Currently, financial markets are being analyzed from the perspective of Behavioral Finance, a subfield of Behavioral Economics that considers aspects of the Psychology and Sociology of Finance, which emerged from several authors' criticism of Classical Finance, including 1. Research shows that investors do not necessarily think rationally, but are also guided by emotions, subjective thoughts, and sometimes the so-called herding mentality 23. Herding behavior is considered one of the most interesting concepts in Behavioral Finance. This concept is not recent, as it was already mentioned in 4.

Herding in financial markets is defined as an imitation, a convergence of action 5 and can be explained as a psychological tendency to follow in the footsteps of others while ignoring one's own skills 6.

In the realm of financial markets, understanding herding behavior among investors is crucial for comprehending market dynamics and predicting systemic risks. Initially recognized for its potential to amplify market movements, herding behavior has been studied extensively to uncover patterns of collective decision-making among investors. Researchers have developed various methodologies to measure and analyze herding, ranging from early metrics focusing on dispersion relative to market returns 27 to sophisticated models exploring the interplay of social learning and economic indicators 89).

Financial markets fluctuate over time. Market anomalies and major deviations from stock market efficiency are likely to be facilitated or even generated during crisis situations, with significant consequences for optimal asset allocation, portfolio diversification, and financial stability in general 10. Thus, searches for the correlation between crises, and the evolution of financial markets have become frequent. Likewise, studies targeting investors, specifically in relation to the rationality of capital allocation and its behavior, have increased dramatically.

According to 7, herding behavior is more common during financial crises, and it can cause prices to deviate from fundamentals. Investors panic when the market is strained during a financial crisis, and they tend to have a free ride on the market information.

The existence of herding behavior in financial markets during crises has been extensively studied by various authors. For instance, regarding the 2008 crisis, 11 analyze investors' herding activity across 18 countries, categorizing them into three groups: advanced stock markets, Latin American markets, and Asian markets. Similarly, 12 examine herding behavior in the Chinese and Indian stock markets, finding evidence of such behavior in both.

More recently, in the context of the Covid-19 crisis, several studies have emerged. Among others, 13 use the Cross-Sectional Standard Deviation (CSSD) of returns and a State Space model to identify herding behavior in the Vietnamese and Taiwanese stock markets. 14 conduct an empirical analysis using daily stock market data from 72 countries, including both developed and emerging economies, for the first quarter of 2020. Their results indicate evidence of investor herding in these markets. 15 also investigate herding behavior triggered by the Covid-19 outbreak in 2020, focusing on 6 Asian stock markets. They employ CSSD and Cross-Sectional Absolute Deviation (CSAD) as key indicators, finding a clear presence of herding from February 2020 to January 2021, with a sharp increase during the market crash in March 2020. Likewise, 16 study herding behavior in the Vietnamese stock market, using the CSAD method and quantile regression, and detect such behavior.

Additionally, several papers have examined both the 2008 and Covid-19 crises, including, e.g. 1 , 17, 18, 19, 20, and 21. However, our study distinguishes itself by covering more countries and a longer time period with regard to the Covid-19 crisis. For instance, 17 focus on a sample of 10 countries, 18 examine only 3 countries, and 19 focus solely on Egypt. 21 restrict their analysis to Brazil, Russia, India, China, and South Africa, while 20 limit their study to China and the United States of America.

Therefore, to the best of our knowledge, there are no works that analyze the effect of the Covid-19 crisis on herding behavior once this crisis finished for a big number of countries with a comparison of this crisis with the 2008 one. Therefore, this article delves into herding behavior amidst financial crises, focusing on the global financial crisis of 2008, the Covid-19 pandemic and comparing both.

An evaluation of this scale can be useful to obtain a general representation of the main global markets as well as to facilitate the ability to compare and contrast market behavior between countries. Therefore, this study aims to verify the existence of herding behavior during the 2008 and Covid-19 crises and analyze the impact of volatility and trading volume in markets that have not yet been thoroughly investigated over an extended period. Additionally, it incorporates recent data. We apply a model to detect herding in 31 main markets, and differentiate between high volatility and low volatility, and high and low trading volume on a large scale. The primary contribution of this article is to provide additional empirical evidence from the world's major financial markets. By analyzing data across different regions and periods, we aim to deepen the understanding of herding behavior. This comprehensive approach not only broadens the scope of existing research but also offers valuable insights that can inform both academic theory and practical policy-making in the context of global finance amid crises like the global financial crisis in 2008 and the Covid-19 crisis in 2020.

