Patterns of return predictabiliby across economically linked firms

(1)

Patterns of Return Predictability Across Economically Linked Firms

Stella Hak 1st of July 2018

MSc Finance: Asset Management

Supervisor: Stefan Arping

(2)

this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

This study documents return predictability across economically linked firms: customers and suppliers. Testing the possible presence of attention constraints, by sorting suppliers on the basis of their customer returns in the previous month, resulted in significant results. I examined a more recent time period, as compared to previous research testing return predictability across customers and suppliers, in order to investigate whether recent technological changes have had an effect on information processing by investors. I find that return predictability across economically linked firms still appears to exist in capital markets, making them less efficient than theory would indicate. A trading strategy based on the observed customer-supplier links results in significant abnormal returns. The results are robust when controlling for size, analyst coverage, industry effects, and trading volume.

(4)

TABLE OF CONTENTS

CHAPTER

I RESEARCH OBJECTIVE

Introduction 1

Summary Approach & Outline 2

II RELATED LITERATURE

Efficient Market Hypothesis 4

Related Research on Investor Inattention 5

Summary Cohen and Frazzini 7

Recent Technological Changes & Effects 9

Speed of Information Diffusion 11

III METHODOLOGY

Portfolio Trading Strategy 13

Speed of Information Diffusion 16

IV DATA AND DESCRIPTIVE STATISTICS

Data Gathering Process 17

Summary Statistics 18

V RESULTS

Market Model 22

Carhart Four Factor 25

Duration of the Inattention Effect 28

(5)

VI ROBUSTNESS CHECKS

Size 33

Analyst Coverage 36

Industry Effects 37

Trading Volume 40

VII LIMITATIONS & SUGGESTED RESEACH 43

VIII CONCLUSIONS 45

REFERENCES 46

APPENDICES

A FIGURES 50

(6)

Introduction

The theory of the efficient markets is based on the assumption that all available information concerning a financial asset is incorporated into its price (Fama, 1970). There is a continuous ongoing debate whether capital markets indeed are efficient. One implication of the failure of the efficient market hypothesis is the perceived predictability of stock returns as well as return autocorrelations, return reversals, and momentum, which is observed in stock markets (Vozlyublennaia, 2014). As stated by Hou and Moskowitz (2005) there is substantial evidence indicating that market frictions are present and that there is a significant amount of investors who fail to diversify. Numerous researches have tried to explain the anomalies that have emerged in asset returns, although opinions concerning the true cause tend to diverge, most research is conducted by investigating some form of reaction to new information by investors. Since the beginning of this century theories of behavioral finance have been receiving more attention and more empirical evidence has been carried forward. The idea that a stock return could be predicted by nothing other than its market beta, the most common measure of risk, has been rejected. According to Vozlyublennaia (2014) investors’ (in)attention must play an important role in the formation of asset prices, stock returns, and the overall efficiency in security markets.

Underreactions of the markets caused by inattentive investors are a widely known phenomenon in behavioral finance. Investor inattention stems from the inability of individuals to process multiple stimuli and execute different tasks at the same time (Hirshleifer, Lim, & Teoh, 2009). Of course, the human minds and their abilities are finite, therefore the assumptions that people have unlimited capacity to process new information is unreasonable. By introducing either subtle or salient links one could test whether investors are incapable of processing these links.

Firms are often linked to each other through customer-supplier relationships. When investors tend to miss these economic links between firms they will miss the shocks that occur to the supplier (customer) when news about the customer (supplier) is released. Therefore, the economic link characterizing the customer-supplier relationship could be a measure of investor inattention. Overlooking the publicly available link will translate into the predictability of stock returns of one firm, as a reaction to an earnings announcement, or other new information that causes price changes, of the other firm. This thesis will therefore investigate whether there is a pattern of return predictability of stock returns by examining the development of stock returns of customer and suppliers. The research will be conducted by

(7)

forming a portfolio consisting of long positions in suppliers stocks of which the customers have experienced high returns in the previous month. Additionally, the portfolio will consist of short positions in supplier stocks of which the customers have experienced low returns in the previous month. The analysis is in a way similar to the one conducted in Cohen and Frazzini in 2008, although it differs in the time period used. I will use a more frequent time frame to investigate whether there has been a change in investors’ inattention in recent years. The increased usage of mobile phones nowadays allows human beings, and investors in particular, to access almost all information necessary, instantly. The increased connectivity would thus cause us to believe investors have become less constrained by the amount of information available. On the other hand, research has shown that the increased usage of mobile phones causes a reduction in the cognitive processing abilities (Ward, Duke, Gneezy, & Bos, 2017). One could therefore either, by the increase in the availability of information, expect a decrease in the underreaction between customer and supplier stocks. Alternatively, more information available could make it harder to process information, a phenomenon called information overload (Lee, 2012). In the latter case, return predictability would still be present in capital markets. Another deviation from Cohen and Frazzini is found by looking at the market model used: the authors use the Fama and French three-factor model, whereas I will be using the Carhart four-factor model, which extends the three-factor model by adding the momentum factor as well. This factor is added because the observed effect could be due to the supplier’s own past returns.

Most of the classic asset-pricing models in finance assume instantaneous information diffusion, even though previous empirical research has shown that frictions do occur to investors, translating into a slower diffusion of information into the marketplace (Hou, 2007). When return predictability is found in the first part of this thesis, pointing out a lead-lag effect, the question then arises how long it takes for the information to be incorporated into the stock prices. This thesis will therefore investigate the speed of information diffusion between long-standing customer-supplier relationships.

The thesis outline is as follows: the literature review will discuss the efficient market hypothesis and its implications. Followed by an examination of previous works concerning the topic of investors’ inattention, a summary of the article of Cohen and Frazzini will be given. In order to justify the reason for this study recent developments in technology, and their effects on the human brain will be discussed. Finally, the literature review will give some background information about the speed of information diffusion and previous studies. The methodology will discuss the models that will be used for the analysis, together with their

(8)

implications. Thereafter the main results of this thesis will be discussed followed by the interpretation of the results and some conclusions. In order to justify the conclusions drawn from the results a few robustness checks will be conducted. The importance of validity will be discussed, the shortcomings of this thesis, and suggestions for further research. Finally, the conclusion of the study will be given.

(9)

Related Literature

The most important question in the literature about investor inattention is whether limited attention affects stock returns and their predictability. This section will give some definitions of the efficient market hypothesis, its implications, and the effects of inattentive investors on the efficient market hypothesis. It will further discuss previous articles about investor inattention, and the known effects on stock returns and their predictability. Thereafter I will discuss the article of Cohen and Frazzini in more detail, which is followed by consideration of the recent changes in technological developments in the past decades and the impact on cognitive processing. Finally more attention will be given to the speed of information diffusion.

