Behavioral economics and behavioral finance are closely related fields that have evolved to be two distinct branches of economic and financial analysis. Behavioral finance is a very new area of research and very little developments have been made so far but it's becoming popular as researchers and policymakers are identifying the importance of scientific research on human and social, cognitive and emotional factors that better explain and help understand economic decision making by the consumers, investors, borrowers and how this decision making affect the market outcomes and allocation of resources.
Some of the main issues in behavioral finance today include "why investors and managers make systematic errors". It basically shows how market inefficiencies are created by these errors that affect prices and returns and how institutional and financial players take advantage of these market inefficiencies (arbitrage behavior). Some of the market inefficiencies are under reactions or overreactions to information, such misreactions have been as a result of limited investor attention, overconfidence or over optimism, noise trading, and herding behavior (mimicry).
The focus of my literature review is "Herd behavior in financial markets". In recent years, this topic has been of great interest, both theoretically and empirically, to see the extent by which financial market trading is affected by herd behavior; such an interest arises from the potential effects that herding can have on financial markets' ability and stability to achieve efficient, allocative and informational efficiency. The theoretical work (e.g., Avery and Zemsky, 1998; Lee, 1998; Cipriani and Guarino, 2001; Dasgupta and Prat, 2005; Sabourian and Park, 2006) have tried to identify the causes and circumstance that can lead traders to herd. My aim is to review the developments made so far in this area and how it has helped traders and policy makers to better understand the market trends and make informed decisions and not be part of the herd.
Herd behavior is commonly observed phenomenon. There are number of economic and social situations in which we easily get influenced by others decision making. The most common examples are from our daily life: we usually choose the same stores, restaurants and schools that are popular among others. Not only that the academic researchers also choose those topics that are currently in debate or are considered hot. The reason why such behavior is considered rational is because others may have information that we don't have. This would mean that everyone is doing what everyone else is doing and ignoring their own information which might be suggesting something else and this is called herd behavior. The very fact that people are ignoring their own information to follow others decision makes their decision less useful and informative for others. Let me quote a very simple example which will make it very clear as to why such behavior can lead to wrong decisions made by a group of people.
Most of us have been in a situation where we have to choose between two restaurants that are both more or less unknown to us now "Consider a population of 100 people who are all facing such a choice. There are two restaurants A and B that are next to each other, and it is known that the prior probabilities are 51 percent for restaurant A being the better and 49 percent for restaurant B being better. People arrive at the restaurants in sequence, observe the choices made by the people before them, and decide on one or the other of the restaurants. Apart from knowing the prior probabilities, each of these people also got a signal which says either that A is better or that B is better (of course the signal could be wrong). It is also assumed that each person's signal is of the same quality. Suppose that of the 100 people, 99 have received signals that B is better but the one person whose signal favors A gets to choose first. Clearly, the first person will go to A. The second person will now know that the first person had a signal that favored A, while effectively cancel out, and the rational choice is to go by the prior probabilities and go to A. The second person thus chooses A regardless of her signal. Her choice therefore provides no new information to the next person in line: the third person's situation is thus exactly the same as that of the second person, and she should make the same choice and so on. Everyone ends up at restaurant A even if, given the aggregate information, it is practically certain that B is better. To see what went wrong, notice that if instead the second person had been someone who always followed her own signal, the third person would have known that the second person's signal had favored B. The third person would then have chosen B, and so would have everybody else. The second person's decision to ignore her own information and join the herd therefore inflicts a negative externality on the rest of the population. If she had used her own information, her decision would have provided information to the rest of the population, which would have encouraged them to use their own information as well. As it is, they all join the herd". (Abhijit V. Banerjee, 1992)
The above example was given in order to give a better understanding of what is "Herd behavior" and how it can affect the decision making of a large group of people. My review paper will only focus on herd behavior in financial markets. There are many causes of rational herd behavior in financial markets and among them the most commonly observed are; concerns for reputation, compensation structures, information-based herding and cascades.
