Geographic industry concentration and the comovement of stock prices

Geographic Industry Concentration and the Comovement of Stock Prices

Author: Stacy Otten Student number: 10654224

Supervisor: Martijn Dröes

University of Amsterdam, Amsterdam Business School July 2015

MSc Business Economics, Finance Track Masters Thesis

Statement of Originality

This document is written by Student Stacy Otten who declares to take full responsibility for the contents of 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.

Abstract

While there is growing literature on the economic importance of geography, there is little research on its effect within financial markets. This thesis attempts to provide insight into how geographic industry concentration affects annual firm stock returns. Its main findings indicate that there is a local effect present within investment decisions but is not related to spatial industry concentration. The findings show a significant comovement of stock prices among firms that are located within the same area and operate within the same industry that is persistent across time. However, there is no evidence to show that this comovement is related to the spatial concentration of industries. The lack of significance in the effect of industry concentration on the comovement of annual stock returns found among firms located within the same industry and area dispute the argument surrounding the phenomenon of local investor bias that geographical proximity plays a role in investor recognition.

Table of Contents  

Section 1: Introduction ...5

Section 2: Literature Review ...6

2.1: Local Investor Bias ...7

2.2: Dimensions of Proximity and Finance ...7

2.3: Comovement of Stock Prices ...9

2.4: Knowledge and Knowledge Transfer ...9

2.5: Knowledge Spillovers and Agglomeration Economies ...10

Section 3: Methodology ...10

Section 4: Data and Descriptive Statistics ...12

4.1: Industry Concentration ...14

4.2: Portfolio Descriptive Statistics ...16

Section 5: Results ...17

Section 6: Conclusion ...20

SECTION 1:INTRODUCTION

There is growing literature theorizing the economic importance of geography and evidence to indicate its relevance within financial markets. In particular, researchers in economic geography theorize that the transfer of information (i.e. facts and data) is geographically bounded due to the tacit nature of knowledge (Audretsch & Feldman, 1996). This theory has significant implications on how markets incorporate information into stock prices. If information is geographically bounded, its incorporation into financial markets will be bounded as well.

While current research on the specific topic of geographically bounded information within finance is limited, there is growing evidence to indicate that the speed of incorporation of information into financial markets is not-at least in the short-term- immediate and part of the reason for this point towards the spatial dispersion of economic activity. Garcia and Norli (2012) create a measure to distinguish between geographically dispersed and truly local firms by counting the number of times firms mention other states in their financial reports. They found that truly local firms (those which operate in a small number of states) have a higher stock return than those which are geographically dispersed. However, they attribute this difference in returns to investor compensation for insufficient diversification of the firm rather than an information advantage that is commonly suggested in literature surrounding home-bias of investors. Addoum, et al.

(2013) builds upon this framework and finds that information is not immediately

incorporated into stock prices. Specifically, the geographic dispersion of firms creates an ability to predict future fundamentals.

Despite limited research within this specific topic, there are other areas of research that touch upon geographically bounded information and its effects within the economy. In analyzing the investment decisions of firms, Dougal, et al. (2015) find that firms’ investments commove with local firms, suggesting that agglomeration economies play a role in investment decisions and firm growth. Supplementary to this, numerous researchers have found a home bias amongst investors which suggest that investors are benefiting from an information advantage and that there is a distinct distance-decay effect (the interaction between the firm and the investor decreases as the distance between them increases) present within those investment decisions (see for example Coval & Moskowitz, 2001; Chen, et al., 2011; Bok, et al., 2011; Ehrlich, et al., 2009). In economic geography, the effect of bounded knowledge has been well established. Economic geographers consistently show that knowledge transfer is geographically bounded and show a distinct distance-decay effect despite using codified knowledge as their measure of knowledge within their studies (Howells, 2002).

