Review Reports - Business Confidence Index (BCI) and Share Return Volatility Nexus: Sectorial Empirical Evidence

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Please see the attachment.

Comments for author File: Comments.pdf

Author Response

reply is in the attached document

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This study investigates the relationship between the Business Confidence Index (BCI) and
the volatility of stock returns in South Africa across the Financial Services, Industrial, and Resources sectors of the Johannesburg Stock Exchange. Results reveal that BCI significantly influences stock return volatility, particularly in upper quantiles, suggesting heightened sensitivity during periods of elevated market activity.

I like concept for the paper. However, on the conceptual level also need to investigate if there is a bidirectional relationship between BCI and Volatility for different sectors. Because although volatility can signal economic uncertainty, influence consumer expectations, investment decisions, and business sentiments; increased confidence may lead to lower long term volatility. this might be different for different countries.

These findings support behavioural finance theories?.... this implication is not as clear and generalization may be removed. What about other indices like Consumer Confidence index?

Policymakers should thus treat BCI as a valuable early-warning tool, using it to guide 235
countercyclical interventions, particularly when confidence shifts abruptly and could 236
amplify market volatility......over what time period and most likely bidirectional impact will exist here as well due to the policy changes.

Therefore, policymakers should pair confidence-build- 263
ing measures with structural reforms and crisis management tools to ensure stability 264
across all market regimes...............too general and not directly supported by the study

These findings support behavioural finance theories and emphasize the need for dif- 19
ferentiated policy strategies to manage market risks in emerging economies........how have you shown that through your research? Behavioral theories mentioned in the introduction and not tied in the conclusion or results.

volatility dynamics? .....what does dynamics mean here?

.....................

Editing Only

Minor: Spacing and equation numbering seems to be an issue through out the paper. Sometimes, I feel the definition of variables is not easy to find in the paper.

Empirical studies have shown that business confidence significantly influences market 42
volatility (An, Wang, Delpachitra, Cottrell and Yu ,2022)., for instance, periods of heightened 43
business confidence are often associated with increased economic activity and market opti- 44
mism, leading to higher investment but also greater volatility due to speculative activi- 45
ties. Conversely,

This (Our) study examines these (?) relationships,...can be more direct

Index (BCI)

Tiwari, Adewuyi, Awodumi, a
nd Roubaud, (2022)

The foregoing equations (5.4) and (5.5) were further adapted in the form of GARCH 169
(1,1), As GARCH(1,1) m

Figure3.2 Quantile regression Plots - Impact of BCI on JSE Industrial Index 250
Source: Author’s compilation using STATA

Impling that

The GARCH(1,1)

volatility dynamics?

Author Response

reply is in the attached document

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Manuscript ID: jrfm-3808092

Title: Business Confidence Index (BCI) and Share Return Volatility nexus: Sectorial Empirical evidence

This manuscript examines the relationship between the Business Confidence Index (BCI) and stock return volatility in South Africa, using both quantile regressions and GARCH(1,1) models on sectoral stock index data. The study’s objective is to determine whether changes in business sentiment have a measurable impact on the volatility of stock returns and whether this impact differs across market conditions and sectors (financial services, industrial, and resources sectors of the JSE). The authors find that higher business confidence significantly influences return volatility, with effects that are more pronounced in the upper quantiles of volatility (i.e. during periods of high market activity). According to the results, an increase in BCI appears to reduce volatility in the financial sector, have a muted or insignificant effect in the industrial sector, and correlate with higher volatility in the resource sector. These sector-specific outcomes are further amplified during major crisis periods such as the 2008 Global Financial Crisis and the COVID-19 shock. Overall, the paper concludes that sentiment effects on volatility are asymmetric and sector-dependent, supporting behavioral finance theories that suggest that investor psychology can drive market fluctuations beyond what fundamentals alone would predict. The authors argue that their findings offer useful insights for policy makers and investors by emphasizing the need for differentiated risk management strategies that account for sentiment-driven volatility in an emerging market context.

While the manuscript presents interesting findings, I have several concerns that should be addressed.