Before we discuss the empirical findings on these main causes of herd behavior it is important that I distinguish between "Spurious (unintentional) herding" and "True (intentional) herding". Intentional herding is usually characterized by fragility and eccentricity. It creates unpredictability and systemic risk. It is very important to determine the causes of investor herding because without that policies cannot be properly designed to mitigate such behavior. From the definition of herd behavior that we have understood herding results from an individuals/investors obvious intent to imitate or copy other individuals/investors decision where as in spurious herding a group of people are facing a similar problem with similar information and they make similar decision and outcome of such type of herding is efficient where as the result of intentional herding may or may not be efficient. It is easy to empirically distinguish them but in real setting there are number of factors that affect the decision of an individual/investor.
Information-Based Herding and Cascades
"Men nearly always follow the tracks made by others and proceed in their affairs by imitation."
Machiavelli (1514)
Banerjee (1992) and Bikhchandani, Hirshleifer and Welch (1992) introduced the concept of observational learning. The literature analyses the Bayesian sequential decision makers who take once-in-a-lifetime decision under asymmetric and incomplete information which as a result shows that even though the information was asymmetric eventually all decision makers will copy the behavior of their predecessor even if its not inline with their own information. The main assumption of this model was that all decision makers are able to observe previously made decisions in other words they have a complete history of actions taken. It is then the decision maker does a comparison of her own information with the set of others but in reality information is never perfect instead it's imperfect.
Bogaçhan Çelen (2003),Shachar Kariv (2003) build on Gale's (1996) model and they relaxed the perfect information assumption and instead they dealt with a situation where decision maker can only look at the decision made by her immediate predecessor. Smith and Sorensen (2000) made a distinction between action dynamics and learning dynamics. They differentiated between herd behavior and informational cascades which were again introduced by Banerjee (1992) and Bikhchandani, Hirshleifer and Welch (1992). According to them information cascades occur when, after a limited time all decision makers disregard their private information while taking an action, where as in herd behavior after limited time all decision makers take the same action which does not necessarily mean ignoring their own information. So an informational cascade involves herd behavior but it is not necessary that herding is a resultant of an informational cascade. In herd behavior traders or decision makers make the same choice but it could be different if they realized their private signals differently but in informational cascade traders find it best to copy their predecessor's action while disregarding their own signals and Smith and Sorensen (2000) argue that this belief is so strong that it outweighs all signals.
(Sushil Bikhchandani, David Hirshleifer, and Ivo Welch, 1998) argue that in reality cascades are very fragile. Many kinds of shocks can dislodge a cascade and there are many examples for it like an arrival of more informed individual, availability of new public information etc. when traders know that they are in a cascade, then they know that cascades are build on very little information as compared to information of private individuals. The imperfect-information model captures the phenomena such as fads, fashion, manias, crashes and booms and give a better understanding of concepts like why do markets move from booms to crashes without settling down? Why some technologies are rapidly adopted by a large group of people and then suddenly taken over by another?
There is another possibility not considered by (Avery and Zemsky (1998) which is to relax the assumption of costless private signal. In the model of sequential trading with fixed costs to information attainment, payoff maximizing agents find it most favorable to stop obtaining informative signals when the price of assets are closer to the extremes, and the benefit of extra information is overshadowed by its costs which results in non convergence of prices to the liquidation value of the asset. There is also a probability that prices get trapped at a level which is far from the actual value of the asset and the probability of this event will increase the cost of the signal.
Mixed success has been achieved by applying herding models to financial markets. The existing literature shows that informational cascades are possible in a sequential trading model. In the short-term there might be fluctuations but in the long-term prices always go back to the liquidation value. Informational cascades might recur because of any frictions such as transaction costs as in Lee (1993), opportunity cost of investment together with endogenous timing as in Chari and Kehoe (2004), or career considerations as in Dasgupta and Prat (2006), are added to the model.
Concerns for Reputation
Scharfstein and Stein (1990); Trueman (1994); Rajan (1994); Zweibel (1995); Brandenburger and Polak (1996); Prendergast and Stole (1996); Graham (1999); Ottaviani and Sorenson (2000) have all shed light on the theory of herding based on the reputational concerns of managers and its relationship with cascades. According to them it is very important to have a positive reputation in order to survive in the stock market and that is the reason why many managers copy the investment decisions of others while ignoring their own information which might be suggesting something else. From a social point of view this behavior will be considered inefficient but if seen from manager's perspective it is very much rational because they have to protect their reputation in the labor market.