The paper is structured as follows. After this Introduction, Section 2 presents a brief literature review, particularly focusing on herding behavior during the 2008 and Covid-19 crises and states the hypotheses our work tests. Section 3 describes the data sample and the research methodology used in the study. Section 4 summarizes and discusses the results, while Section 5 explores the conclusions and provides suggestions for further research.

Herding behavior is the result of an intention or action by one group of investors to imitate or copy the behavior of another group 5. 22 defined it as the correlated movement of investors, which present investment decisions similar to a particular group. The herding effect on the financial market is also marked by a homogenization of the activities of its members, who act in the same way at a given time. In other words, it occurs when a market agent attempts to follow the herd despite having a different viewpoint 2324. It can also occur when investors prefer to follow the market consensus above their own personal information and ideas 2. According to 25, this conduct is driven by emotions, and it frequently occurs because of societal pressure to comply. Another reason given is the notion that a vast number of people cannot all be wrong.

This movement has been studied extensively in a variety of international contexts (stock market, bond market, derivatives market, commodities market, exchange rates, mutual funds, hedge funds), referring to institutional investors, analysts, individual investors, and financial markets in both developed and developing/emerging markets 2611 1 For a brief description of the content of these articles, see Section 2.

3 271028.

Herding can occur in the event of markets' uncertainty 12 and this uncertainty increases during times of crisis, when most investors panic and strive to protect the value of their investments. As a result of this scenario, investors reduce their confidence when allocating investments and this may lead to greater volatility in the market and high trading volume.

As for volatility, some studies verified that herding behavior is more pronounced under high volatility markets 2629Arjoon et al., 2020). When using a herding model, 30 found a positive link between transaction volume and excessive volatility. Meanwhile, 10 found that herding is more present during volatility market conditions.

Regarding trading volume, 31 examined its impact and reported that herding is present during high trading volume periods. 32 also found a positive and significant correlation between market trading volume and herding. 33 used trading volume as an investment sentiment gauge, arguing that when investors are more optimistic, they bet on rising stocks and contribute liquidity to the market, resulting in increased trading volume. During periods of pessimism or crisis, on the other hand, investors were found to avoid trade altogether.

The global financial crisis of 2008 and the Covid-19 pandemic are two prominent instances of times when herding behavior became very noticeable. Because investors were so fearful and uneasy during these crises, they followed the herd, which increased market volatility as well as trading volume. These two crises can be compared to get important insights into how herding behavior appears and affects market dynamics under high stress. This provides a rare opportunity to assess the influence of an unanticipated and dreaded disease on the behavior of investors in increasingly interconnected stock markets (Maquieira and Espinosa-Mendéz, 2022;18.

Several studies have focused on herding behavior during the 2008 and Covid-19 crises. It is beyond the scope of this article to give an extensive description of all those studies. However, it is worth highlighting some of them to better identify the research gap that our work aims to fill.

Thus, 17 investigate herding in 10 equity markets from January 2001 to August 2021 using a methodology that considers movements in assets due to changes in fundamentals. They find heterogeneous patterns in herding across the 10 countries during the pandemic, with limited evidence of herding overall. However, Italy, Sweden, and the United States of America displayed signs of herding. The authors note that fear, uncertainty, and rapid information dissemination during crises could lead to significant deviations from rational market behavior.

On the other hand, for the period January 2001 -June 2021, and using a modified herding model with the Kalman filter and GARCH methodology, 18 investigate the presence of herding in the United States of America, China, and Taiwan and find that investors exhibited herding behavior during the 2008 crisis, but not during the Covid-19 crisis.

In the case of Egypt, 19 check for the existence of herding for the whole period from January 2003 to December 2022. They employ the CSSD and the CSAD models and, additionally, use the quantile regression approach. For the whole period, they find evidence of herding behavior only in down-market conditions using the CSAD model. Conversely, when the market was up, herding behavior was absent. Therefore, when the market was down, investors were afraid of the condition of uncertainty, neglecting their own private information, avoiding acting independently and consequently, following other investors; and when the market was up, investors became rational and acted fully independent. Moreover, when the whole period is split into subperiods and, among others, the 2008 and Covid-19 crises are considered, the authors find evidence of herding before, after and during the five significant crises examined in the study, except for 2008 crisis where no herding behavior was observed.

Likewise, based on data starting from January 2005 to May 2020 for the 2008 crisis, and from January 2019 to December 2021 for the Covid-19 crisis, 20 investigate the impact of these two global crises on herding behavior in the stock markets of China and the United States of America. They find no evidence of herding in the United States of America during these crises but significant herding in the Chinese stock market during the Covid-19 crisis. Their results highlighted the differences in the effects of financial and public health crises on herding behavior and the variations between emerging and developed stock markets.