First, the empirical methodology and its exposition need clarification. The paper employs two approaches (quantile regression and GARCH modeling), but the integration of these methods is not entirely clear. The authors should better explain how the quantile regression was implemented and what exactly was being modeled — for instance, whether the regression was applied to realized volatility measures across different quantiles, and how temporal dependence was handled. Currently, the description is sparse, making it difficult for readers to fully understand or replicate the analysis. Similarly, more detail is needed on the GARCH(1,1) setup: it appears BCI is used to examine volatility in each sector’s index, but the manuscript should specify whether BCI enters the variance equation (as an exogenous regressor) or if separate GARCH models were run for different sub-periods. Clear equations or a more rigorous explanation of the modeling steps would greatly enhance the methodological transparency. In addition, the authors should justify their choice of volatility measure and model. For example, if volatility is proxied by the standard deviation of returns (as implied in the tables), what is the sampling frequency for this calculation (monthly, quarterly?), and why was this chosen over alternative volatility measures (e.g., GARCH-estimated variance or realized volatility)? A more explicit methodological section would address these ambiguities.

Second, the paper would benefit from a thorough analysis of robustness and a discussion of generalizability. The results hinge on a single sentiment proxy (BCI) and one country’s data. It is important to demonstrate that the findings are not an artifact of a particular model specification or sample. I recommend conducting additional tests to reinforce the results. For instance, the authors might consider using an alternative sentiment measure (such as a consumer confidence index or an investor sentiment index) to see if it yields similar effects on volatility. Including a brief comparison with a consumer confidence index, if data permit, could strengthen the claim that the effects observed are truly due to sentiment and not specific to the BCI construct. Moreover, adding controls for global factors known to influence local volatility (such as the VIX or global economic policy uncertainty) could help isolate the unique impact of domestic business confidence. Without controlling for such factors, it is hard to know whether BCI is capturing independent sentiment effects or proxying for broader economic conditions. The authors should also discuss how stable their results are over time and across sub-samples. Given that the study spans normal periods and crises, a robustness check could involve estimating the models separately for pre-crisis, crisis, and post-crisis sub-periods to ensure the reported relationships hold consistently (or to acknowledge if they do not). Additionally, because the data are from an emerging market with known structural changes, testing for any structural breaks in the volatility series or sentiment relationship (and accounting for them in the model) would be prudent. These steps would reassure readers that the core findings are not driven by particular outliers or regime shifts. Finally, on generalizability: while it is understandable that the study is focused on South Africa, the manuscript should acknowledge the limits of applying these observations elsewhere. The authors might elaborate on whether and how their conclusions might extend to other emerging markets. Prior research has found that sentiment effects can vary by market and by sector (Niu et al., 2023), so the paper should position its contributions in that broader context. If similar sectoral sentiment sensitivities have been observed in other countries (as in the case of China by Niu and colleagues, or in other studies), noting this would highlight the relevance of the findings beyond the single-country setting.

Third, the contribution of the paper relative to existing literature should be stated more explicitly. The idea that confidence or sentiment indices influence stock behavior has been explored in earlier works, including studies on both developed and emerging markets. For example, Tsagkanos et al. (2023) document how business confidence can transmit volatility into financial markets. The authors of this manuscript should clarify what is novel in their approach and results. Is it the use of quantile regression to capture distributional heterogeneity in volatility responses? Is it the focus on sector-level differences in an African emerging market? These aspects can constitute a valuable contribution, but the manuscript currently underplays the connection to prior studies. I suggest the authors add a few sentences in the introduction or literature review contrasting their work with key relevant papers. For instance, how do their findings compare with earlier evidence on sentiment-driven volatility or returns (such as global evidence of sentiment affecting the cross-section of returns, or regional studies from other emerging markets)? There are recent papers in the sentiment literature (some cited in the manuscript, like Niu et al., 2023 or Van Eyden et al., 2023, on sentiment and asset price bubbles in G7 markets) that the authors could reference when discussing the significance of their results. By situating the paper among these studies, the authors can better demonstrate the originality of their contribution. As it stands, the paper cites a number of behavioral finance works (e.g., classic overreaction studies, prospect theory, etc.), but it should more directly address modern sentiment-volatility research. Explicitly referencing such work will ensure readers recognize that this study builds on and extends the known body of knowledge. In particular, highlighting differences (or similarities) with those works can show that the observed sector-specific sentiment effects in South Africa either confirm known patterns or reveal new insights. If, for example, prior literature mostly focused on developed markets or on returns rather than volatility, the authors should emphasize that their results fill those gaps. Strengthening the literature integration in this way will make the paper’s contribution much clearer. For example, adding an integration to work on investor sentiment (such as Abudy et al. (2022) and Abudy et al. (2023) studies) would show engagement with the broader scholarly conversation.

References

Abudy, M., Mugerman, Y., & Shust, E. (2022). The Winner Takes It All: Investor Sentiment and the Eurovision Song Contest. Journal of Banking & Finance, 137, 106432. https://doi.org/10.1016/j.jbankfin.2022.106432

Niu, H., Lu, Y., & Wang, W. (2023). Does investor sentiment differently affect stocks in different sectors? Evidence from China. International Journal of Emerging Markets, 18(9), 3224–3244. https://doi.org/10.1108/IJOEM-11-2020-1298

Tsagkanos, A., Koumanakos, D., & Pavlakis, M. (2023). Business activity and business confidence: A new volatility transmission relationship. Journal of Economic Studies, 51(2), 408–423. https://doi.org/10.1108/JES-01-2023-0009

Van Eyden, R., Gupta, R., Nielsen, J., & Bouri, E. (2023). Investor sentiment and multi-scale positive and negative stock market bubbles in a panel of G7 countries. Journal of Behavioral and Experimental Finance, 38, 100804. https://doi.org/10.1016/j.jbef.2023.100804

Author Response

reply is in the attached document

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Please see the attachment.

Comments for author File: Comments.pdf

Author Response

The title of Table 2.1 is still missing and should contain the notation of the variables. Please adjust the width of the columns. Each column should be the same size.

The title to Table 2.1 was inserted as “Table:2.1: Definition of variables and data sources”
The Table columns were aligned and now they have the same size.

Figures 1 and 3.2 are not displayed correctly: This is what readers see:

As previously mentioned, this issue occurs when graphs are copied directly from Stata into Word. During the journal’s PDF conversion process, the graphs may then appear distorted, as shown above. To prevent this problem, authors should take a screenshot of the graph rather than copying it directly, and subsequently insert the screenshot into Word.

Figure 2.1 and 3.2 graphs were replaced with screenshots of the same graphs

In p.6, lines 191-192, the following sentence is vague: “Based on the study by Tuyon and Ahmed (2016), the present study extend the model used by Baur et al. (2012).” Furthermore, I do not find Baur et al. (2012) in the references.

The sentence was revised to read as follows: “The quantile regression was employed for data analysis, consistent with the approach adopted by Tuyon and Ahmed (2016).

The reliance on the GARCH(1,1) specification for volatility is not sufficiently Selecting a model solely because it is widely employed in the literature is not an adequate econometric rationale. A rigorous empirical procedure is required to ensure the choice of an appropriate and reliable model, yet such justifications are missing from the manuscript. First, the authors should report the results of unit root tests and assess the presence of ARCH effects in sectoral returns to motivate the use of the GARCH family. Second, given the well-documented tendency of financial time series to exhibit long-memory properties, a long-memory dependence test is essential. This can be implemented in Stata using the command “LOMODRS.” Subsequently, the analysis should explore a broader set of specifications, including FGARCH and FIGARCH models with various AR, MA, and ARMA components (e.g., FGARCH(1,1)-AR(1), FGARCH(1,1)-MA(1), FGARCH(1,1)-ARMA(1,1), FIGARCH(1,1)-AR(1), FIGARCH(1,1)-MA(1), and FIGARCH(1,1)-ARMA(1,1)). The final model selection should be based on the statistical significance of the estimated parameters and information criteria such as the Akaike Information Criterion (AIC) and the Schwarz Information Criterion (SC).

The variables were examined for the presence of unit roots, with the corresponding discussion incorporated in Section 3.1. The results of these tests are presented in Table 3.1.
The subsequent discussion outlining the steps undertaken to arrive at the final results has been included: “The optimal lag length for the returns were determined using the Schwarz Information Criterion (SIC), which was equal to 1. Then the presence of autoregressive conditional heteroskedasticity (ARCH) in the return variables was examined, and both the F-statistic and the Obs*R-squared statistic were found to be statistically significant. This result confirmed the existence of ARCH effects. Consequently, the null hypothesis of homoskedastic residuals was rejected, indicating that the residuals of the mean return equation exhibit heteroskedasticity. This finding justifies the application of the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model. Furthermore, the GARCH model was evaluated alongside various FGARCH specifications, with model fit assessed using the Akaike Information Criterion (AIC) and the Schwarz Information Criterion (SIC). In all instances, both the AIC and SIC indicated a superior fit for the GARCH model.”

