero in their example. 

In your case, please see eq. (4.6) and (4.7) in the link above. Your regressions are in the form of 4.6. However, you should first difference your data and reestimate your models as in the form of eq. 4.7. After this, your intercept, your betas and therefore findings will be changed. Most importantly, they will be reliable in the context of stationarity. You should correct it.

There are other issues as well that are not addressed. You should control autocorrelation, include necessary lagged dependent variable to the models as independent variables while doing the above-correction. Also, similar to heteroskedasticity, check autocorrelation with Breusch's serial correlation test. 

Afterwards, revise your paper. Your regression results are reported just the same as they were in the last round. This is not acceptable. Then, revise interpretations as well in necessary places including conclusion, model evaluation and abstract. 

One last comment. In the abstract, your findings are not reported in detail and the significance of these findings are not reported in the last sentences where your concluding marks are at. If you do not highlight the importance of these findings in the abstract, the contribution will not be clear to the reader who reads abstracts at first sight. Abstract is important. If it is catchy, the reader will be more interested. This is expected to affect the citation scores of the paper and the journal as well.  I suggest revision of the abstract by highlighting the importance of the paper, its contribution and how it fills the research gap. 

Overall, I regret to inform that my decision is major revision again which is very unexpected for me to see that after the unit root tests, authors continued to report their suspicious results. This suspicion is a very well known issue in econometrics. Authors cannot avoid the influence of nonstationary data on econometric models. I don't understand why they did not do it in the last round. This is the cause of one additional round.  Steps are taken, paper has improved after the last round.  However, still another round is needed due to the problem with empirical section which I tried to warn in the last round in a detailed letter. Now, I am writing another detailed instructive review with the hope that authors will correct the paper. 

Regards. 

Minor. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx