Analysis and Comparison of Bitcoin and S and P 500 Market Features Using HMMs and HSMMs †
Round 1
Reviewer 1 Report
I enjoyed reading "Analysis and Comparison of Bitcoin and S&P 500 Market Features using HMMs and HSMMs" very much. The research issue at hand is very interesting, focusing on the trending bitcoin prediction in literature. The manuscript is very well-written and requires only a minor polishing prior to publication. My only comment refers to the presentation of the methodology and the results, Given that this is an applied research paper, I would appreciate a mathematical presentation of the methodologies and in this way a direct linkage with the coefficients used in the Tables. In its present form the reader has to be familiar with HMMs and their functional form to understand the meaning fo the various a, v, μ etc. Moreover, prior to using an acronym it would be useful to provide the full naming of each methodology, in the first time it is used (could not find the exact menaing of HMM, HSMM and others).
Author Response
Dear editor and reviewers,
Kindly accept our gratitude for your support and valuable suggestions.
The following points refer to an exhaustive list of modifications made to the paper following reviewers’ suggestions. Since the paper was done in latex, changes in text are marked in red.
(1) One of the reviewers has asked for a better mathematical presentation of the methodologies and a direct linkage to the parameter estimates tables. This is a very valid suggestion, and is done on Pg 3 of the paper (see text marked in red). It must be noted that, while as much mathematical detail as possible is given, a balance had to be found between keeping the paper concise and adding too much of it. While theory relating to their formulation is better clarified, theory relating to their estimation and filtering (via Viterbi algorithm) is referenced as this is considerable. Hopefully, the current level of detail achieves better understanding while retaining the conciseness.
(2) One of the reviewers has suggested that acronyms to be displayed. This is currently available in the abbreviations list on Pg 26, but we admit that it may not be noticeable to the reader initially. However, we abided by the suggested format of the journal. After consultation with the editor, it was suggested to keep things as they are. We have, however, found two missing acronyms and we added them.
We hope that the changes implemented are to the reviewers’ satisfaction.
Sincerely,
David Suda, Luke Spiteri
Reviewer 2 Report
Summary: This paper models time series for BTC/USD426 and S&P using normal-HSMM. This is a nice empirical application. Below there are some minor comments and suggestions which should be addressed:
General comments
The summary must be rewritten. There is only one sentence about the results (this one: “We conclude that our modelling approaches yield9 positive results in both BTC/USD and S&P 500, and both are best modelled via four-state HSMMs.”)
The summary reads like a description of the purpose of the paper. To arouse the interest of the readers, please highlight the most important results a little more.
Please clarify the overlap between this paper and the following reference
Suda, D. and Spiteri, L. Comparing market phase features for cryptocurrency and benchmark stock index453 using HMM and HSMM filtering. To appear in Lecture Notes in Business Information Processes, 2019
Introduction first para: Add references
After reading the introduction, it is not clear to me what the main contribution of the work is to previous literature. Please add a sentence about the contribution of the work (may be at the end of the first section after the literature search). The hidden semi-Markov model (HSMM) allows the underlying stochastic process to be a semi-Markov chain. Is this the first application to BTC/USD and S&P time series?
There is too much material (Graphs, tables). Please condense some of this material. Some figures are difficult to read (Figure 7. Expanding window: 4-state normal-HMM filtering via the Viterbi algorithm on BTC/USD. The colours vary by the mean while the sizes vary by the volatility.)
Limitations of the paper
The time series of the BTC/USD rate is only three years. This is a limitation which should be mentioned
It would be fair to mention the limitations of the HSSM approach (e.g. prior specification of the number of states for the hidden chain)
References: Too many unpublished working papers are cited. Please check
Author Response
Dear editor and reviewers,
Kindly accept our gratitude for your support and valuable suggestions.
The following points refer to an exhaustive list of modifications made to the paper following reviewers’ suggestions. Since the paper was done in latex, changes in text are marked in red.
(1) One of the reviewers has suggested that the summary is rewritten, as there is only one sentence about the results, and the more important results highlighted. This is a very valid point, and indeed the abstract has been rewritten with the aim of highlighting the main results in this paper.
(2) One of the reviewers has suggested that the overlap between paper and original conference paper is highlighted. We have dedicated the first paragraph of the paper to clarify this.
(3) One of the reviewers has suggested that references are added to (what was formerly) the first paragraph, which included some discussion on cryptocurrencies. This is now the second paragraph. A lot of this has been rewritten, and it now focuses more on general discussion on Bitcoin/cryptocurrency price movements and volatility than news events which may have greatly affected it, to avoid quoting news and internet discussion, and focus on scholarly references and official reports. These are included throughout the discussion in the second paragraph.
(4) One of the reviewers has indicated that it is not clear what the main contribution of the work is to previous literature. The last paragraph in Section 1 (Introduction) has been dedicated to elaborate on this more clearly.
(5) One of the reviewers has mentioned that there is too much material (in terms of tables and figures) and it needs to be condensed. This has been quite a difficult task, as a lot of the outputs presented are important. As much as possible, we have condensed certain tables into one rather than have multiple ones. There were cases when this was not possible. We have opted to eliminate what were formerly Figure 2 and Figure 13, where mixture distributions overlapped the histogram, from the text. This is because these figures were not so important within the main applications of the paper.
(6) One of the reviewers has mentioned that some figures are difficult to read, in particular referencing Figure 7 and similar. Indeed, the authors agreed that the captions were not clear enough to explain the figure. Figure 6, 10, 16, 20 now have more detailed captions. Figure 21 and 22 have also been given some extra detail in the caption.
(7) One of the reviewers has suggested that the limitations of the study are mentioned, also suggesting a couple. This is a very valid point, and the last section of the conclusion is now partly dedicated to a number of limitations.
(8) One of the reviewers has suggested that there are unpublished working papers cited. We have identified three that we can omit. These have been removed from the bibliography, as has their reference in the text.
We hope that the changes implemented are to the reviewers’ satisfaction.
Sincerely,
David Suda, Luke Spiteri
Round 2
Reviewer 1 Report
The authors have responded effectively to my concerns from the first round of review, so I am happy to propose this version of the paper for publication.
Reviewer 2 Report
Comments: The revisions are satisfactory.