According to classical economic theory investment decisions that are made by the agents are as a result of their logically and realistically formed expectation which means that they have used all the information available in the market in the most efficient way. On the other hand there is another view point which says that investments can also be made as a result of group psychology which will weaken the connection between available information and the market outcome. Keynes (1936,pp. 157-58) in The General Theory shows his disbelief on the long-term investors ability to make efficient investment and cash the market trends. He believes that managers will not be very willing to make an investment decision based on their own information because their contrarian behavior can ruin their reputation in the market.
Thus according to him managers will become part of the herd because they are concerned about their reputation to make efficient decisions. This kind of behavior can have many important repercussions. If for example we take a simple hypothetical scenario of a stock market where there is a common believe among managers that the price levels in the market are very high and they will most likely go down instead of going up. In such a situation there will be some managers just like always who would like to sell their equity but they fear that if the market continues its trend and their calculation didn't prove to be accurate then they will be considered fools for not benefiting from such an opportunity whereas on the other hand if market did go down than everyone will be suffering from the same fate so they will not look bad because everyone else is in the same situation.
There are two kinds of managers: the "smart managers" who get informative signals and the "dumb managers" who get noisy signals. Neither managers nor the market can determine between the two types unless they make an investment decision and ones they do then the market can adjust their beliefs according to these two pieces of information: did the manager make a profitable investment? And two whether the manager's behavior was inline with or was different to other managers. The first piece of information is not wholly used because it's natural that all the mangers get unlucky or they receive an uncorrected signal that's why the second information also plays an important role. Keeping the absolute profitability constant, managers will be preferred if they make a decision which is inline with others rather then behaving in a contrarian manner. So from reputation point of view being unprofitable is not that bad if everyone is in the same boat because this way the all share the blame.
This "sharing-the-blame" effect arises because it is believed that the smart managers receive correlated information whereas the dumb managers receive uncorrelated or noisy signals, so if they all act in a similar manner it shows to the market that they have all received correlated signals. On the other hand if the manager sticks to his own piece of information which is different from others that manager is more likely to be considered dumb.
Stock market volatility can be partly explained because of such behavior by money managers. By copying the behavior of each other (for example buying stocks when others are also buying and selling when others do the same) instead of responding to personal information, the herd members tend to increase exogenous shocks to stock prices. This can provide the foundation for stock market phenomena's such as mass euphoria, groupthink or panic.
Compensation based Herding
One of the most powerful reason for managers and investors in making a investment decision is compensation or profits. If the manager is solely concerned about his/her reputation then there is 99.9% chance that they will herd but investors who not only care about reputation but also about their gains and profits will have to analyze the tradeoff between their gains and loss of reputation. According to Scharfstein and Stein, (1990) as margin for profits increases, the incentives or parameter values which cause an investor to herd shrinks. Massimo Massa (2003), Rajdeep Patgiri (2003) have used mutual fund industry to study the tradeoff between reputation and compensation. As discussed above reputation is a very important factor for managers/investors but Masimo Massa and Rajdeep Patgiri (2003) study shows that the more compensation structure is incentive-loaded which puts extra weight on performance may tradeoff for the negative impact of loss of reputation. So one can deduce from these findings that there will be more of risk taking and less of herding when the compensation structure is incentive-loaded.
When the incentives get higher then there is a greater chance of higher performance on the manager's part, because his performance is very closely related to his payoff. If higher incentives lead to increased effort from the manager then this effort should translate into greater net-of-risk performance. But if higher incentive means higher risk then there is very little possibility to observe superior performance once the risk factor is controlled. Infact they argue that increased risk taking can lead to survival problems of the manager/Fund and there is also little possibility of performance to last overtime. Masimo and Rajdeep (2003) argue that increase in performance cannot solely be explained by higher risk-taking and it is related to increased effort of the manager that leads to persistent performance overtime.
A multivariate regression analysis proves that as higher level or increase in incentives results in persisting performance by the funds year after year. According to their analysis type of compensation also played a very important role as it has a direct relation with herding and risk taking. Which means it is very important for both the market and investor, but most of the studies revolve around the high cost of giving incentives and little attention is given to the benefits which can potentially translate into lower levels of herding.