The aim of this paper is to build upon current financial literature by incorporating the theories surrounding agglomeration economies and investigate whether the concentration of economic activity can cause a comovement of stock prices. Barberis, Shleifer, and Wurgler (2005) have created a theoretical framework which incorporates market friction and investor irrationality into the comovement of stock prices; this is supported by Pirinsky and Wang (2006) who found a significant comovement effect in stock prices of firms headquartered within the same area and a change in the comovement when those

firms moved out of the area. I postulate that geography is one type of market friction, which can be detected by comparing the comovement of firm stock returns within an industry of a particular region, and intend to examine its effect on market prices. This leads to my primary research question:

What is the effect of the geographic concentration of an industry on the comovement of firm returns?

Looking at the United States between 1969 and 2000, I borrow from the framework of Dougal, et al. (2015) and classify each firm under a specific economic area (location) and industry. Using a dynamic fixed effects model, I compare whether the annual stock return of that firm is related to a portfolio other firms that are located within the same area and same industry. The annual stock returns of the firm were also compared against a measurement of geographic concentration for the industry and location in which the firm is placed. Included in my analysis was an interaction term between the same

industry-same area portfolio of firms and industry concentration.In addition, I compared whether

the returns are related to portfolios of other firms that are located within the same area but do not operate within the same industry, as well as those which are within the same industry but located outside of the firm’s economic area.

The results of the regression analysis show there is a persistent, positive significant relationship between firm j annual stock returns and the annual stock returns in the same industry-same area across time, which is not related to industry concentration. This provides evidence of strong local effects within the non-fundamental comovement of stock returns but there was no indication that industry concentration affects firm annual returns. The findings confirm the findings of Dougal, et al. (2015) and dispute the argument by Coval and Moskowitz (2001) that geographical proximity plays a role in a local information advantage among investors.

The remainder of this thesis is structured as follows:Section 2 examines current literature surrounding theories related to local investor bias, knowledge transfer and finance. Section 3 describes methodology used in this analysis. Section 4 describes the data and provides descriptive statistics of the sample. Section 5 discusses the results of the analysis. Finally, section 6 concludes this paper and provides suggestions for future research.

SECTION 2:LITERATURE REVIEW

Within the literature, there is little attempt to look at the relationship between spatial industry concentration and an information advantage among investors. However, there are areas of research that document an investor bias for local stock which point towards an information advantage and separately how agglomeration economies can provide economic benefits. In addition, other researchers describe the how geography affects innovation, knowledge transfer, and finance.

2.1: LOCAL INVESTOR BIAS

Perhaps the most well known example of how geography affects financial markets is the phenomenon of investor home-bias. The tendency for investors to invest in a large amount of domestic securities despite the benefits of an internationally diverse portfolio has been well documented within financial literature. The explanations surrounding the puzzle of home-bias range from institutional to behavioral to informational reasons. A large amount of literature surrounding local investor bias, both on a national and international scale, explains the phenomenon as being rooted in a local information advantage among investors. Coval and Moskowitz (2001) study mutual fund managers’ investment decisions and finds that they have a strong preference for geographically local stocks, which they argue is due to an informational advantage of stemming from geographic proximity. Ivković and Weisbenner (2005) test whether the local bias stems from an information advantage or a preference for familiarity and find that local investments generate a larger annual return which suggests investors are benefiting from an information advantage. Francis, et al. (2007) find that bondholders have a preference for local stocks and suggest this is due to a distance-decay effect which causes information asymmetries as well as a desire to mitigate agency problems between managers and investors. Garcia and Norli (2012) argue that local firm will have lower investor recognition than geographically dispersed firms and so will have higher expected returns to compensate for insufficient diversification. Berry and Gamble (2013) find that local retail investors’ trading prior to earnings announcements were able to predict announcement returns and investors not living near corporate headquarters, i.e. nonlocal investors, were unable able to make the same predictions. This, they argue, suggests investors are benefiting from a local information advantage rather than a preference for familiarity.