Efficient Market Hypothesis

The theory of the efficient market hypothesis (EMH, from now on) is based on the assumption that all available information concerning a financial asset is incorporated into its price (Fama, 1970). According to Fama (1970), this implies that even information that is not available to the public will be incorporated into stock prices. Using this assumption, together with the assumption that all economic agents are rational, this would imply the impossibility of assets to be mispriced (Sybramanian, 2010). In the case a financial asset is mispriced, trades of arbitrageurs would quickly correct it. Grossman and Stiglitz (1980) state that investors can earn an excess return by using superior information by taking in positions that are better than the positions of traders who do not possess this information. When information is costly, prices will not reflect all available information and competitive markets break down (Grossman and Stiglitz, 1980). The EMH is known to be described by many different forms. When the “strong’’ form of the EMH is in place it implies that the market will be efficient with respect to all the information available (Basel & Stein, 1979). In the case the strong form holds securities will not be mispriced and no profitable trading strategies arise. This would in the light of investor inattention implicate that investors are aware of the customer-supplier links and new information would be followed by an immediate price response (Zhu, 2014). Although the efficient market hypothesis states that capital markets will incorporate all available information, previous research of Hirshleifer and Teoh (2003) has shown investors are constrained by their own cognitive barriers of the mind. The observed predictability and cross-predictability of returns in previous research conflicts with the notion of efficient markets (Hou & Moskowitz, 2005). Another form of the EMH is known as the “semi-strong’’

(10)

form, which is more in line with the existence of costly information. The semi-strong form states that prices will reflect all publicly available information (Fama, 1970). When the assumption of fully efficient markets is loosened the existence of the predictability of stock returns can be explained by the gradual diffusion of information along, for example, customer-supplier links (Menzly & Ozbas, 2010).

Another assumption of the EMH is that all investors are rational and markets are frictionless (Hou & Moskowitz, 2005). In the light of rationality multiple approaches have been conducted, including representative-agent models with rational beliefs but unconventional preferences, such as those associated with prospect theory (Barberis & Huang, 2001). Prospect theory states that investors do not behave irrational necessarily, but they derive utility from fluctuations in their wealth instead of consumption alone. Barbaris, Schleifer, and Vishny (1998) propose a representative-agent model with standard preferences but biased beliefs. The theory of rational agents does not correspond with the existence of underreaction in the short-run and an overreaction in the long-run (Hong & Stein, 2007). Therefore, Hong and Stein (1999) propose a model with investors who are boundedly rational, allowing for the possibility of slow information diffusion. Hirshleifer, Lim, and Teoh (2006) suggest that even if there are rational investors in the market, the issue that arises is that they cannot fully eliminate the patterns by trading upon them since their capacity is finite. When it comes to limited attention, Hirshleifer, Lim, and Teoh state that all investors are constrained by their cognitive abilities. The reasoning behind this statement is that even if there is a rational investor present in the market, he or she will allocate his/her full attention to one security, this would mean their resources are drawn away from other activities and stocks. The strong form of the EMH appears to have too many shortcomings and there is too much available evidence to be able to assume that the strong form is present. It is therefore more compelling to assume a semi-strong form, in which patterns of return predictability can arise.

Predictability of Stock Returns & Limited Attention

Multiple researches in finance have attempted to answer the question whether stock returns are predictable, and if so, which factors contribute to the prediction of these returns. While a large variety of factors exist that try to explain the predictability, this section will solely focus on the literature that is associated with the processing on available information of related firms who are publicly traded and the notion of investor inattention.

Multiple reasons for return predictability based on publicly available information were put forward. Since traditional asset-pricing models appear to lack the ability to explain

(11)

differences in future returns, behavioral models have been introduced to allow for the departure of the classical EMH assumptions (Hong & Stein, 1999).

Hayek (1945, as cited in Menzly & Ozbas, 2010) proposes the possibility of dispersion of information about fundamentals. This is supported by the disagreement model of Hong and Stein (2007), who argue that investors have heterogeneous beliefs about asset prices and the fundamentals of a stock. Previous work of Hong and Stein (1999) discusses how information is processed among different types of investors, and in particular how some can only process a fragment of the available information. Differences in information distribution and/or technology cause relevant information to arrive earlier to some investors than others. Therefore some investors will react sooner to the arrival of information, which is referred to as gradual information flow by Hong and Stein (2007). Hong and Stein (2007) mention that the process of gradual information diffusion is not necessarily inconsistent with the assumption of rational investors. It solely demonstrates how patterns of above-average trading volume and prices arise, it does not advocate in favor to some form of inattention or bounded rationality.

Less focused on the dynamics of information diffusion is the idea of limited attention, which refers to limitations in the cognitive processing abilities of investors (Hong & Stein, 2007). Together with disregarding the assumption of sophisticated investors (EMH), investor inattention leads to valuations by investors based on only a part of the relevant information available.

Vozlyublennaia (2014) investigated the links between investor inattention, as measured by Google search probability, and the performance of security indices. The findings indicate that when an increase in attention is observed, index returns change significantly, for a short period. Contrariwise, a long-term effect is detected in investor attention when a shock to the index return has taken place, especially when it concerned a negative shock. (Vozlyublennaia, 2014).

Another study relating investor inattention to stock returns is one by Hirshleifer, Lim, and Teoh (2009), testing the investor distraction hypothesis. The hypothesis states that there is a slow reaction in market prices and trading when irrelevant information is revealed about a firm. Earlier studies concerning the investor distraction hypothesis presented results indicating the negligence of investors to information signals. The negligence of these information signals then leads to mispricing of securities whose information is publicly available (Hirshleifer & Teoh, 2003 as stated in: Hirshleifer, Lim, & Teoh, 2009). The authors suggest that one’s interpretation of, for example, the future profitability of a firm is hindered

(12)

by extraneous news about other firms, which creates a distraction for investors. When there are more distractions, or the magnitude of the distraction is large, there will be a greater underreaction to a firm’s earnings announcement, leading to a greater post-earnings announcement drift.

Distraction can sometimes come from quite obvious explanations as well, DellaVigna and Pollet (2009) investigated limited attention by hypothesizing that on Friday, when investors are distracted by nothing other than the weekend, earnings announcement have less effect on the abnormal returns of a stock. Friday earnings announcements experience both lower trading volume and more post-earnings announcement drift.

Zhu (2014) explored the possibility of abnormal returns of supplier stocks around a customer’s announcement date. Also based on the incapability of investors to acknowledge the existing links between customers and suppliers, Zhu’s hypothesis is based on the responsiveness of one firm’s stock returns to another, linked, firm’s release of new information. Zhu (2014) finds that the abnormal returns of a supplier are positively related to the earnings announcements of their customers.

Menzly and Ozbas (2010) investigated the possibility of cross-predicting returns within customer-supplier relationships. They propose a limited-information model to examine the existence of return predictability across the supply chain. The findings indicate that returns are correlated along the supply chain and self-financed strategies result in significant returns. Menzly and Ozbas (2010) also find that the degree to which stocks are predictable within economically linked firms, depends on the level of information available in the market. The level of information available in the market is measured by the level of analyst coverage. Menzly and Ozbas exclude small stocks from their sample in order to ensure the lead-lag effect does not arise from a delayed price response among small stocks.

Finally, the lead-lag effect between customer and suppliers seems to hold up internationally as well. Shahrur, Becker, and Rosenfield (2010) investigated return predictability based on the lead-lag effect in developed countries, excluding the United States.

Cohen & Frazzini

The following section will discuss the main article on which this thesis is based. Cohen and Frazzini (2008) investigated the customer-supplier links, and in particular the presence of attention constraints which leads to the existence of abnormal returns. Their research showed that buying supplier firms whose customers had positive returns in the previous month, and selling short supplier firms, whose customer returns were negative in the previous month,

(13)

yielded abnormal returns of 18.6% per year. The intuition behind their research is the existence of a limited ability of investors to process all available information in the market. A considerable shock in the earnings of a customer should, subsequently, have a shock to the suppliers’ earnings. Although the translation of an earnings announcement of a customer should have an immediate effect on the supplier, when the information is publicly available, investor tend to miss these links. This leads one to consider the idea of return predictability when studying economic links. Cohen and Frazzini (2008) impose two conditions to test for limited attention. The first states that the information, that is possibly overlooked, by investors needs to be readily available to the public. The second condition requires that the information is salient, meaning that it is noticeable and therefore that it can be expected of investors to gather this information.

In order to test for the existence of return predictability, due to investor inattention, Cohen and Frazzini composed a portfolio containing either long or short positions in the stock of the supplier firm. By sorting suppliers based on the returns of the customers in the previous month, a portfolio was created containing long positions in suppliers whose customer(s) had performed well the previous month, whereas short positions were held in suppliers whose customer(s) had performed bad/worse in the previous month. If the hypothesis of the existence of investor inattention is confirmed one would expect that returns of either one of the firms could forecast the future returns of the other firm. The trading strategy translated into monthly abnormal returns of 1.45% when compared to the Fama and French three-factor model. The authors refer to their findings as customer momentum, which entails the predictability of stock returns on the basis of customer-supplier links. Before moving on to recent changes in the technology and the effects on the cognitive processing abilities of investors, it might be interesting to finish with an example of the failure of investors to process information about economically linked firms. The example was given by Cohen and Frazzini and it concerns the publicly traded firm Coastcast, a manufacturer of golf club heads, and one of its most important customers: Callaway Golf Corporation, publicly traded as well. On the 8th of June 2001 Callaway’s stock dropped by 30% in response to the downgrading of an analyst covering it the previous day. In a press release on the 8th Callaway lowered their second-quarter revenue prediction by half, the stock price thus dropped accordingly. The forecasted earnings per share (EPS) changed from 70 cents per share to around 35 cents per share. So far, the resulting price drop and change in earnings per share of Callaway lies all within the expectations. However, the negative news of Callaway did not have any impact on Coastcast’s share price, nor on its EPS forecasts. Finally, after nearly two months Coastcast

(14)

declared EPS of -4 cents, which was followed by negative returns in the consecutive two months. The course of both firms’ returns is depicted in the figure below.

Figure 1

Recent Changes in Technology & the Effects on Cognitive Processing

I will, as well as Cohen and Frazzini, investigate publicly traded firms who are, by law, obliged to disclose their main customers. Although different in more than one perspective, the main difference is the time period over which this study will span. Since more recent data will be used to investigate whether investors’ inattention to the customer-supplier links has changed in recent years, two possible outcomes exist:

I. Due to the publication of the article of Cohen and Frazzini, investors have become more aware of the existence of these economic links, and have incorporated this into their trading strategies. Additionally, due to the increased availability of data and information, investors have become more attentive.

II. Contrarily, due to the increased availability of information an information overload effect could arise. Previous studies have shown that an increase in information has led to worse decision making amongst individuals, since it requires more effort to process more information (Casey, 1980, and Simnett, 1996, as cited in Lee, 2012). When more information is available it takes more time and resources to extract the information efficiently.

(15)

Due to the tremendous amount of information that is readily available the processing powers of the human mind are constrained, and therefore only able to focus on understanding the implications of, for example, one financial report of one firm at a time (Hirshleifer & Teoh, 2003). The more recent period I will be investigating is characterized by a fast growing technology sector. Especially when it comes to smartphones, there has been a significant increase in the usage. Not only is an increasingly large number of the world a smartphone user nowadays, the frequency by which smartphones are used is increasing as well. The paper of Wilmer, Sherman, and Chein (2017) discusses the effect of smartphone usage on cognitive functioning. The authors point out that there is too little empirical support to state the use of smartphones has a significant effect on attentional capacities. However, this concerns long-term effect, supposedly there are short-long-term consequences. Engagement with smartphones does lead to attention distraction in the short-run. When it comes to learning, data has shown human beings learn less and remember less from smartphones (Wilmer, Sherman, & Chein, 2017). Ward, Duke, Gneezy, and Bos (2017) test the “brain drain” hypothesis, by investigating that the usage of smartphones affects the cognitive performance. The results illustrate that even the mere presence of a smartphone reduces and negatively affects cognitive functioning. The continuous encompassing of (new) information has increased, yet the authors state that the ability to process all this information is constrained. The great amount of available information and the limited competence to process this information results in a constant mismatch. The final thing worth mentioning concerns the priority of the stimulus, which is dependent on its salience as well as its relevance. In the light of investor inattention, this would imply that investors are constrained in the processing of the economic links, because of cognitive limitations as well as a lack of salience, and perhaps the inability to acknowledge the relevance.

A final thing worth mentioning is the behavior of anomalies in finance, especially after they have been addressed by researchers and after they have been published. McLean and Pontiff (2016) tested whether patterns of return predictability were still present after publication. The authors investigated 97 characteristics that have been presented as possible factors to explain stock returns, and, if these factors still contribute to the explanation of stock returns post-publication. They find a significant deterioration in the predictability of stock returns by the characteristics, portfolio returns were 58% lower post-publication.

(16)

Speed of Information Diffusion

Earlier evidence has shown there is a gradual diffusion of information between economically linked firms (Hirshleifer, Lim, & Teoh, 2009). In the light of return predictability, the speed of information diffusion has been one of the primary subjects. Most of the literature concerning the diffusion of information has been focused on the momentum factor, which refers to the tendency of stocks to have a delayed price response to their own past returns (Menzly & Ozbas, 2010). Menzly and Ozbas however, investigate the speed of information diffusion within customer-supplier relationships. The authors find evidence of cross-predictability across customer and supplier industries.

Lee (2012) examined how information diffuses into prices by studying the degree of availability of quarterly earnings announcements. The author discusses the notion of stock prices drifting after an earnings announcement, known as the post-earnings announcement drift. Multiple explanations have been suggested to analyze and justify this price delay. Lee (2012) proposes an idea concerning the readability of earnings reports. To a greater extent promoting the idea of limited cognitive ability of the human mind, Lee states that when reports are more difficult to read and interpret, the underreaction to earnings announcements will be more severe.

Hou and Moskowitz (2005) studied the austerity of market frictions affecting the incorporation of new information into the price of a stock by investigating the average amount of time it takes for a stock to incorporate this new information. The authors demonstrate that frictions associated with investor recognition seem to be related to the slow information diffusion. It is worth mentioning the possibility of an observed price delay, which stems from lack of liquidity of a particular stock (Hou & Moskowitz, 2005). When investigating the main research question of this study, some conditions concerning the liquidity of a stock should be introduced, in order to ensure that any price delay detected does not stem from a shortfall in liquidity of the stocks. Hou and Moskowitz (2005) further find that cross-predictability is more apparent across small firms, firms who are less visible, and firms who receive less attention from market participants. According to Hou (2007), any observed lead-lag effect in stock markets is driven by big firms leading small firms within the same industry. Hou (2007) suggests that this observed can arise due to multiple reasons; asymmetric information, noise traders, limited investor attention, short sale constraints, and other sorts of frictions or constraints.

Concluding, the strong form of the EMH is inconsistent with the patterns of observed return predictability across stocks. Concerning investor inattention, multiple areas of study

(17)

have already been conducted, all finding results of price delays across economically linked firms. Investor inattention can come from a dispersion in beliefs about fundamentals, different sorts of information processing, cognitive constraints, attention distractions and so on. The vast amount of studies is conducted with data that is considerably less recent than this study. Recent changes in the increased amount of mobile phones have serious consequences for the processing abilities of individuals. Overall research suggests that the use of mobile phones decreases cognitive performance. Therefore the expectations are that return predictability is still observed in markets. It is important to keep in mind that observed price delays and therefore, cross-predictability, may also arise due to other market shortcomings, these will be discussed when the robustness checks are performed.

(18)

Methodology

The methodology will consist of two parts. The first part will discuss the return predictability, and how this can be measured by constructing a long/short portfolio. After that, the speed of information diffusion will be discussed.

The first part will consist of building a trading strategy based on the lead-lag effect. In this way the possibility of abnormal returns can be inspected. In order to investigate the return predictability, information is needed about long-standing customer and supplier relationships, and in particular customers who account for at least 10% of a supplier’s sales. Following the methodology of Cohen and Frazzini (2008) and Shahrur, Becker, and Rosenfield (2010) the stocks of the suppliers will first be grouped based on their customer returns in the previous month, customer returnt-1. After this, the sample will be divided into five quintiles, on the

basis of the customer returns in the previous month. The first quintile, Q1, will represent suppliers whose customers have experienced the lowest returns in the previous month and the fifth quintile, Q5, represented the suppliers whose customers have experienced the highest returns in the previous month. When the stocks are sorted, a long/short strategy can be constructed by buying the stocks of the top quintile, Q5, and selling short the stocks of the bottom quintile, Q1.

In order to investigate whether abnormal returns are present, a model must be used in order to estimate the expected stock returns. Two models will be used to test for the relation between characteristics of stocks and the expected returns of these stocks. The normal return is defined as the expected return of a security that would have occurred in the case the event did not take place (MacKinlay, 1997). As MacKinlay (1997) describes, there are basically two choices for estimating the normal return; the constant mean return model, and the market return model. In this thesis the market model will be used as well as Carhart’s four-factor Model, which will be discussed hereafter. The market model states that the return of any given stock is related to the return of the market (MacKinlay, 1997).

Hence, starting with the market model:

𝑟_𝑖 − 𝑟𝑓 = 𝛼_𝑖 + 𝛽_𝑖(𝑟_𝑚− 𝑟𝑓) + 𝜀_𝑖

The model describes how the return of a stock, or in this case quintile, is related to the return of the market: 𝑟_𝑖 is the average monthly return of all the suppliers in the specific quintile and

(19)

𝑟𝑓 is the risk-free return rate, 𝑟_𝑚𝑡 is the return of the return of the market, adjusted with the risk-free return rate as well. The factors 𝛼_𝑖, 𝛽_𝑖, and the variance of the error term are the parameters. MacKinlay (1997) explains how the market model is an improvement as opposed to the constant return model, since the variance is reduced and therefore an increase in the prediction of returns. The intercept of the model is captured by the alpha, 𝛼_𝑖, is also known as a measure of the abnormal return.

Ahern (2009) states that when a multifactor model is used, the results are less skewed in general, and are therefore better for the conduction of statistical tests. Therefore, this thesis will also use a multifactor model to estimate the normal returns. Factor models are known for reducing the variance of the abnormal returns by explaining more of the variance in the normal returns. For this reason the widely known factor model of Fama and French (1996) will be used, extended with the momentum factor suggested by Carhart (1997). Fama and French argue that the three-factor model reduces the variance of the market model by introducing two other factors. Namely, the difference between the return on a portfolio consisting of small stocks and the return on a portfolio consisting of large stocks, this is called the small minus big factor, SMB. This factor is also known as the size factor and is based on the market value of equity (ME) of a firm (Fama & French, 1993). It is calculated by subtracting the average return of large portfolios from the average return of small portfolios, as measured by their market capitalization. The third factor is the difference between the return on a portfolio consisting of stocks with a high book-to-market value and the return on a portfolio consisting of low-book-to-market stocks, this is called the high minus low factor, HML. The HML factor is also known as the value premium, meaning that companies known to have high market ratios are value stocks, whereas companies with low book-to-market ratios are growth stocks. Firms with high book-to-book-to-market ratios are known to earn positive excess returns (Pontiff & Schall, 1998). The final factor is the MOM factor. A momentum strategy is characterized by selling stocks that have performed badly in the past and buying stocks that have performed well in the past (Hong & Stein, 2007). Therefore, the expected return on a stock or portfolio is:

𝑟𝑖 − 𝑟𝑓 = 𝛼𝑖+ 𝛽𝑖(𝑟𝑚− 𝑟𝑓) + 𝑠𝑖𝑆𝑀𝐵 + ℎ𝑖𝐻𝑀𝐿 + 𝑚𝑖𝑀𝑂𝑀 + 𝜀𝑖

The MOM factor is an important control variable, since the effect, if existing, could be due to the suppliers’ own past return. Stock return momentum is a widely known anomaly in asset pricing theories (Hirshleifer & Teoh, 2003). The intercept of the model, the alpha, captures

(20)

the abnormal performance of a stock. The alpha can be seen as a measure of performance which compares the predicted return, the one that would have been earned in the case the model entirely explains returns, with the realized return.