Bikhchandani and Sharma, (2001) argued that herding in the financial markets can aggravate volatility, make it more vulnerable to shocks and can destabilize the market. So if looked at it from their prospective then herding caused by compensation or reputation can increase market fluctuations and higher compensation stabilizes the market by reducing the rate of herding. Wermers, 1999, Chen et al., (2001) argue, if herding is based on information then increasing the compensation level may not have any affect.
Jensen and Meckling (1976) looked at the effect of higher compensation on the behavior of the manager. They argue that profit-maximizing managers will prefer higher asset volatility even if the firm itself is on the stake. Grinblatt and Titman (1989) looked at the fund manager who can protect his incentive fee will try to increase the fee value by increasing fund leverage. Ross (2004) study shows that no incentive scheme has the ability which can cause all utility maximizers to take on extra risk and this relationship is also studied by Leland and Senbet (2002), Masimo Massa and Rajdeep (2003).
Herding and Contrarian behavior in financial markets
Mathias Drehmann, Jörg Oechssler, Andreas Roider (2005) designed an internet experiment to address some important issues which included testing of informational cascade theory introduced by Avery and Zemsky (1998). They used an exact design of their model so that they are able to potentially reject it. The main aim of the theory was to see whether herding can be prevented by market prices. They also wanted to test whether traders follow their private signals or they engage in herd behavior, or follow contrarian strategies. Over 6,400 people participated in the experiment and they used this diversity and size of their pool to see the difference in their behavior because of their different backgrounds.
Since they used AZ's model the market maker assumed while setting prices that all the participants behave rationally. But the experiment result showed that large part of the pool ignored its private signals. Due to this discovery Mathias Drehmann, Jörg Oechssler, Andreas Roider (2005) added two new "error treatments" while setting prices, they kept in mind that subjects do not always follow their private signals. The first treatment dealt with noise traders, where the market maker assumed that there is a proportion of noise trader making uninformed decision along with rational traders. The second error treatment was based on Richard D. McKelvey and Thomas R. Palfrey (1995, 1998) quantal response equilibrium where market marker considers that subjects ability to follow his/her own signal depends on the history of his past decisions.
They also used two benchmarks; first treatment was without market prices related to BHW model of AZ; and in second benchmark subjects could observe both their own as well as predecessors signals. Their experiment supports large set of empirical literature which include Lakonishok et al. (1992), who looked at security analysts (Welch,2000), Bikhchandani and Sunil Sharma, 2000; Kent Daniel et al., 2002; or David Hirsh- leifer and Siew Hong Teoh, 2003), Lisa R. Anderson and Charles A. Holt (1997), and Cipriani and Guarino, 2005 who were the first ones to conduct an expriement on cascades with flexible prices and Mathias Drehmann, Jörg Oechssler, Andreas Roider (2005) used some of the same methods or treatments as they did.
Mathias Drehmann, Jörg Oechssler, Andreas Roider (2005) study shows that AZ model made the right predictions as no evidence of herding was found when prices were flexible but their few predictions are not supported by actual data and one such prediction was that all subjects will follow their personal information, but according to data only 50 to 70% do that which means prices are not on the level where they should be when 100% use their private information. As mentioned earlier herding was not observed but there was presence of contrarian behavior. Subjects preferred buying asset B when the price of A was higher and vice versa. This result shows that contrarian behavior is fruitful at extreme prices. When subjects have doubts on rationality of others they might loose faith in their own decision. The error models they introduced do consider this aspect and provides an explanation for contrarian behavior.
Since the pool of subject was large an diverse Mathias Drehmann, Jörg Oechssler, Andreas Roider (2005) conducted a series of behavioral comparison with respect to educational background and demographics etc. the results showed that gender and college education (some or none) had no significant difference but subjects who had Ph. D degree were more accurate with theory predictions. Subjects belonging to different field of studies performed differently in term of rationality where the best performance was by the physicists and worst by psychologists. But performing well in rationality does not guarantee profits because profitability only comes when others are also being rational. In terms of profits the case was completely opposite. This internet experiment was quiet successful as it confirmed a large set of theoretical literature on herd behavior.
Herding Measures
In order to identify and measure the extent of herding in the stock markets several measures were developed by many researchers and among them the most popular are
LSV (Lakonishok, Shleifer, and Vishny, 1992),
PCM (portfolio change measure by Wermers, 1995),
CH Measure (Christie and Huang, 1995),
Chang, Cheng and Khorana (2000).
