Liquidity Risk and Investors’ Mood: Linking the Financial Market Liquidity to Sentiment Analysis through Twitter in the S&P500 Index

Round 1

Reviewer 1 Report

The paper is currently not well focused. Most of the explanatory material on social media in the introduction is unnecessary. The relevant focus of the study is not introduced until the paragraph on lines 43-49. Concerning the background on social media, research papers that link market behavior to twitter seem to be relevant here but are not well summarized. Some suggestions are:

Blankespoor, E., Miller, G.S. and White, H.D., 2013. The role of dissemination in market liquidity: Evidence from firms' use of Twitter™. The Accounting Review, 89(1), pp.79-112.

 

Guindy, M. A., 2017. Is corporate tweeting informative or is it just hype? Evidence from the SEC social media regulation. Social Science Research Network. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2824668.

 

Liew, J., Wang, G., 2016. Twitter sentiment and IPO performance: a cross-sectional examination. Journal of Portfolio Management (Summer), 129–135.

 

Renault, T., 2017. Market manipulation and suspicious stock recommendations on social media. Unpublished working paper. IÉSEG School of Management and Pathéone-Sorbonne University, Paris, France. http://www.thomas-renault.com/wp/market-manipulation-suspicious.pdf .

 

Rakowski, David A. and Shirley, Sara and Stark, Jeffrey, Twitter Activity, Investor Attention, and the Diffusion of Information (June 6, 2018). Available at SSRN: https://ssrn.com/abstract=3010915 or http://dx.doi.org/10.2139/ssrn.3010915

Agrawal, S., Azar, P.D., Lo, A.W. and Singh, T., 2018. Momentum, Mean-Reversion, and Social Media: Evidence from StockTwits and Twitter. The Journal of Portfolio Management, 44(7), pp.85-95.

Following the paragraph in lines 43-49, most of the following background information on liquidity is irrelevant for this study. The definition of the bid-ask spread in lines 158-165 is the next paragraph with relevance. This should follow with literature citations that examine the bid-ask spread as a measure of liquidity, but none are given here. Some suggestions here are (some of these are referenced later in the paper, but would fit well here also):

Amihud, Y. and Mendelson, H., 1986. Asset pricing and the bid-ask spread. Journal of financial Economics, 17(2), pp.223-249.

Glosten, L.R. and Harris, L.E., 1988. Estimating the components of the bid/ask spread. Journal of financial Economics, 21(1), pp.123-142.

Copeland, T.E. and Galai, D., 1983. Information effects on the bid‐ask spread. the Journal of Finance, 38(5), pp.1457-1469.

Roll, R., 1984. A simple implicit measure of the effective bid‐ask spread in an efficient market. The Journal of finance, 39(4), pp.1127-1139.

The empirical model needs to be given in the methodology section. Lines 202-203 only state that a linear regression model was used. This model needs to be clearly specified and justified, along with concerns about control variables, estimation procedure (OLS?), and adjustments for time-series concerns such as autocorrelation or error terms, stationarity, etc.  Were contemporaneous or lagged terms used, and why? The description of the sentiments measure (lines 209-2015) belongs in the methods section and not the findings. A detailed description of the machine learning process should also be given there. It is unclear what significant difference is being described in lines 205-206. What variables are being compared? Where is the test for statistical significance? The tables need more descriptive variable names. There is no reason to use unreadable abbreviations. None of the results appear significant in any meaningful way.

Author Response

The authors would like to thank the reviewers for their comments, which significantly helped improving the manuscript. We believe we have addressed the issues raised by the reviewer.  For each comment, we have first highlighted the issue, then we provide a detailed point-by-point answer, and finally we describe how the manuscript was adjusted. We provide our response to both reviewers in the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper is devoted to the examination of the relationship between information provided in tweets and liquidity. The aim of the paper is clearly stated and the contribution of such examination is obvious. However, there are some major issues and minor remarks stated below which should be considered.  

Abstract seems to be too general - there are five sentences about the overall issues and just two showing what is the core of the study. This should be more balanced.

The paper would benefit if the source of the definition of liquidity (verse 46) was provided. The reasons for reducing financial market liquidity are not exhaustive - the increase in volatility also has an impact on liquidity. It would be better to focus on the informational aspects as these are examined in the paper.

In verses 107-109 something is wrong with this sentence. Do you mean that a market microstructure is related to financial analysis? In what aspects? ("in the market By means"??). Perhaps it is too much to call Random Walk a theory (it is rather hypothesis). Even if, a quotation is missing. In RW innovations are iid, not prices (see e.g. formula 1.14 in https://orfe.princeton.edu/~jqfan/fan/FinEcon/chap1.pdf). 

Verses 127-128 include a repetition of v. 69-71.

It is better to introduce figures in the right order (first Figure 2 and then Figure 3). Figure 2 would be more readable if there were any dates on x axis. What has caused 600 tweets within one day? Perhaps instead of using frequency call it number of tweets.

How is GMS calculated? Is it a simple difference between ask and bid prices or something else? What are the theoretical foundations for using plain range of prices as a spread measure? There are such proposals in the literature (e.g. Bedowska-Sojka and Echaust 2019), but usually the relative range is used. I strongly recommend to check the proxies for spread measures presented there as well as in Fong et al. (2017) or Ahn et al. (2018). There is no information about what is Ask(t) and Bid(t) - is this the last ask and bid price within day t or something else ("the ask (high) quote and the bid (low) quote at a given time" - so is ask price the high price?). This is not clear.