The model will be used to explain the returns of the suppliers, however, as even Fama and French indicate, it is “just a model”. Multiple factors can contribute to, if observed, abnormal returns of suppliers. It is therefore unwise to draw conclusions from the data and outcomes before any robustness checks are performed.

Thus, monthly excess returns will be regressed on the excess return of the market, the size, the value, and the momentum factors in order to compute the betas for each factor. Since these factors are known to explain most of the variability in stock returns, the expectation is that the intercept will be zero. If investors are not limited by their cognitive processing powers, and are able to incorporate information about economically linked firms, there will be no price drift across securities. Due to the recent technological changes there are two outcomes possible: investor inattention has disappeared because investors have become aware of these economic links, or, due to the incapability of investors to process these links and because of the increase distractions caused by mobile phone usage investor inattention has increased. This thesis will assume some amount of inattention, therefore the expectation is that there is, indeed, an underreaction of stock prices between related firms, and therefore the possibility to obtain abnormal returns, hence the alpha.

Both the market model as well as the Carhart four-factor model will be performed twice. Once for the quintile portfolios formed in which each stock contributes equally: the equal weighted portfolios. Since firms tend to differ in terms of their size, quintile portfolios based on the market capitalization will be formed as well: the value weighted portfolios. The value weighted portfolios illustrate the contribution of large-cap stocks to the quintile portfolios.

Duration of the Inattention Effect

To investigate the speed of information diffusion the lags will be increased up to 12 months, to examine the development over time. Supplier stocks will thus be assigned to one of the five quintiles based on their customer returns in the 2, 3, 4, 5, 6, or 12 months before. The basic regressions discussed earlier on in this section will be performed on all the quintiles for all the months mentioned. This method allows us to study whether there are still abnormal returns, alphas, present after the first month. Additionally, the lagged customer returns will be regressed on the suppliers’ current returns in order to investigate whether they are able to explain the current returns.

(21)

If there is a lagged response of the suppliers’ stock price that is beyond the one-month lag examined earlier, one of the lags will be significantly different from zero. The Carhart four-factor model is used, extended with one of the lagged customer returns:

𝑟_𝑖 − 𝑟𝑓 = 𝛼_𝑖 +𝛽_𝑖(𝑟 − 𝑟_𝑓) +𝑠_𝑖𝑆𝑀𝐵 + ℎ_𝑖𝐻𝑀𝐿 + 𝑚_𝑖𝑀𝑂𝑀 +∑ 𝐶𝑢𝑠𝑡𝑜𝑚𝑒𝑟𝑅𝑒𝑡𝑢𝑟𝑛_𝑡−𝑛

12

𝑛=1

+ 𝜀_𝑖

If customer returns continue to affect supplier returns, one of the coefficients of the lagged customer returns will be significantly different from zero.

(22)

Data and Descriptive Statistics

Through the Compustat database the Segments database was entered to retrieve the customer-supplier relationships. The database provides publicly listed companies together with their customers, if these customers count for at least 10% of the total sales of the suppliers. Beginning the sample by using all firms listed in the CRSP/Compustat database allows us to work with a sufficiently large enough sample. Customer Type includes the method used by a company to organize their customers, for example: Company or Government. The customer name and the amount of sales generated by the particular customer were reported. Since some customer names did not supply any information concerning that customer, for example when their name was “US Government” or “Arizona”, these companies were manually deleted from the sample.

As opposed to the period of Cohen and Frazzini (2008) from 1980 until 2004, I will use a more recent period, namely, the period will run from 1999-2017. This not only gives more insides into whether recent developments in the financial markets have made investors more attentive, but it also circumvents the problem of using phonetic string matching algorithm. Following the approach of Cohen and Frazzini (2008) the corporate customer names will be matched with their Compustat identifiers (GVKEY). Companies that did not have a unique match were deleted from the sample. From the CRSP database monthly stock returns were subtracted together with the shares outstanding, trading volume and the price per share. As suggested by Menzly and Ozbas (2010), small stocks are excluded from the sample in order to ensure the lead-lag effect does not arise from a delayed price response. Therefore stocks of which the price was lower than 5 dollar were deleted from the sample. Even though penny stocks are excluded from the sample, liquidity constraints can still be a contributing factor to the observed lead-lag effect, especially for stocks that are traded infrequently. Therefore to control for possible liquidity effects, only stocks that enjoyed a strictly positive volume every trading day in the previous 12 months will be included. In order to control for these volume effects, daily data from CRSP is needed, to ensure all traded stocks are liquid within the limits set. Since this thesis will use monthly data, the daily data is only used to inspect the trading volume of stocks, in the case the volume was not strictly positive, the data was manually deleted from the main file with monthly data.

From Compustat the book value of equity was calculated by deducting a firm’s total liabilities from total assets. The market value of equity is calculated by shares outstanding times the price, the price can be retrieved from CRSP, and shares outstanding from Compustat

(23)

or CRSP. Additionally, Compustat provided the earnings before income and taxis (EBIT) and the net income of all firms as well. SIC (Standard Industrial Classification) codes were obtained in order to further analyze whether lead-lag effects, if detected, stem from intra-industry effects.

From the WRDS BetaSuite database the factors from Carhart’s model were subtracted. Betas and factor loadings were reported for all US common stocks together with their ticker code. The factors were based on daily data, meaning consisting of 252 trading days, therefore there were 252 factor data points per firm per year. Since this study uses monthly stock data, the factors were collapsed to monthly. From I/B/E/S the number of analysts providing forecasts for any given security throughout the sample are subtracted. For calculating the speed of information diffusion the returns of the customer as well as the supplier must be obtained. The CRSP database, via WRDS, provides the returns but also the stock prices of customers as well as suppliers.

Entire Sample Return Statistics

The table shows the summary statistics for the returns of the customer firms, supplier firms, and market throughout the sample. The sample period covers the period between January 1999 and December 2017. Returns are not displayed in excess of the risk-free rate, they thus represent gross returns.