Soosung Hwang, Mark Salmon, 2001.
The LSV criteria of measuring herd behavior is based on the trades carried out by a section of markets participants over a period of time but this measure of herding does not critically look at the amount of stocks the investors buy or sell. Wermers (1995) on the other hand designed a portfolio change measure which takes into account both the critical factors like the direction and intensity of trades. Another method proposed by Christie and Huang (1995) examine the magnitude of cross-sectional volatility of individual stock returns during periods of large price movements. If dispersion over that time period is little then that means there is presence of herding. But focusing on variance of returns bounds the value of the CH measure greatly. The reason for that is cross-sectional volatility on returns is dependant on the time series volatility of returns and it might not be used directly to examine herding behavior.
Soosung and Mark (2001) developed a new measure of herding and used linear factor models. This was very similar to CH measure but it also examined the information contained in the movements of the cross-sectional markets. They focused on the variability of factor sensitivities of cross-sectional markets rather than the returns by CH. Soosung and Mark (2001) also developed a statistical testing measure unlike CH measure. This method made it easy to calculate herding which was a bit difficult in LSV and PCM measures where very detailed records were required for trading activities and changes in portfolios which are mostly unavailable and instead of measuring herd behavior of a group Soosung and Mark (2001) examined the market-wide herding.
Recently another measure of herding has been suggested by Chang, Cheng and Khorana (2000) which is a variation of CH method. According to them "under CAPM assumptions, rational asset pricing models suggest that the equity return dispersion, measured by the cross-sectional absolute deviation of returns, should be a linear function of market returns" (Soosung and Mark, 2001, pg 2). They used a data set of South Korea, Hong Kong, Taiwan and US markets and found that there is presence of herding behavior in South Korea and Taiwan. The assumption that is made in the Soosung and Marks (2001) model is that herding is something that can only be looked at in relative terms but it is seen that most of the other methods of herd measures e.g LSV, PCM, CH and Chang, Cheng and Khorana (2000) have tried to look at herding in an absolute terms.
Herd Behavior in Laboratory Financial Markets
A number of empirical papers have recognized the presence of herding in the financial markets (see, e.g., Lakonishok et al., 1992; Grinblatt et al., 1995; Wermers, 1999; Sias, 2004) but none of them have directly tested the theoretical models of herding; but cipriani and Guarino (2006) have made an exception by estimating a structural model of informational herding. They have compared two different treatments; in the first one the price adjusts to the order flow in a way that herding should not occur, and in the second one the existence of event uncertainty makes herding possible.
Testing herding models with actual financial market data is very difficult because in every model herding means taking the same decision as others but with the private information that one receives. The main problem for the researchers is that data on private information is not available which makes it very hard to understand the trader's rationale behind making similar decision. To mitigate this problem many authors and among them Cipriani and Guarino (2005); Drehman et al (2005) tested presence of herd behavior in a laboratory financial market. A big advantage of testing models in such a setting is that one is able to look at variables that are not available in the real markets.
Many studies in the past have looked at the way experimental subjects who are in most cases undergraduate students play strategic games in similar manner professionals might do so. But Cipriani and Guarino (2006) took a different approach they used a sample of financial market professionals and looked at their activity in a controlled financial market environment. They believed that by using professionals from real financial markets will highlight their true activity. They also believed that this will help to connect empirical analyses with theoretical studies. This was the first study that used financial market professionals, Alevy et al (2007) also used financial professionals but they tested a standard cascade game and not trading. Drehmen (2005) study on herding behavior used both students and professionals sample but they didn't use financial market professional and had similar limitations to others.
The result of their study were inline with the theoretical predictions made by Avery and Zemsky (1998) as the proportion of herding was very low in the first treatment which was evident from the experimental data. In the second treatment the rate of herding increases which was again in accordance with the theory. But there were few discoveries made by Cipriani and Guarino (2006) which were not accounted for in the original theory. In the first treatment some of the professionals engaged in contrarianism and sold when the price was high and bought when the price was low (irrespective of their private signals). In the second treatment herd behavior is not as much as predicted by the theory and in both the treatment one thing which is not accounted for in the theory at all was that professionals had the tendency to avoid trading which would mean that market is unable to comprehend the private signals of professionals and this in turn decreases the informational efficiency of the market.