Other researchers have suggested that overconfidence plays a role in home-bias. Karlsson & Nordén (2007) look at Swedish pension plans and find that men have a greater tendency to perceive an information advantage in domestic assets than women and are more likely to hold portfolios with large amounts of domestic assets, which they attribute to overconfidence. However, Graham, et al. (2009) argues investors that display a tendency towards overconfidence will hold more internationally diverse portfolios. Further studies point towards institutional explanations for home-bias. Solnik and Zou (2012) say there are explicit barriers in investing internationally from regulations, transaction costs, etc. Lau, et al. (2010) supports the institutional argument by finding an inverse relationship between the cost of capital and home-bias.

2.2: DIMENSIONS OF PROXIMITY AND FINANCE

The literature surrounding home-bias indicate investors are gaining from some sort of local benefit; however, they have yet to identify along which dimension of proximity this benefit comes from. Boschma (2005) identifies five dimensions of proximity which help facilitate knowledge transfer and innovation: geographic, cognitive, organizational, social and institutional. Geographical proximity is the physical distance between two actors.

Geographers point out that actors are fundamentally rooted in space-time which should not be dismissed as it causes them to be territorially embedded within a local network of interdependencies between buyers and sellers (Pollard, 2003). This observation can be used to explore how securing capital for firms is rooted within geography and history; however the literature within finance predominantly identifies firms as placeless entities. Coval and Moskowitz (2001) and Addoum, et al. (2013) both suggest that their findings are related to geographical proximity; however, do not control for other types of proximity. Addoum, et al. (2013) simply state that market participants are unable to aggregate information very quickly.

Boschma (2005) describes institutional proximity as the distance between the set of rules under which economic actors conduct their activities. This is commonly recognized to be the governing body under which actors operate; however, it is not limited to country boundaries. For example, Wolf (2000) argues that if trade barriers were the culprit for home-bias then this should not be seen in trade within a country. He finds there is still home-bias present within the United States between states and attributes these findings to non-institutional reasons for why home-bias exists. What he fails to take into account is that the states within themselves have distinct institutional setups which greatly vary and so trade could still be negatively affected by this institutional distance as described by Boschma (2005). There are significant national differences in financial markets and corporate governance. National rules and a centralized financial market do not necessarily mean national homogenous corporate governance (Wojcik, 2002). Actors within the financial market, both firms and financial intermediaries, are repeatedly affected by national, regional and cyclical shocks which affect the supply and demand of credit (Pollard, 2003). These theories support the explanation that home-bias is a result of institutional settings.

Cognitive proximity is the shared knowledge base between two actors (Boschma, 2005). To give an example, when discussing ideas surrounding mathematical models, it is important for the two actors to have a base level of knowledge regarding these models in order to fully share their ideas. Related to cognitive proximity is organizational proximity, which is the extent to which actors share an organizational arrangement, either within or between organizations (Boschma, 2005). Two employees working for the same firm would be in close organizational proximity. Similarly, two employees working in separate firms but are engaging in a joint venture will also be within close organizational proximity of each other. Contracts are contextualized and formulated through a non-contractual background and learning, knowledge, rules and norms shape the expectations, beliefs and decisions in conditions of uncertainty (Pollard, 2003). Agents try to manage credit flows by participating in a network of peers and consumers while operating under historically and geographically specific working practices and conventions. In order to obtain finance, firms must conform to conventions of financial reporting that are recognized by financial intermediaries, and even as they conform to these standardized conventions, the lens under which their statements are viewed is based on the nature, intentions and constraints of the individual financial intermediaries (Pollard, 2003).

Boschma (2005) describes social proximity as the social embeddedness of two economic actors based on trust due to friendship, kinship, and experience. Geographers emphasize the social nature of economics, inherently supporting the behavioral explanations for home-bias. Investors primarily interested in a short-selling strategy for their portfolios will view an annual report much differently than those interested in holding the stock for the long-term (Pollard, 2003). Pirinsky and Wang (2006) explain their findings of local comovement of stock returns through a “word of mouth” effect where investors derive utility by trading information among local investors and tend to herd into and out of local securities as they compete for resources. Adding to this is the reality that the individual is born into a network of social relations –whether it is race, class, religion, gender, etc. – which shape attitudes towards entrepreneurship, risk tolerance, financial intermediaries, and the like.