N Mean Standard

Deviation Min Max

Customer Return 56,403 0.0112 0.106 -0.710 1.776

Supplier Return 56,183 0.0217 0.167 -0.813 4.227

Market Return 56,403 0.0042 0.041 -0.169 0.108

CRSP Stocks 1,248,381 0.0158 0.140 -0.984 15.98

In my sample the return of the suppliers is higher on average than the return of the customers. The average return for the customers was 1.12%, while the average return for the suppliers was 2.17%. The standard deviations of both the customer as well as the supplier returns are quite high in relation to their means; 10.6% and 16.7%, respectively. The final row in the table represents all traded US stocks for the period 1999-2017 in the CRSP database. The column was included to give some insights into how the returns of the customers and suppliers behaved in relation to all other traded stock during the period. Stock prices lower than 5 USD were excluded from the sample in order to make a legitimate comparison. The average return for all stocks was 1.58% with a standard deviation of 14%. Even though the statistical tests will be performed on returns in excess of the risk-free rate, in the table returns are not displayed in excess of the risk-free rate. It is worth mentioning that the risk-free rate

(24)

was around zero throughout the sample, in contrast to the market in general which experienced quite some fluctuations, especially during the crisis of 2008. Figure 2 in the appendix shows the evolution of the customers’ returns as well as those of the suppliers. Figure 3 in the appendix shows the development of both the risk-free rate as the market return throughout the sample.

Summary Statistics Stock Fundamentals

The table shows the average price, trading volume, and shares outstanding for suppliers, customers, and the overall market, respectively. The table additionally shows standard deviations, minimum values, and maximum values. The period over which averages are calculated spans between 1999 and 2017, the period of interest. Stocks with prices lower than 5 dollar were excluded from the sample. Additionally, only stocks that enjoyed a strictly positive volume every trading day in the previous 12 months were included.

Mean Standard

Deviation Min Max

Stock Information

Price Supplier 29.09 26.44 5 477.8

Volume Supplier 545,989 1,856,874 14 30,441,987

Shares Outstanding Supplier 276,870 1,186,115 154 10,880,222

Price Customer 47.22 38.43 5 1216

Volume Customer 2,462,512 4,067,354 237 87,967,800

Shares Outstanding Customer 1,296,221 1,962,513 2,633 11,144,681

Average Price CRSP 53.05 1,556 5 297600

Average Volume CRSP 187,873 925,583 0 118,895,338

Average Shares Outstanding CRSP 942,281 359,976 1 11,144,681 While both minimum prices were bounded to 5 dollar per share, the mean, as well as the maximum price of supplier stocks, was somewhere between two and three times smaller than the mean and maximum price of the customers. The average price of the supplier was 29.09, whereas the average price of the customer was 47.22, both of them being lower than the average price in the overall market of 53.05. The average trading volume of customers was approximately 4.5 times greater than the average trading volume of suppliers. The number of shares outstanding of customers was approximately 4.7 times greater than the number of shares outstanding for suppliers. Trading volumes, as well as the number of shares outstanding, display large standard deviations, possibly meaning that the sample has (very) large firms as well as (very) small firms. In comparison to all traded US securities during that period, prices of both supplier and customer firms were lower than the average of all stocks traded. Suppliers, as well as customers, had a higher trading volume than the overall market and the number of shares outstanding was lower than the market in the case of suppliers and

(25)

higher than the market in the case of customer firms. The results appear to be in line with the results of Shahrur, Becker, and Rosenfield (2010), who found that supplier firms were on average smaller than the customer firms. The reason for this finding is partly due to the data gathering process: since suppliers are only obliged to report customers who count for, at least, 10% of their sales, a customer needs to represent a large part of the annuals sales, it is therefore more likely to identify a large sized firm.

Summary Statistics Firm Characteristics

The table shows the summary statistics of accounting data of both suppliers and customers. The table shows the averages of each variable, together with the standard deviation, the minimum value, and the maximum value that have occurred in the sample period. The sample period spans between January 1999 and December 2017. The variables displayed are; Total Assets, Revenue, Earnings Before Income and Taxes (EBIT), Net Income, Total Liabilities, and Book Value are all in millions. B/M represents the book-to-market ratio and is calculated by dividing the book value of equity of a firm by the market capitalization of that firm.

Mean Standard

Deviation Min Max

Suppliers Total Assets 5,189 14,123 0.795 212,949 Revenue 4,859 13,582 0.83 107,552 EBIT 746 2,741 -147 27,956 Net Income 533 2,068 -6,729 23,150 Total Liabilities 2,722 7,596 0 122,503 Book Value 2,512 8,266 -2,951 90,446 B/M Ratio 0.681 3.420 -0.558 323 Customers Total Assets 99,575 368,595 1.213 2,281,234 Revenue 43,183 59,748 0 43,352 EBIT 5925 10,099 -25,913 66,290 Net Income 3453 6,498 -23,119 4,835 Total Liabilities 77,465 241,388 0.152 2,036,661 Book Value 22,160 38,952 -5293 267,146 B/M Ratio 0.449 0.627 -5.55 19.78

In terms of total assets, the average customer is around 30 times larger than the average customer. Revenue, earnings before income and taxes (EBIT), and net income all vary substantially as well. The book-to-market ratios are both smaller than 1, indicating that the average sample can be seen as growth stocks. Since previous literature has shown that value stocks, instead of growth stocks, are known to earn positive excess returns, the contribution of the B/M to the model value might be biased (Pontiff & Schall, 1998). Although it cannot directly be concluded from the ratio alone, a low B/M ratio can also stem from a company with many intangible assets, or a natural resource company with high short-lived earnings or a

(26)

stock with low risk causing the future cash flows to be discounted by a lower rate (Lakonishok, Schleifer, & Vishny, 1994). As the average customer has a large size in comparison to the average supplier in the sample, there possibly is a correlation between the customer return and the return of the corresponding industry (Cohen & Frazzini, 2008). During the robustness checks this issue will be addressed by adjusting the sample to have a similar exposure to the underlying industry.

Matrix of Correlations

Table shows the correlation coefficients over the entire sample. The first variable reported, ReturnSup, is the monthly gross return of the suppliers in the sample. ReturnCust is the monthly customer gross return. Size is defined as the market capitalization of the supplier, calculated by multiplying shares outstanding with the stock’s price. B/M is the book-to-market-ratio, which is calculated by dividing the book value of assets (total assets minus total liabilities) by the market value of equity. Profitability is the suppliers’ profitability, which is in this case defined as the ratio of net income/total assets.

ReturnSup ReturnCust Size B/M Profitability

ReturnSup 1.000

ReturnCust 0.275 1.000

Size -0.008 -0.009 1.000

B/M -0.011 0.000 -0.029 1.000

Profitability 0.010 -0.013 0.055 -0.027 1.000

As Hong and Stein (1999) mention, correlated fundamentals is a necessary condition for cross-predictability to exist. The correlation between customer returns and supplier returns is positive with a value of 0.275. Suppliers tend to have a small, yet negative correlation with their returns, with a correlation coefficient of -0.008. Banz (1981) demonstrated that small stocks tend to have higher risk-adjusted returns than large stocks, therefore the observed negative correlation can be quite reasonable. A supplier’s size and book-to-market ratio appear to have a small, but negative correlation with the return of the supplier. According to Fama and French small firms are more prone to experience negative earnings for a longer period, as opposed to big firms who are frequently bypassed by these earnings depressions. Profitability has a small but positive correlation with the supplier returns, and a small but negative relationship with customer returns, with correlation coefficients 0.010 and -0.013, respectively.