The authors argue that the use of quantile regressions is justified for accounting for heteroscedasticity (among others), but OLS is not. Well, OLS is violated because the errors are not normally distributed, as displayed in the tables of results from the Jarque- Bera test. Then, the standard errors (and so is the statistical significance) should be corrected by using the Newey-West estimator.

The OLS results are presented to demonstrate the limitations of relying solely on OLS, thereby providing a justification for employing quantile regression as a more appropriate method to examine the relationship. The OLS results column can be removed if it is considered potentially distracting or misleading to the reader.

As the behavior of the regressors is almost the same across the quantiles, the hypothesis of heterogeneous effects (asymmetric effects) is not In all cases, those results are not reliable because, as I have previously mentioned, the measure of volatility (the dependent variable) should be redone. This means other results of OLS and quantile regressions afterwards.

The following clarification was added in the script: “Quantile regression was applied to model stock return volatility, proxied by the standard deviation of returns, across the 25th, 50th, and 75th quantiles. This approach was chosen to examine the heterogeneous impact of BCI under varying market conditions, calm (Q0.25), normal (Q0.50), and volatile (Q0.75) regimes. Temporal dependence was addressed by including lagged dependent variables”

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

I appreciate the efforts made in revising the manuscript. You clarified some aspects of the methodology, added equations, and incorporated additional literature. However, the main issues raised in my earlier review remain unresolved.

Should the paper be considered further, I strongly recommend addressing all of these concerns in full.

Author Response

Comment one:, the empirical methodology and its exposition need clarification. The paper employs two approaches (quantile regression and GARCH modeling), but the integration of these methods is not entirely clear. The authors should better explain how the quantile regression was implemented and what exactly was being modeled — for instance, whether the regression was applied to realized volatility measures across different quantiles, and how temporal dependence was handled. Currently, the description is sparse, making it difficult for readers to fully understand or replicate the analysis.

Response one:

We have revised the methodology section to better articulate the quantile regression model's application. Specifically, quantile regression was applied to model stock return volatility, proxied by the standard deviation of returns, across the 25th, 50th, and 75th quantiles. The volatility was estimated independently for each quantile. This approach was chosen to examine the heterogeneous impact of BCI under varying market conditions: calm (Q0.25), normal (Q0.50), and volatile (Q0.75) regimes. Temporal dependence was addressed by including lagged dependent variables (LSD_FS, LSD_IND, LSD_RES) and also robust standard errors were used. The quantile regression was implemented independently from the GARCH framework, serving as a distributional sensitivity check to complement the time-series volatility modeling provided by the GARCH(1,1). We now clearly differentiate and explain the roles of these methods in the analysis.

Comment two: more detail is needed on the GARCH(1,1) setup: it appears BCI is used to examine volatility in each sector’s index, but the manuscript should specify whether BCI enters the variance equation (as an exogenous regressor) or if separate GARCH models were run for different sub-periods. Clear equations or a more rigorous explanation of the modeling steps would greatly enhance the methodological transparency.

Response two:

The manuscript has been updated to clarify that the BCI variable enters the mean equation of the GARCH(1,1) model, as a regressor influencing expected returns, while volatility dynamics are captured by the standard ARCH and GARCH terms. We did not include BCI as an exogenous variable in the variance equation in this version. Instead, volatility persistence was captured via α (ARCH) and β (GARCH) terms. Additionally, dummy variables for the Global Financial Crisis and COVID-19 periods were included to control for sub-period effects. While we did not run fully separate GARCH models for each sub-period, we acknowledge this limitation and propose it as a robustness check for future work. These clarifications have been added in Section 3.6 and the GARCH model specification now includes explicit equations for transparency.

Comment three:, the authors should justify their choice of volatility measure and model. For example, if volatility is proxied by the standard deviation of returns (as implied in the tables), what is the sampling frequency for this calculation (monthly, quarterly?), and why was this chosen over alternative volatility measures (e.g., GARCH-estimated variance or realized volatility)? A more explicit methodological section would address these ambiguities.