Also, the machine learning strategy is only poorly discussed. First, there is an asymmetry in positive and negative sentiment (and surprisingly there is more positive sentiment than negative). This is not in line with well-known leverage effect. Second, the most positive sentiment that got the highest point is awkward - what does it have in common with S&P? Perhaps some conditions and filters should be used on the tweets in order to get these that might influence traders' decisions. There is information in the paper that "the most negative sentiment was valued at -3.9", while in Table 1 minimum for sentiment is -6.950 (and maximum 101.6). How does it correspond to Figure 1? More information is needed on what is exactly done with tweets, what is their distribution within the time and how they are transformed into sentiment variable.

Third, there is no specification of the regression in the paper. The number of digits in the tables is changing from one line to another. It's not clear which parameters are significant and which are not. By the way, the regression cannot be significant (v. 249-250). How do you interpret the value of R square here? It is really very low. What is this p-value for? Have you examined if spread series and sentiments are stationary? Perhaps you should consider the differences in spread series. So far, too many issues are not clearly explained in the paper, although they should be.  

 

 

References:

Ahn, H.-J.; Cai, J.; Yang, C.-W. Which Liquidity Proxy Measures Liquidity Best in Emerging Markets? Economies 2018, 6, 67.

Będowska-Sójka, B.; Echaust, K. Commonality in Liquidity Indices: The Emerging European Stock Markets. Systems 2019, 7, 24.                           

Fong, Kingsley Y L ; W Holden, Craig;  Trzcinka, Charles A. What Are the Best Liquidity Proxies for Global Research?, Review of Finance, Volume 21, Issue 4, July 2017, Pages 1355–1401, https://doi.org/10.1093/rof/rfx003

 

 

Author Response

The authors would like to thank the reviewers for their comments, which significantly helped improving the manuscript. We believe we have addressed the issues raised by the reviewer.  For each comment, we have first highlighted the issue, then we provide a detailed point-by-point answer, and finally we describe how the manuscript was adjusted. We provide our response to both reviewers in the attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The new version is much improved.  I have no further major concerns.

Author Response

Thank you very much for your consideration.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper is singificantly improved. Few issues should be considered, however. 

Authors investigate how "Twitter microblogging social network influences transaction cost, liquidity immediacy and volatility" (v.43). It seems to me that volatility is in fact not considered in the paper. In the model there are liquidity and sentiments variables only. Which of the measures used is a proxy for volatility? And in Conclusions there is a "level of volatility". Also the relation between spreads and transaction costs is not clearly stated. 

I do not agree with the final conclusion that for models with 2-day moving average "the results are significantly improved" (perhaps have significantly improved?). In three remaing models nothing has changed, so is it significant improvement?  

Table 3 and 4 - we need estimates and either standard errors, or t-statistics, or p-values. Not all of them at the same time. What I miss are any tests for residuals. 

Few typing errors e.g. "information assymetry" (v.131), R-Squared (v. 278) - rather R-squares?

I do not understand this sentence: "This, however, does not mean that another type of relation could be considered. " (v.282) 

Author Response

The authors would like to thank the reviewer for his/her comments, which significantly helped improving the manuscript. We think that the issues raised by the reviewer have been addressed in the new version of the manuscript.
For each comment, we have first highlighted the issue, then we provide a detailed point-by-point answer:

 

Question 1: Authors investigate how "Twitter microblogging social network influences transaction cost, liquidity immediacy and volatility" (v.43). It seems to me that volatility is in fact not considered in the paper. In the model there are liquidity and sentiments variables only. Which of the measures used is a proxy for volatility? And in Conclusions there is a "level of volatility". Also the relation between spreads and transaction costs is not clearly stated.

Answer: We have removed all references to volatility or transaction costs. Certainly we are addressing the relationship between social network Twits and liquidity.

 

Question 2: I do not agree with the final conclusion that for models with 2-day moving average "the results are significantly improved" (perhaps have significantly improved?). In three remaining models nothing has changed, so is it significant improvement?  

Answer: We agree with the reviewer. We have modified the original adjective (significantly). The new version states that "the results are slightly improved".

 

Question 3: Table 3 and 4 - we need estimates and either standard errors, or t-statistics, or p-values. Not all of them at the same time. What I miss are any tests for residuals. 

Answer: We have removed the columns with standard errors and t-statistics. Tables 3 and 4 include just coefficients' estimate and p-values. We also provide some statistics regarding residuals in both tables. We perform several test to analyze the residuals of regression in Tables 3 and 4:

Regarding Table 3: "In addition, we have checked the residuals of all regression models. The Kolmogorov-Smirnov test shows that residuals are not normally distributed except for the ES liquidity measure (p-value = 0.421).
Heteroscedasticity was also checked. In this case, the Breusch-Pagan test showed that residuals were homoscedastic for all regressions."

Regarding Table 4: "Residuals of all regression models have been also checked. The Kolmogorov-Smirnov and Breusch-Pagan tests show that residuals are normally distributed and homoscedastic in all cases."

 

Question 4: Few typing errors e.g. "information assymetry" (v.131), R-Squared (v. 278) - rather R-squares?

Answer: Both typos have been corrected.

 

Question 5: I do not understand this sentence: "This, however, does not mean that another type of relation could be considered. " (v.282) 

Answer: We have removed that sentence.

Author Response File: Author Response.pdf