Herding in Emerging Stock Markets
1997 financial crises in the emerging markets attracted much attention and many researchers focused on the important issue of whether international investors herding behavior leads to higher volatility in the developing countries capital flows. Most of the research done is based on Korean financial markets and the main reason for that was the availability of data which could directly uncover the trading practices of the investors.
Kim and Wei (1999) collected data from Dec 1996 to June 1998 and analyzed the trading practices of the Korean investors. This data had information on investor's origin (Foreign or Korean), whether they were an institution or individual, resident or nonresident, ceiling on ownership etc. they used the Lakonishok, Shleifer, and Vishny (1992) LSV measure of herding and shed light on some important findings. They identified that nonresident institutional investors were using positive feedback trading strategies before the crisis and after the crisis they started using momentum strategies on the other hand resident institutional traders before the crisis were contrarian traders and during the crisis period they also started using positive feedback trading. They also discovered that nonresident traders were herding more then the resident traders also herding was higher for individual investors as compared to institutional investors. According to them herding did increase during the crisis period but it wasn't that significant statistically.
Choe, Kho,and Stulz (CKS), 1999, came to a similar conclusions by using KSE data of daily transactions. But there was a difference in their findings which was that Kim and Wei (1999) concluded from their research that herding increased during the crisis period but Choe, Kho and Stulz (1999) had an opposite conclusion that herding was actually lower. My understanding is that the difference in the two finding was as a result of different data set or sample periods. CKS classified their investor into three broad categories which included foreign investors, domestic institutional investors and domestic individual investors. They analyzed the behavior of foreign investors before the crisis and when the crisis was at its peak. They also used LSV measure to test for presence of herding and figured that herding was quiet significant in Korean stocks and these investors used positive feedback strategies before the crisis. They made an estimate of the herding ranges of theses foreign investors and saw a decline in the herding figures during the crisis and also observed that investors were less inclined towards using momentum strategies. CKS came to the conclusion based on their findings that foreign investors did not destabilize the Korean stock market during their observation period and in general they don't seem to have a destabilizing influence.
Another study was conducted by Borensztein and Gelos (2000) but that did not focus on any one country instead they took a sample from the Emerging Markets Funds Research, Inc. of 467 funds in developing countries on monthly geographic asset allocations between the period of 1996 to 1999. These funds were quiet significant in the markets they operated in and had a market capitalization of 4 % to 7%. Borensztein and Gelos used the LSV herding measure for these funds and found that the herding figures they obtained were in the lower range as identified by Kim and Wei (1999) for nonresident institutional investors in Korea and there wasn't much difference in the herding figures before or during the crisis period. They also found that funds that are offshore tend to herd more than others and that is also true for funds operating in larger markets.
Conclusion
Herding is a very important aspect of behavioral finance and especially herding in financial markets. Much research has been conducted to see the impact of herding on financial markets but much research work is done in the developed countries and the result suggests that investor in these countries do not show significant herd behavior and their tendency to herd is related to their ability to use momentum investment strategies. Efficiency of positive feedback and momentum strategies depend on how quickly market absorbs new information in the prices. Very little work has been done on Emerging Markets and they need to be focused more because that's where the herding behavior is more pronounced.
In the emerging markets the functioning is not very clear because of weak infrastructure which includes weak reporting and accounting standards, negligent behavior of regulatory body and higher cost of acquiring information. All these things lead to higher rate of information cascades and reputational herding. Another important issue is that information is very slowly incorporated by the market in the prices; and that is the reason it is more profitable to engage in momentum investment strategies.
The methods used in the research studies need to be selected with caution and need to be more refined so that its easier to differentiate between actual herding behavior from mere reaction of investors to new information or public announcements. It is hard to estimate the degree of herding behavior in the markets even if there is existence of highly technical statistical tools because there is certain required information which is not available. But ambiguity is very important for the proper functioning and liquidity of the markets and it is not apt to ask the financial participants to reveal information regarding their investment strategies.
There will always be some level of information irregularity and that is the reason there is possibility of formation of informational cascades and compensation and reputational based herding. It is very important to make the market transparent by designing better compensation contracts, timely availability of data, disclosure rules. If the markets are transparent then it is possible that most of the profit maximizing investors will respond to the changes in the situation of economic units in the similar manner.