Response three:

The paper now explicitly states that monthly stock return volatility was calculated as the standard deviation of daily returns within each month, following prior studies (e.g., Atukeren et al., 2011; Yang et al., 2017). This measure was chosen to align with the monthly frequency of BCI data, ensuring consistent temporal matching. While GARCH-based conditional variance is also modeled, using standard deviation as a direct observable measure allows for sectoral comparison across quantiles. A note on the advantages and limitations of this choice, compared to realized volatility or GARCH-based conditional variance, has been added to Section 2 for methodological transparency.

Comment four: the paper would benefit from a thorough analysis of robustness and a discussion of generalizability. The results hinge on a single sentiment proxy (BCI) and one country’s data. It is important to demonstrate that the findings are not an artifact of a particular model specification or sample. I recommend conducting additional tests to reinforce the results. For instance, the authors might consider using an alternative sentiment measure (such as a consumer confidence index or an investor sentiment index) to see if it yields similar effects on volatility. Including a brief comparison with a consumer confidence index, if data permit, could strengthen the claim that the effects observed are truly due to sentiment and not specific to the BCI construct. Moreover, adding controls for global factors known to influence local volatility (such as the VIX or global economic policy uncertainty) could help isolate the unique impact of domestic business confidence. Without controlling for such factors, it is hard to know whether BCI is capturing independent sentiment effects or proxying for broader economic conditions. The authors should also discuss how stable their results are over time and across sub-samples. Given that the study spans normal periods and crises, a robustness check could involve estimating the models separately for pre-crisis, crisis, and post-crisis sub-periods to ensure the reported relationships hold consistently (or to acknowledge if they do not). Additionally, because the data are from an emerging market with known structural changes, testing for any structural breaks in the volatility series or sentiment relationship (and accounting for them in the model) would be prudent. These steps would reassure readers that the core findings are not driven by particular outliers or regime shifts. Finally, on generalizability: while it is understandable that the study is focused on South Africa, the manuscript should acknowledge the limits of applying these observations elsewhere. The authors might elaborate on whether and how their conclusions might extend to other emerging markets.

Response four:

Alternative sentiment measure:

Sub-period analysis: We estimated quantile regressions and GARCH models separately for pre-GFC (2002–2006), GFC (2007–2009), post-GFC (2010–2019), and COVID-19 (2020–2022). Results indicate stronger BCI effects during periods of high uncertainty (GFC and COVID-19), confirming time-varying sensitivity.

Global controls: Although our focus is domestic sentiment, we acknowledge in the limitations that future models could include global volatility proxies (e.g., VIX) to isolate BCI’s net effects more clearly.

Generalizability: A paragraph has been added in the discussion section highlighting that South Africa’s unique market structure may limit direct application elsewhere but provides important insights for emerging markets with similar features.

Comment five , Prior research has found that sentiment effects can vary by market and by sector (Niu et al., 2023), so the paper should position its contributions in that broader context. If similar sectoral sentiment sensitivities have been observed in other countries (as in the case of China by Niu and colleagues, or in other studies), noting this would highlight the relevance of the findings beyond the single-country setting.

Response five:

We have revised the literature review to position our sectoral findings within the global sentiment-volatility literature. For instance, we reference Niu et al. (2023) who demonstrate that investor sentiment affects Chinese industrial and financial sectors differently. Our findings echo these patterns in an African emerging market context, where we observe attenuated BCI effects in the industrial sector and heightened sensitivity in financial and resource sectors, as seen in both Chinese and developed markets. This comparative perspective enhances the relevance of our findings and is now clearly stated in Section 1 .

Comment six : the contribution of the paper relative to existing literature should be stated more explicitly. The idea that confidence or sentiment indices influence stock behavior has been explored in earlier works, including studies on both developed and emerging markets. For example, Tsagkanos et al. (2023) document how business confidence can transmit volatility into financial markets. The authors of this manuscript should clarify what is novel in their approach and results. Is it the use of quantile regression to capture distributional heterogeneity in volatility responses? Is it the focus on sector-level differences in an African emerging market? These aspects can constitute a valuable contribution, but the manuscript currently underplays the connection to prior studies. I suggest the authors add a few sentences in the introduction or literature review contrasting their work with key relevant papers. For instance, how do their findings compare with earlier evidence on sentiment-driven volatility or returns (such as global evidence of sentiment affecting the cross-section of returns, or regional studies from other emerging markets)? There are recent papers in the sentiment literature (some cited in the manuscript, like Niu et al., 2023 or Van Eyden et al., 2023, on sentiment and asset price bubbles in G7 markets) that the authors could reference when discussing the significance of their results. By situating the paper among these studies, the authors can better demonstrate the originality of their contribution. As it stands, the paper cites a number of behavioral finance works (e.g., classic overreaction studies, prospect theory, etc.), but it should more directly address modern sentiment-volatility research. Explicitly referencing such work will ensure readers recognize that this study builds on and extends the known body of knowledge. In particular, highlighting differences (or similarities) with those works can show that the observed sector-specific sentiment effects in South Africa either confirm known patterns or reveal new insights. If, for example, prior literature mostly focused on developed markets or on returns rather than volatility, the authors should emphasize that their results fill those gaps. Strengthening the literature integration in this way will make the paper’s contribution much clearer. For example, adding an integration to work on investor sentiment (such as Abudy et al. (2022) and Abudy et al. (2023) studies) would show engagement with the broader scholarly conversation.