2.3: COMOVEMENT OF STOCK PRICES

Barberis, Shleifer, and Wurlger (2005) argue that an economy with frictions or irrational investors will have comovement in prices that are separate from comovement in fundamentals. There are three ways in which this type of comovement can manifest itself: the category view, habitat view, and information diffusion view. The category view states that investors group assets into categories to simplify the decision-making process. This then causes prices to move based on a correlated sentiment among noise traders. The habitat view states that investors only trade a subset of all available securities due to transaction costs, international trading restrictions, or lack of information and will adjust their exposure to these securities as their risk aversion, sentiment, or liquidity needs change. Lastly, the information diffusion view states that information is incorporated into stock prices at different speeds due to market frictions. Under this framework, geography would be type of market friction that causes delays in the speed of transfer of information, thereby creating a local investment bias.

2.4: KNOWLEDGE AND KNOWLEDGE TRANSFER

Underpinning information flows is the concept of knowledge and knowledge transfer. Howells (2002) defines knowledge is defined as “a dynamic framework or structure from which information can be stored, processed and understood (pp 872).” Ambos and Ambos (2009) define knowledge as “accumulated practical skill or expertise that allows one to do something smoothly and efficiently (pp 2).” There are two types of knowledge: (1) codified knowledge - information that is transferred through a systematic, formal language and (2) tacit knowledge - information that is transferred through hands-on experience. It is important to note that codified and tacit knowledge are not discrete but exist on a continuum where some tacit knowledge is required for the interpretation of codified knowledge (Howells, 2002). It is generally recognized that tacit knowledge is not easily transmitted and that distance has a negative impact on its assimilation. However, researchers have also found that there is a distinct distance-decay effect present within the transfer of codified knowledge. Even though codified knowledge is more easily transferred, its interpretation is still influenced by geography. There are no clear market mechanisms which facilitate the transfer of tacit knowledge, and because tacit

knowledge is by definition difficult to articulate, its demand and consumption becomes a difficult process (Howells, 2002).

2.5: KNOWLEDGE SPILLOVERS AND AGGLOMERATION ECONOMIES

Related to the concept of the transfer of knowledge are knowledge spillovers, the process where knowledge is appropriated by other firms through a spillover effect. This process, like knowledge transfer, is largely assumed in neoclassical economics to be costless and easily transferred between people and organizations. However, researchers in economic geography have provided evidence to show that knowledge spillovers are spatially bounded through a series of studies mapping localized patent citations and universities as well as the movement of people and knowledge that moves with them (see Howells, 2002, for examples). It is assumed that industries, especially those which are knowledge-intensive, are concentrated, in part, because of knowledge spillovers.

Firms are theorized to realize scale economies when being in close proximity to each other by sharing pooled resources (Krugman, 1991). These economies are referred to as externalities or spillovers, of which two types are hypothesized to play a role in knowledge creation and diffusion (Beaudry & Schiffauerova, 2009). The first, called MAR spillovers or specialization externalities, states that firms which specialize within the same industry will realize scale economies when they locate close to each other because they share the same labor pool. As employees transfer across firms, they take with them ideas and knowledge related to the industry, which they then share with their colleagues and vice versa. The most famous example of an industry that resembles benefiting from MAR spillovers is Silicon Valley. Closely related to MAR spillovers are Porter spillovers which state that specialization externalities are realized through competition rather a local monopoly and cite jewelry industries as an example of how the combination of competition and close proximity facilitate knowledge growth and transmission. The second type of externality is called Jacobs spillovers, or diversification externalities, which state that firms from different industries benefit from being in close proximity to each because they bring with them different perspectives which stimulate the exchange of ideas and foster knowledge growth and creation.