Finally, the paper by 21 H1. There is significant presence of herding behavior in the analyzed period. H2. Herding behavior is more prominent in times of crisis. H3. Market volatility has a significant effect on herding behavior. H4. Trading volume has a significant effect on herding behavior.

Regarding the data collecting, sample selection, and evaluation processes for the objective of this article, the data used for the study has been obtained from Refinitiv Datastream. The sample consists of daily adjusted closing prices of stocks listed in the most capitalized markets (current US$), as included in the World Bank (2021). To avoid compromising the results due to lack of data in the observations and to have reliable regression results, the following criteria has been adopted: 1) Only countries with data from January 2000 onwards have been included in the study. 2) Following recommendations by 34 and 35, indexes and companies whose stocks lack available data for 10 consecutive years have been excluded from the sample. 3) Single-stock trading days have not been incorporated in the sample.

To ensure the accuracy and reliability of our analysis, we have taken into consideration the dynamic nature of the index compositions over the sample period. Specifically, we have accounted for additions and deletions of constituents within the indices. By incorporating these changes, we mitigate the risk of survivorship bias, which could otherwise skew the results. This comprehensive approach ensures that our findings reflect the true performance and behavior of the indices over time. As a result, the following countries or markets have been selected: Argentina, Australia, Brazil, Canada, Chile, China, Denmark, Egypt, Finland, France, Germany, Hong Kong, India, Indonesia, Ireland, Israel, Italy, Japan, Mexico, Norway, Portugal, Qatar, Russia, Saudi Arabia, South Africa, Spain, Sweden, Turkey, the United Arab Emirates, the United Kingdom, and the United States of America for the period from 02 January 2000 to 05 May 2023.

The selected countries, based on their high market capitalization (World Bank, 2021) and the criteria outlined in points 1) to 3), represent a significant portion of global stock markets. This diverse selection ensures that the findings are comprehensive and applicable across a wide range of financial contexts.

Data has been adopted on a daily basis due to the sensitivity of the information to reflect any movement in the financial market, and because this reflects change more efficiently than utilizing weekly or monthly data, as adopted in several studies 261112273637.

The final sample is composed of 31 markets and 195.174 observation days, as per Table 1.

In order to examine our sample and test the hypotheses described in Section 2, we split the sample into 4 periods: Several methods have been proposed in the literature to measure herding behavior. Initially, 38 introduce the LSV (Lakonishok, Shleifer, and Vishny) indicator, but its inability to capture intertemporal herding behavior was noted by some authors, e.g., 39. Subsequently, using measures of dispersion in relation to market returns during periods of significant market changes or times of crisis, 2 introduce the CSSD measure. Inspired by the CSSD measure, a widely used method proposed by 7 examines herding behavior based on the degree of return dispersion, which is measured by the CSAD of returns. According to the CAPM, 7 demonstrate a positive linear correlation between CSAD and stock market return in a rational market. However, this linear relation is disrupted by herding behavior, leading to a nonlinear relationship. Therefore, the relationship between market return dispersion and market return rate serves as an indicator for identifying the presence of herding behavior:

being CSAD t the CSAD at time t defined as:

N the number of assets; R X ,t the return of stock X at time t calculated as a continuous rate: According to 7, a market is in equilibrium when CAPM holds, and CSAD t , as the measure of return dispersion, should be linearly related to average market return R m, t . This is represented in equation (1) Unlike CSSD, CSAD's emphasis on the proximity of individual returns to market averages improves its sensitivity to market movements 640. Therefore, in this paper, we use the CSAD of returns method. Based on equation (1), H1 is tested by considering the whole period analyzed. In order to test H2, the same equation is estimated but, in this case, using the data related to the 4 periods the whole period has been split into.

As far as the hypothesis H3 is concerned, we examine the role of volatility on herding by splitting the sample into high and low volatility days. According to 26, high volatility is defined as a day's volatility above the last 30-day moving average, and vice versa. In accordance with that definition, we first calculate the average of the market returns for the last 30 days, and then its standard deviation, according to the following formulas:

being σ t ❑ = √ σ t 2 the volatility of day t . In order to verify if the volatility of day t is higher or lower/equal than the previous 30 days, we calculate the average of the volatilities of the last 30 days and compare it with the volatility of the respective day. The average of the volatilities of the last 30 days is given as:

the day t is considered of high volatility, otherwise it is low/equal.

Then, equation (7) and equation (8) allow us to replicate equation (1) but with respect to high and low volatility days, respectively:

where

) is the market return during day t when the volatility is high (low/equal).