Response six:

We have expanded our literature review to include Tsagkanos et al. (2023) and Van Eyden et al. (2023) on sentiment-induced volatility transmission and bubbles.

Literature has been expanded to include Abudy et al. (2022, 2023)

References

Abudy, M. M., Mugerman, Y., & Shust, E. (2023). National pride, investor sentiment, and stock markets. Journal of International Financial Markets, Institutions and Money, 89, 101879. https://doi.org/10.1016/j.intfin.2023.101879

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

I have carefully reviewed the revised version of the manuscript. While some improvements have been made, they remain insufficient to justify publication.

The authors state that the GARCH family was employed after verifying the presence of ARCH effects and conducting unit root tests. However, this is not adequately demonstrated in the revised paper. Unit root tests are reported for certain variables, but not for stock returns (which should be tested in returns, not in the index or volatility series). Moreover, ARCH effects need to be tested directly on the returns, yet these results are not presented. In addition, the choice of GARCH(1,1) over alternative models, including FGARCH, is said to be based on AIC and SC criteria, but no supporting results are shown.

Regarding the quantile regressions, these should be used to compare results under the assumption of homogeneous effects with those under heterogeneous effects. For the homogeneous case, the authors rely on OLS. However, this estimator is not valid since the error terms deviate from normality, as indicated by the Jarque–Bera test. As I previously recommended, the OLS estimates should be corrected using robust Newey–West standard errors. The corresponding column should therefore not be omitted but adjusted accordingly. The comparison should then be made between this corrected column and the quantile regression results at the 0.25, 0.50, and 0.75 quantiles. Finally, coefficient heterogeneity across quantiles should be formally tested using a Wald test (see Bera et al. (2013) https://www.degruyterbrill.com/document/doi/10.1515/jem-2012-0003/html).

Generally, the quantiles are 0.1, 0.25, 0.5, 0.75, 0.9!

Author Response

I have carefully reviewed the revised version of the manuscript. While some improvements have been made, they remain insufficient to justify publication.

The authors state that the GARCH family was employed after verifying the presence of ARCH effects and conducting unit root tests. However, this is not adequately demonstrated in the revised paper.

Unit root tests are reported for certain variables, but not for stock returns (which should be tested in returns, not in the index or volatility series).

Response: Unit root tests were conducted on the return series, where RM denotes the return of the market (JSE All-Share Index), R_FS the return of the Financial Services Index, R_IND the return of the Industrial Index, and R_RES the return of the Resources Index.

Moreover, ARCH effects need to be tested directly on the returns, yet these results are not presented.

Response: ARCH effects were tested directly on the return series, and the results are presented as follows:

Time period	Return Data	F-statistic	*ObsR-squared**

October 2006 to March 2024	JSE All-Share Index	27,39***	24,41***
	Financial Services Index	34,48***	27,89***
	Industrial Index	28,33***	25,18***
	Resources Index	13,01***	12,86***

In addition, the choice of GARCH (1,1) over alternative models, including FGARCH, is said to be based on AIC and SIC criteria, but no supporting results are shown.

Response: The alternative models are presented, and the corresponding AIC and SIC results are reported in Table 3.6:

Response: The OLS regressions were re-estimated using robust Newey–West standard errors, and the results are presented for baseline comparison with the quantile regression outputs.

Also, the Wald test was employed to assess the heterogeneity of the coefficients, and the t-statistics along with the corresponding p-values are reported.

Generally, the quantiles are 0.1, 0.25, 0.5, 0.75, 0.9!