SECTION 3:METHODOLOGY

The methodology of this thesis borrows from the framework of Dougal, et al. (2015) by creating a model where the firm annual stock return is compared against portfolios of other firms’ annual stock returns based on their location and industry. Each firm was classified under a specific area, a, and industry, i. A portfolio of returns was then created for every other firm based on the following four mutually exclusive categories: same industry-same area (i, a), same industry-different area (i, -a), different industry-same area (-i, a), and different industry-different area (-i, -a). Thus, as an example, if firm, j, operated within New York under the finance sector, four portfolios of annual stock returns were created for every other firm within the dataset. The first would be a grouping of firms which were also located within New York and operated in the finance sector

(same industry-same area). The second would be those firms which operated within the finance sector but outside of New York (same industry-different area). The third would be firms that are located within the area but operate outside of the industry, for example firms within New York but in the manufacturing sector. The last portfolio would be those firms that operated completely outside of the area and industry of firm, j, for example firms located in Los Angeles and operate within manufacturing (different industry-different area).

Doing this allows for a portfolio of returns to be created based on these categories and compared against firm, j, annual returns. The main regression equation is as follows:

𝑅_!,!!,! =  𝛼 +   𝛽!,!𝑅!,!!!!,! ! !!! +   𝛽!,!𝑅!,!!!!,!! ! !!! + 𝛽!,!𝑅!,!!!!!,! ! !!! +     𝛽!,! ! !!! 𝐿!,!_!!! + 𝛽_!,! ! !!! 𝐿!,!_!!!𝑅_!,!!!!,! +  𝛽_!𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠_!!,!+ 𝜀_!,!!,!   (1) Where the dependent variable 𝑅_!,!!,! is the annual stock return of firm j operating in industry i and area a during year t. On the right side of the equation, the first explanatory variable 𝑅_!,!!!!,! is a portfolio p of annual returns for firms which fall into the category same industry-same area for the current year t and two previous years (t – 1, t – 2). The firms that fall under this category operate within the same industry and same area as firm

j. Thus, if the firm was located in New York and operated within the finance sector, all

firms in this portfolio are also located in New York and operate within the finance sector. The coefficient on this variable indicates how investors perceive the riskiness of the firm based on its location and industry. This coefficient is of particular interest in this analysis because it effectively shows the comovement of annual returns for an agglomeration of firms within a region. A positive and significant coefficient on this variable would indicate that there is a non-fundamental comovement of stock prices between firms that are area and industry specific.

The second explanatory variable, 𝑅_!,!!!!,!! follows the same concept as the first but is for firms which fall into the category same industry-different area; thus, the firms that are in this portfolio operate within the same industry as firm j but in a different area. This variable controls for industry trends. The coefficient on this variable indicates the perceived riskiness of the firms that are operate within the industry but located outside of firm j’s region. A comparison of 𝛽! with 𝛽! will show the contrasts between the

perceived the riskiness of an industry inside and outside an area. A positive and significant coefficient would indicate a comovement of stock prices that are related to the industry rather than a spatial effect.

To add to the comparison, the third explanatory variable 𝑅_!,!!!!!,! is included to control for area-specific shocks. This variable follows the same concept as the above but is for firms located with the same area as firm j but do not operate within the same industry. Its

coefficient will indicate how investors perceive the riskiness of firms operating within the area but outside the industry. Comparing with the first variable, the same industry-same area portfolio, will provide an insight into how investors perceive the riskiness of the entire region relative to firm j. A positive and significant coefficient on this variable will indicate that there are spatial effects present within investment decisions, which cause stock prices to co-move with each other.