Instead of estimating equations (7) and (8) in equations (7) and (8). Specifically, the null hypothesis to be tested is

. This hypothesis can be assessed by determining whether the coefficients β 3 and β 4 in regression (9) are equal. Conducting the Wald test prior to analyzing the impact of volatility ensures the statistical validity of our subsequent findings.

where D t σ 2, H is a dummy variable equal to 1 when volatility is high and 0 otherwise.

According to 27 Therefore, we only verify herding behavior under high and low volatility by using equations (7) and (8) for those markets where the Wald test shows statistical significance.

Similarly, to test hypothesis H4, trading volume is characterized as high if on day t it is greater than the previous 30-day moving average and low/equal if it is less/equal than the previous 30-day moving average. Then, we estimate equation (10) and equation (11) hence:

where V , H and V , L refer to high and low/equal trading volume in the day t .

Again, instead of estimating equations (10) and (11) . This hypothesis can be tested if coefficients δ3 and δ4 in regression (12) are equal:

where D t V , H = 1 when trading volume is high and 0 otherwise.

We only estimate equations (10) and (11) to check the influence of herding behavior under high/low trading volume for those markets where the Wald test shows statistical significance and, therefore, in markets where the Wald test is not statistically significant, indicating equal herding coefficients, we do not proceed with that estimation.

To perform the data analysis and draw the corresponding conclusions, we use EViews. Moreover, all equations related to regressions have been estimated by using Ordinary Least Squares (OLS) regression technique. Tables 3 and 4 report the results of the estimations of equation (1) when considering the whole sample from 02 January 2000 till 05 May 2023 and the four periods into which it has been divided, respectively. (1) for the whole period Note: Grey highlighted countries for which y 2 is negative and statistically significant; ***,** and * represent statistical significance at the 1%, 5% and 10% levels, respectively. Source: Own elaboration.

Results of herding behavior in the four periods considered as per equation (1) Note: Grey highlighted countries and periods for which y 2 σ 2, H = y 2 σ 2, L ; ***, ** and * represent statistical significance at the 1%, 5% and 10% levels, respectively, of t-statistics. Source: Own elaboration.

Regarding the volatility, results of the Wald test described in Section 3 are shown in Table 6 and 7 present the results of equations (7) and (8) and statistically significant; ***,** and * represent statistical significance at the 1%, 5% and 10% levels, respectively. Source: Own elaboration.

During high volatility periods, herding behavior was reported (Table 6) during the crisis of 2008 in Argentina, Brazil, China, Finland, Indonesia, Italy, Mexico, Qatar, South Africa, Saudi Arabia, Turkey, and the United Arab Emirates. Table 6 also shows that during the Covid-19 crisis, herding was evidenced in Argentina, Brazil, China, Egypt, India, Indonesia, Italy, Portugal, Qatar, Russia, South Africa, Saudi Arabia, Spain, Turkey, the United Arab Emirates, and the United Kingdom suggesting that volatility is a driving factor of this behavior in these markets. A characteristic that can be evidenced from these results is that most of the markets are developing or emerging markets. The results show that herd behavior was more noticeable during the Covid-19 crisis. and statistically significant; ***,** and * represent statistical significance at the 1%, 5% and 10% levels, respectively. Source: Own elaboration. Table 7 includes results during low volatility days. During the period of the 2008 crisis, China, India, Mexico, Russia, South Africa, and Saudi Arabia presented a significant negative result for the herding coefficient. In the Covid-19 crisis, Brazil, China, India, and South Africa showed presence of herding.

Based on the results described in the previous paragraphs concerning volatility, we do not reject hypothesis H3. Our findings indicate that market volatility has a significant effect on herding behavior, being more prevalent during periods of financial crisis than in non-crisis periods.

In relation to the analysis of trading volume and the presence of herding behavior, Table 8 shows the results of the Wald. It can be seen that most countries have different herding coefficients in the analyzed periods (grey highlighted those countries and periods for which the Wald test is not statistically significant, i.e., the null hypothesis is not rejected and so

When the two periods of crises are compared, we notice that the number of markets that rejected the null hypothesis are very similar in both crises.