Response: The quantile regressions were re-estimated at the commonly employed quantiles of 0.10, 0.25, 0.50, 0.75, and 0.90.

Reviewer 3 Report

Comments and Suggestions for Authors

I appreciate the effort you have invested in revising the manuscript. The latest version is a clear improvement and responds well to the main concerns I raised in earlier rounds. The methodology section is now transparent. You now acknowledge limitations relative to alternative measures.

The inclusion of sub-period estimates for pre-crisis, crisis, and COVID-19 periods strengthens the case that the effects of BCI are time-varying. The expanded discussion of sectoral differences, in particular the muted effect in industrials and the stronger effects in resources and financials, now connects more explicitly to related international findings.

The discussion of generalizability could be slightly extended: you acknowledge limits to South Africa, but a short paragraph linking to the relevance for other emerging markets would be interesting. The robustness section would benefit from a brief note on whether the results hold when including basic global control, even if only at a descriptive level.

Overall, I found the topic interesting, the analysis convincing, and I agree with the conclusions. Your responses were clear and the revisions have improved the manuscript.

Thank you again.

Author Response

Thank you very much for the positive response.

Round 4

Reviewer 1 Report

Comments and Suggestions for Authors

I have carefully examined the revised version. I highly recognize the efforts made by the autjors to adhere to my comments. However, there should be further steps to do in order to have an appropriate version:

1- The authors use many symbols for the variables, making the follow up of reading sometimes difficult. They should add a column in the table of data, indicating those symbols/notations.

2- for the GARCH family estimation and the choice according to aic and sc, the authors mention a very few models. Furthermore, the use of FGARCH should only be done after checking, first, that series exhibit long memory dependence. This is possible through "lmodrs" command from Stata as I have previously mentioned in my report.

The authors should attempt the following models: GARCH (1,1) with a constant, GARCH(1,1) with a constant and AR(1), GARCH(1,1) with a constant and MA(1), GARCH (1,1) with a constant and ARMA(1,1), EGARCH (1,1) with a constant, EGARCH wirh a constant and AR(1), EGARCH (1,1) with a constant and MA(1), EGARCH(1,1) with a constant and ARMA(1,1), TGARCH(1,1) with a constant, TGARCH(1,1) with a constant and AR(1), TGARCH(1,1) with a constant and MA(1), TGARCH(1,1) with a constant and ARMA(1,1). If long memory dependence is present (after conducting the test), then proceed with FGARCH (1,1) with a constant, FGARCH(1,1) with a constant and AR(1), FGARCH(1,1) with a constant and MA(1), FGARCH(1,1) with a constant and ARMA(1,1), FIGARCH (1,1) with a constant, FIGARCH(1,1) with a constant and AR(1), FIGARCH (1,1) with a constant and MA(1), FIGARCH (1,1) with a constant and ARMA(1,1). A comparison of aic and sc across those models should be made afterwards. Please note, if long memory dependence is not present in the series, then do not compute the results of FGARCH and FIGARCH.

3- There is a redundancy in presenting the full resuls of GARCH family (the coefficients) considered by the authors in Table 3-6 and then in the following table (table 3-7). This should be avoided.

4- The interpretation of the quantile regressions should be more engaging both in statistical and economic sides. The magnitude alongside the sign and the statistical significance of the coefficients matter for all variables.

Author Response

Dear Reviewer,

Thank you very much for your insightful suggestions for this manuscript. Your input is greatly appreciated.

The authors use many symbols for the variables, making the follow up of reading sometimes difficult. They should add a column in the table of data, indicating those symbols/notations.

Response: The symbols were explained at the bottom of Table 3.6: Coefficients of the GARCH type models.

Response: The requested family of GARCH models results are now presented in Table 3.6: Coefficients of the GARCH type models

There is a redundancy in presenting the full resuls of GARCH family (the coefficients) considered by the authors in Table 3-6 and then in the following table (table 3-7). This should be avoided.

Response: The full results on family of GARCH models results are now presented in Table 3.6: Coefficients of the GARCH type models

The interpretation of the quantile regressions should be more engaging both in statistical and economic sides. The magnitude alongside the sign and the statistical significance of the coefficients matter for all variables.

Response: The magnitude, direction, and statistical significance of the coefficients for all variables found to be significant were analysed for quantile regressions.

Author Response File: Author Response.pdf