Both fourth and fifth explanatory variables measure the effect geographic industry concentration has on firm annual returns. 𝐿!,!_!!! is a direct measurement of industry concentration for the area and industry in which firm j is located. Its coefficient is best explained in comparing it with the first explanatory variable. 𝛽! captures the total local

effect of the agglomeration of an industry within an area without regard to exact distance between firms. 𝛽!, on the other hand, captures the effect of the density of the industry

agglomeration within the area. It accounts for the degree of compactness of firms within the same industry as firm j and its effect on firm j annual returns. 𝐿!,!_!!!𝑅_!,!!!!,! is an interaction term between the measurement of industry concentration and the same area-same industry portfolio. This variable is of particular interest to this study as it directly relates to the research question in how geographic industry concentration affects the comovement of stock returns. The coefficient on this term shows how the perceived riskiness of firms within the same area-same industry category is dependent upon the geographic concentration of the industry, i.e. how investors perceive the riskiness of firm annual returns in relation to the density of the industry within the region. A significant coefficient on this variable would confirm that any comovement found within the same industry-same area portfolio is at least partly due to the geographic density of the industry. This would provide evidence to support the hypothesis that the incorporation of information into financial markets is geographically bounded.

Five firm fixed effects regression with clustered standard errors (by firm) are then performed where contemporaneous portfolios, industry concentration and the interaction term were first compared, and then lagged variables were successively added. Thus, the first estimate compares firm annual returns with only the same industry-same area portfolio, industry concentration and the interaction term between industry concentration and same-industry same area portfolio for the current year. The second includes the same variables as the first but adds the contemporaneous different industry-same area portfolio. The third follows the same method as the second by adding the same industry-different area portfolio for the same year. The forth estimate includes the 1 year lagged variables for each variable. Finally, the fifth reaches the final form of the model by including the 2nd

year lagged variables.

SECTION 4:DATA AND DESCRIPTIVE STATISTICS

To gather the data, monthly stock prices, location and industry information for all public firms listed on the NYSE, NASDAQ, or Amex between January 1969 and December 2000 were first obtained. Monthly stock prices were downloaded from the CRSP

database, from which monthly returns were calculated for each firm and then annualized. To minimize the influence of outliers, annual returns were winsorized at the 1% level. Both location and industry information were obtained from the CRSP/COMPUSTAT Merged Database and matched to the CRSP monthly stock prices by date and permanent company number (permco for CRSP database; lpermco for CRSP/COMPUSTAT Merged Database). The location of a firm is identified by its ZIP code (addzip), which was then matched to its relevant economic area (CSA) as described below.

For every firm, their relevant SIC code was obtained and sorted into a broad grouping of industries in the same manner as Dougal, et al. (2015). Like Dougal, et al. (2015), these grouping are intentionally broad to capture the local effects that operate both within and across industries. Doing so reduces the risk of picking up meaningful cross-industry effects when comparing industries. Dougal, et al. used the Fama-French 12 industry categories as a method of classification; however, I used the first-tier hierarchal classification system that the US Department of Labor uses for its SIC system, called a Division structure. Divisions are divided into ten primary categories: (1) Agriculture, Forestry and Fishing; (2) Mining; (3) Construction; (4) Manufacturing; (5) Transportation, Communications, Electric, Gas and Sanitary Serves; (6) Wholesale Trade; (7) Retail Trade; (8) Finance, Insurance and Real Estate; (9) Services; (10) and Public Administration.

Regional employment data was then gathered for the United States between 1969 and 2000 from the US Bureau of Economic Analysis (BEA). The region of analysis (economic area) was the Combined Statistical Area (CSA). A CSA is grouping of adjacent micro- and metropolitan statistical areas that have been identified by the US Office of Management and Budget (OMB) as having significant social and economic ties (see OMB Bulletin No. 13-01). CSAs are used as an area identifier in this study as opposed to major metropolitan areas because it captures the employment exchange between areas allowing for a more complete overview of how far information is transferred. Specifically, CSAs capture firms that are close enough to their respective workers to share and transfer knowledge but may not necessarily be located within a major metropolitan area. Further, they capture the exchange of workers between two areas that would otherwise be lost in using metro- or micropolitan areas as unique area identifiers. For every CSA, employment levels were categorized into the same broad groupings as the industries that the firms operate.1