(27)

Results

This section will present the primary results of the study. First of all the regression outcomes using the market model, both equal and value weighted, will be presented, followed by a short discussion of the results. Thereafter the quintile regressions will be performed using the Carhart four-factor model, subsequently, the monthly abnormal returns are presented, for both the market model and the Carhart four-factor model, which is followed by the table listing the abnormal returns for each model, both equal- and value weighted. A discussion about the results of the Carhart four-factor model and the main results in the table with abnormal returns will be given. Then the test concerning the speed of information diffusion will be conducted, followed by the conclusions.

Table Ia

Market Model: Equal Weighted

The table shows the excess returns for all quintiles using the market model and all stocks in each quintile are equally weighted. The returns shown in the table are all in excess of the risk-free rate. The intercept represents the alpha obtained from the monthly regressions. The explanatory variable is the excess return on the market. Due to significant results when tested for heteroscedasticity, heteroscedasticity-consistent standard errors were used, which are shown in parentheses.

Q1 Q2 Q3 Q4 Q5 Market 1.461*** 1.337*** 1.278*** 1.3285*** 1.387*** (0.0329) (0.0353) (0.0324) (0.0342) (0.0409) Alpha -0.00940*** 0.00438*** 0.0111*** 0.0225*** 0.0380*** (0.00130) (0.00134) (0.00129) (0.00133) (0.00152) Observations 11,602 11,304 11,242 11,227 10,808 R-squared 0.152 0.127 0.121 0.117 0.110

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(28)

Table Ib

Market Model: Value Weighted

The table shows the excess returns for all quintiles using the market model and all stocks in each quintile are value weighted, meaning that the stocks are weighted based on their market capitalization. The returns shown in the table are all in excess of the risk-free rate. The intercept represents the alpha obtained from the monthly regressions. The explanatory variable is the excess return on the market Due to significant results when tested for heteroscedasticity, heteroscedasticity-consistent standard errors were used, which are shown in parentheses.

Q1 Q2 Q3 Q4 Q5

Market 0.0267*** 0.0228*** 0.0266*** 0.0229*** 0.0262***

(0.00164) (0.00150) (0.00201) (0.00159) (0.00213) Alpha -0.000341*** -9.08e-05 0.000132** 0.000560*** 0.000882***

(5.27e-05) (4.80e-05) (5.82e-05) (5.77e-05) (6.53e-05)

Observations 11,602 11,304 11,242 11,227 10,808

R-squared 0.039 0.034 0.032 0.022 0.022

Table Ia and Ib show the results of the quintile regressions using the market model. The equal-weighted market model has an intercept of -0.94%, which represents the monthly alpha, for the suppliers with the worst performing customers in the previous month. The suppliers whose customers had the highest returns the previous month had a monthly alpha of 3.80%. The betas of the excess market returns are all significant and higher than 1, implying the securities used in this sample are on average more volatile than the market. The first quintile of the value-weighted market model has a monthly alpha of -0.03% whereas the fifth quintile has a monthly alpha of approximately 0.09%. The alphas appear to rise monotonically across all quintiles and are all significant except for the second quintile in the value-weighted portfolio. Even though the market betas are all significant as well, as in the equal-weighted model, they appear to be much lower in the value-weighted model, which corresponds to the observed returns as well. The higher betas observed can be due to sample selection, implying that some high-beta industries are overrepresented in the sample used in this study. For example, SIC codes 2834, 3674, and 6062 account for approximately 15% of the sample and are known to have an average beta of 1.21 given the industry in which they operate1. When the analysis is conducted using the value-weighted method there appears to be a substantial difference, which indicates that firms with a large market capitalization seem to underperform firms with a small market capitalization. It also indicates that small stocks seem to exhibit

(29)

more cross-predictability than large stocks. Shahrur, Becker, and Rosenfield (2010) found similar results when inspecting the differences between value- and equal-weighted portfolios.

Table IIa

Carhart Four-Factor Model: Equal Weighted

The table shows the abnormal returns using the Carhart Four-Factor Model in which each stock is equally weighted. Each month the supplier stocks are ranked in ascending order based on the returns of their customers in the previous month. Market represents the beta of the market excess returns. The SMB and HML factors are the size and book-to-market factors, respectively. Due to significant results when tested for heteroscedasticity, heteroscedasticity-consistent standard errors were used, which are shown in parentheses. Q1 Q2 Q3 Q4 Q5 Market 1.204*** 1.161*** 1.074*** 1.011*** 1.058*** (0.0363) (0.0366) (0.0372) (0.0369) (0.0480) SMB Factor 0.882*** 0.678*** 0.743*** 1.047*** 1.144*** (0.0742) (0.0742) (0.0726) (0.0715) (0.0753) HML Factor -0.325*** -0.0479 -0.0264 -0.0724 -0.255*** (0.0542) (0.0509) (0.0514) (0.0524) (0.0658) MOM Factor -0.119*** -0.0962* -0.125*** -0.154*** -0.179*** (0.0356) (0.0493) (0.0384) (0.0431) (0.0531) Alpha -0.0101*** 0.00323** 0.00994*** 0.0209*** 0.0370*** (0.00126) (0.00134) (0.00126) (0.00129) (0.00153) Observations 11,602 11,304 11,242 11,227 10,808 R-squared 0.194 0.142 0.148 0.165 0.160

(30)

Table IIb

Carhart Four-Factor Model: Value Weighted

The table shows the abnormal returns using the Carhart Four-Factor Model in which each stock is weighted based on their market capitalization. Each month the supplier stocks are ranked in ascending order based on the returns of their customers in the previous month. Market represents the beta of the market excess returns. The SMB and HML factors are the size and book-to-market factors, respectively. Due to significant results when tested for heteroscedasticity, heteroscedasticity-consistent standard errors were used, which are shown in parentheses.

Q1 Q2 Q3 Q4 Q5 Market 0.0256*** 0.0213*** 0.0246*** 0.0194*** 0.0232*** (0.00179) (0.00160) (0.00210) (0.00161) (0.00258) SMB Factor 0.00365* 0.00620*** 0.00351 0.0112*** 0.00970*** (0.00211) (0.00228) (0.00320) (0.00304) (0.00277) HML Factor -0.00572** -0.00430* -0.00854*** -0.00815*** -0.00687** (0.00234) (0.00240) (0.00265) (0.00286) (0.00335) MOM Factor 0.000129 0.000213 -0.00163 -0.00173 -0.000988 (0.00131) (0.00123) (0.00161) (0.00167) (0.00238) Alpha -0.000334*** -9.67e-05** 0.000154*** 0.000563*** 0.000886***

(5.38e-05) (4.70e-05) (5.76e-05) (5.62e-05) (6.78e-05)

Observations 11,906 12,694 12,662 12,623 11,906

R-squared 0.025 0.037 0.034 0.027 0.025

Tables IIa and IIb show the regression results using the Carhart four-factor model. Below the results of a monthly long/short trading strategy will be given, the main results of this study. Thereafter the results of the Carhart four-factor regressions will be discussed as well, which are followed by the tests for the speed of information diffusion.

