Search

This thesis investigates herding behaviour among major world stock markets from 2002 to 2018. It also studies the herding behaviour at sector level from 2001 to 2020. In the first chapter, we introduce the background and motivation for this study. In the second chapter, we review herding behaviour and relevant prior research. In chapters 3, we use the standard CCK method on a recent data sample to detect herding behaviour in a comprehensive study of the world's major stock markets that have previously been investigated for herding. This allows comparison with previous results in the literature. We have captured clear evidence of anti-herding behaviour in most of the world's major stock markets, and the presence of herding behaviour in emerging markets during larger price movements in the market. Then in chapter 4, we explain and evaluate the theoretical and empirical difference between using the log and simple return calculation methods in tests for herding. Most of the theoretical work on herding would tend to indicate that one would expect to observe herding in the financial markets. In practice, most empirical studies to date have not found this to be the case. This could be due to problems with the procedures used to test for herding. In chapter 5, theoretically and empirically, we discuss the major drawbacks of the CAPM based CCK method which is the method most commonly used to find herding in the literature. We show that the test is highly biased against finding herding. The bias arises because the test assumes that, in the absence of herding, stock prices follow the CAPM but does not account for the implications of the CAPM not being a perfect asset pricing model. We provide alternative and tractable ways to overcome the disadvantages of the CCK method. Also, we show these methods theoretically may give very different conclusions to the CCK method. In chapter 6, we then apply the new testing methods we have developed to the comprehensive world data we have previously investigated for herding using the CCK method in chapter 3. The empirical results give quite strong evidence of herding which is in contrast to our results in chapter 3 and most of the prior literature. In chapter 7, we investigate herding at the industry sector level for the major European economies of Britain, France and Germany. This allows us to detect whether certain sectors are particularly likely to herd. We can also detect how different sectors react over different time periods which is clearly a question of interest given the experience in the financial crisis and later in the COVID pandemic. We again use the CCK method and the new methods we have developed. We capture clear evidence of herding behaviour in different time periods, we have observed significant herding behaviour in most sectors among the different markets. We can observe there is more herding behaviour in different sectors than in the entire market. We also find there is more herding behaviour when the market is in turmoil or has larger movements which is consistent with prior literature. Then we compare the strength of herding behaviour between the Financial Sector and the Banking Sector. These sectors are of particular interest because of their interconnected nature and the fact they have been implicated in system risk particularly in the case of banks. There are, however, some important differences between the two sectors involving their business models, the extent and nature of regulation and perhaps the extent to which they are monitored by sophisticated investors. The results show that the Financial Sector has more herding behaviour than the Banking Sector under most market conditions. In chapter 9, we Investigate the impact of herding on market volatility. Drawing on previous research in the area we use GARCH models linked with measures of herding. Past work in this area directly uses dispersion as a measure of herding and we initially duplicate these studies. However, dispersion is unlikely to be a good herding measure as it is probably itself driven by volatility, regardless of whether herding is present. Hence, we adopt a new approach by measuring herding using the residual values from a model estimating the amount of dispersion expected if no herding is present which is a more valid approach. We find that using CSAD results as the proxy of herding, market volatility has a positive relationship with herding behaviour in the market while using residual values as the measure of herding, we have mixed results for the contemporaneous link between herding and volatility which is consistent with prior research. We do, however, find that market volatility is positively influenced by our lagged measure of herding. The final chapter presents the conclusions of our research.