Moreover, considering the crisis of 2008, we find that for Argentina, Australia, Brazil, China, Egypt, Finland, Germany, Hong Kong, India, Indonesia, Israel, Italy, Mexico, Norway, Portugal, Qatar, Russia, South Africa, Saudi Arabia, Spain, Sweden, Turkey, the United Arab Emirates, the United Kingdom, and the United States of America the null hypothesis of symmetric herding behavior is rejected. However, Canada, Chile, Denmark, France, Ireland, and Japan evidenced a symmetry of herding behavior. During the Covid-19 pandemic crisis, all markets rejected the null hypothesis, except Australia, Canada, Ireland, Israel, and Sweden, indicating symmetric in herding coefficients during high and low trading volume periods. is negative and statistically significant; ***,** and * represent statistical significance at the 1%, 5% and 10% levels, respectively. Source: Own elaboration.

This article aims to verify the existence of herding behavior during the 2008 and Covid-19 crises, particularly in the aftermath of the latter, and to analyze the impact of volatility and trading volume on markets that have not been thoroughly investigated over an extended period. By examining the Covid-19 crisis up to the point it ended, we provide a detailed assessment of how herding behavior evolved throughout the crisis. This approach enhances our understanding of market dynamics during critical periods and offers new perspectives for future research.

Using a sample from the main indexes of 31 financial markets over the period from January 2000 to May 2023, we find evidence of herding during the whole period, during the different periods of crises, during both high and low volatility periods, and during both high and low trading volume periods.

Throughout the entire period analyzed, Argentina, Canada, Chile, China, Egypt, Finland, India, Indonesia, Mexico, Portugal, Qatar, Russia, South Africa, Saudi Arabia, Spain, Turkey, and the United Arab Emirates exhibited evidence of herding in their markets. These results partially align with the findings of 17, 18, and 21. 17 examine herding behavior from 2001 to 2021 across 10 countries, including all those analyzed in our study, and also find that herding persisted throughout the period. Yang and Chuang (2022) identify significant evidence of herding in China, Taiwan, and the United States of America, although their study covers only the period from 2001 to 2021, without the analysis of the complete Covid-19 pandemic. 21, considering a shorter period (from 2006 to 2022) and using monthly data, observe presence of herding in Brazil, China, India, Russia, and South Africa.

Our study shows that herding behavior was more prevalent during crises, particularly throughout the Covid-19 crisis, across all observed markets. Similar findings are reported by 20, who analyze two time periods (2005-2010 and 2019-2021), and observe the greatest presence of herding during Covid-19 in China, while the United States of America showed no evidence of herding in any crisis, which aligns with our results despite their smaller sample. 17, however, find herding to be more pronounced before crises rather than during them, contrasting with our findings of increased herding during crises, which is based on a larger sample and broader set of markets. Specifically, during the Covid-19 period, they report no evidence of herding in Australia, Belgium, Japan, and the United Kingdom, while Italy, Sweden, and the United States of America exhibited herding, and Brazil and France only showed herding for shorter periods. In contrast, our study finds evidence of herding in Argentina, Canada, China, Egypt, Finland, India, Indonesia, Italy, Mexico, Portugal, Qatar, South Africa, Saudi Arabia, Spain, and the United Arab Emirates, with no herding detected in the other countries during the Covid-19 crisis. 18 report the presence of herding only in China and Taiwan during the Covid-19 pandemic and not in the United States of America, converging with our study.

When considering volatility, we find that herding was more prevalent during high volatility periods compared to low volatility periods, which is consistent with Arjoon et al. (2020). When examining the crises individually, herding was more frequent during high volatility periods of the Covid-19 pandemic and during low volatility periods of the 2008 crisis.

Regarding trading volume, herding was slightly more common during high trading volume periods than during low trading volume periods. Additionally, herding was more frequent in both crises during high trading volume days compared to low trading volume days, which is consistent with 32 and 31.

Considering the analysis of both crises in their entirety, including up to the point when they were officially declared over, our article provides substantial empirical evidence on the effects of herding behavior in markets. This comprehensive approach enhances the existing literature by offering new insights into how herding behavior evolves across different crises and its persistence until their conclusion. This contribution deepens our understanding of market dynamics and informs future research in this area.

Moreover, the increased occurrence of herding during crises such as the 2008 global financial crisis and the Covid-19 pandemic underscores the dominance of emotional responses over rational decision-making in turbulent market conditions. This study highlights the need for greater awareness and measures to mitigate market disruptions in future crises.

Future research should explore the impact of macroeconomic policies and government interventions on herding behavior during crises. Given that herding was more pronounced during periods of high volatility and trading volume, it would be valuable to examine how different policy responses influence this behavior. Additionally, investigating the role of technological advancements and digital trading platforms in shaping herding behavior could provide deeper insights into contemporary financial market dynamics. On the other hand, our study could be complemented by using alternative models or techniques to measure herding behavior, such as the CSAD quartile 41Arjoon et al., 2020) or window regression 3142.

6.