Table III

Monthly Abnormal Returns

The table shows the monthly abnormal returns/alpha for the market model as well as Carhart’s four-factor model using a long/short strategy. The abnormal returns displayed are obtained by a long position in suppliers belonging to the fifth quintile, and going short in suppliers belonging to the first quintile.

Market Market Carhart Carhart

Equal Weighted Value Weighted Equal Weighted Value Weighted

Alpha 0.0474*** 0.0012*** 0.0471*** 0.0012***

Table III shows the main results of this thesis. It shows the monthly abnormal returns obtained by following the long/short strategy. The alpha represents the abnormal return that can be achieved in month t if a portfolio of supplier stocks is formed on the basis of the customer

(31)

return in the previous month, t-1. For the equal-weighted portfolio, using the market model, a monthly abnormal return of 4.74% can be obtained, whereas the value-weighted portfolio is able to achieve a monthly abnormal return of 0.12%. When the Carhart four-factor model is used to control for the factors, the results do not seem to alter much. The Carhart four-factor model delivers a monthly abnormal return of 4.71% in the case of an equal-weighted portfolio, while the value-weighted portfolio delivers a monthly abnormal return of 0.12%. Overall the results so far suggest the existence of investor inattention, which results in the predictability of stock returns among customer-supplier links.

After the discussion of the main results of the other explanatory variables, a brief discussion concerning these apparent differences between value and equal weighted portfolios will be discussed. The small minus big (SMB) factor is positive throughout the sample and significant at the 1% level for all quintiles in the case of the equal-weighted portfolio, and significant for the quintiles two (5% significance level), four (1% significance level), and five (5% significance level) in the case of the value weighted portfolio. This result indicates that the returns on small stocks are larger on average, throughout the sample, than the returns on big stocks.

The high minus low (HML) factor appears to be negative for all quintiles in both the equal and value weighted portfolios. For the equal weighted portfolio, the first and fifth quintile are significant at the 1% level, all other quintiles are insignificant. In the case of the value weighted portfolio, the first quintile is significant at the 5% level, the second quintile is insignificant, the third quintile is significant at the 1% level, and the fourth and fifth at the 10% level. Firms with high book-to-market ratios, according to the literature, outperform firms with low book-to-market ratios. The portfolios seem to have a negative relationship with the value premium, which implies they overall behave more like growth stock portfolios. When looking at the average book-to-market ratio of the sample one can conclude the ratio is smaller than 1. According to Lakonishok, Shleifer, and Vishny (1994), stocks with low book-to-market ratios are overpriced, on average. The sample appears to be overrepresented with companies who have low book-to-market ratios.

Finally, the equal weighted portfolio the momentum factor is negative and significant at the 1% level for all quintiles. This would imply that buying stocks that performed well in the previous month results in negative returns in the future. Whereas buying stocks that have performed badly would yield a positive return. Jegadeesh (1990) found negative returns for individual stock momentum strategies as well. This applies to short-term strategies since the

(32)

first-order serial correlation is negative and significant for monthly stock returns (Jegadeesh, 1990).

The apparent large difference between equal weighted and value weighted results will be discussed shortly. Below are the summary statistics for both equal and value weighted returns.

Summary Statistics Equal and Value Weighted Returns

The table shows the average return statistics over the entire sample using either the equally weighted method, or the value weighted methods. Means, standard deviations, the minimum, and maximum values are given. Returns are displayed in excess of the risk-free.

Variable N Mean Standard

Deviation Min Max

Equal Weighted 56,183 0.0201 0.167 -0.817 4.222

Value Weighted 56,183 0.0003 0.007 -0.154 0.455

When looked at how much each firm contributes in terms of market capitalization to the portfolio, large firms seem to underperform the small firms in terms of the alphas. With monthly rebalancing, the equal-weighted portfolio outperforms the value-weighted portfolio in terms of one and four-factor alphas. The equal weighted does, however, have a higher standard deviation. A possible explanation for these results has been carried forward by Plyakha, Uppal, and Vilkov (2014) who found similar results when studying differences between value and equal-weighted portfolios. The reason for a higher alpha in equal-- portfolios, both the market and the four-factor model, is due to the monthly rebalancing to maintain equal weights. The reasoning behind this theory is that it requires a so-called “contrarian” strategy, which entails buying stocks whose prices have decreased and selling stocks whose prices have increased. Fama and French suggest using value-weighted returns since the portfolios will then better capture the different return behaviors between small and big firms, and value and growth firms. According to Loughran and Ritter (2000), the method of equal weighting will result in higher abnormal returns if investor inattention is more apparent among small firms than in big firms.

When the robustness tests are performed, the equal-weighted four-factor model, as well as the value-weighted four-factor model, will be reported, however, tests using the market model are not, these tables can be found in the appendix.

Duration of the Inattention Effect

The duration of the inattention effect is studied by investigating the development of the stock returns over time. Each customer is assigned to one of the five quintiles based on the lagged

(33)

customer returns. The market model and the Carhart four-factor model are estimated in a likewise manner as done above. As seen in the table below, the returns of suppliers in the subsequent months are studied by increasing the lag up to 12 months.

Table IVa

Speed of Information Diffusion: Equal Weighted

The table reports the alphas resulting from the market model as well as the four-factor model using equally weighted portfolios. Each month the supplier stocks are ranked in ascending order based on the returns of their customers in either 2, 3, 4, 5, 6, or 12 months before. The market beta, the SMB factor, and the HML factor are left out of the table. Due to significant results when tested for heteroscedasticity, heteroscedasticity-consistent standard errors were used.

Panel A: 2-month lag Q1 Q2 Q3 Q4 Q5

Market alpha 0.0128*** 0.0119*** 0.0156*** 0.0112*** 0.0123*** Four-factor alpha 0.0113*** 0.0105*** 0.0150*** 0.00969*** 0.0116***

Panel B: 3-month lag Q1 Q2 Q3 Q4 Q5

Panel C: 4-month lag Q1 Q2 Q3 Q4 Q5

Panel D: 5-month lag Q1 Q2 Q3 Q4 Q5

Panel E: 6-month lag Q1 Q2 Q3 Q4 Q5

Panel F: 12-month lag Q1 Q2 Q3 Q4 Q5