Starting from 1969, Table 1 provides a snapshot of the dataset in five-year intervals. The number of firms, CSAs and descriptive statistics (mean, median, min, and max) for firms within each CSA is provided. The table shows a progressive increase in the quantity of firm and CSA data as time becomes more current, where the number of firms (as well as                                                                                                                

1

The location of a firm is identified by its ZIP code (addzip) which was then matched to its relevant ZIP Code Tabulation Area (zcta) using the 2010 ZIP code to ZCTA relationship file provided by the US Census Bureau, which can be downloaded at:

http://catalog.data.gov/dataset/tiger-line-shapefile-2013-2010-nation-u-s-2010-census-5-digit-zip-code-tabulation-. ZIP codes are area identifiers used by the U.S. Postal Service (USPS) while ZCTAs are approximate area representations of ZIP codes that are used by the Census Bureau to present statistical data. Some ZCTAs represent multiple ZIP codes therefore it is important to take this step. The ZCTAs were then matched to their relevant metro- or micropolitan statistical area (CBSA) using the 2010 relationship file provided by the Census Bureau (which can be download here: https://www.census.gov/geo/maps-data/data/zcta_rel_layout.html). In turn, these were matched to their relevant CSA.

the average number of firms per CSA) doubles from 1975 to 1995 while the number of CSAs within the dataset increased by 20%.

Table 1. Firm and Area Statistics for Selected Years. This table shows the number of

firms, number of CSAs and descriptive statistics (mean, media, min and max) for firms within CSAs for the years 1969, 1975, 1980, 1985, 1990, 1995 and 2000.

Year

Number Firms

Number CSAs

Firms per CSA

Mean Median Min Max

1969 577 51 66 38 1 168 1975 1167 65 112 64 1 303 1980 1226 64 118 66 1 322 1985 1669 71 170 91 1 454 1990 1879 75 183 145 1 484 1995 2435 78 226 179 1 596 2000 2290 75 212 175 1 535 4.1: INDUSTRY CONCENTRATION

For every industry, i, and area, a, a location quotient (LQ) is calculated per year. A LQ is a commonly used to measurement of the industry concentration a region. The Bureau of Labor Statistics (BLS) defines a location quotient as a ratio that allows “an area’s distribution of employment by industry to be compared to a reference of base area’s distribution” (“Location Quotient”, nd). Location quotients compare the spatial distribution of an activity, in this case employment, to that of a base. The base is normally the country in which the activity is located (in this case the United States) but can also be a region or metropolitan area. For industry, i, in area, a, in year, t, a LQ is calculated as:

𝐿

=

!,!

!_!,! !_!

Where 𝐸_!!,! 𝐸!,! is the ratio of industry employment within a region by total industry

employment. This ratio represents the level of activity for an industry within the region, or in this thesis, the level of employment for an industry within a given CSA. 𝑋!,! 𝑋! is

the ratio of total regional employment by national employment. This gives the total employment of the region relative to the country. A LQ is interpreted in the following manner:

- A LQ = 1 indicates that the industry has the same share of its regional employment as it does on a national level.

- A LQ > 1 indicates that the industry is more concentrated within the region relative to the nation.

Figure 1 shows the average location quotient for each division within the United States between 1969 and 2000. Looking at this figure, the majority of industries within the United States have an average location quotient which is greater than 1. That is, throughout the sample time period, the industries tended to be more concentrated within a CSA. This is with exception to the Manufacturing industry, which was less concentrated on average relative to the United States. The tendency towards higher concentration provides some evidence that industries within the United States may be benefiting from agglomeration economies.

Figure 1. Average Location Quotient for each Division. Figure 1 graphs the average

LQ of each industry division for all regions across the entire sample period.

The level of concentration between industries through the time period shows how the economic landscape has changed. Figure 2 plots the average annual LQ over the sample time period (line) and the average annual LQ for each industry division (x’s). Throughout the 1970’s and into the early 80’s, the economic landscape within the United States resembled a series of Silicon Valley’s, where the trend was towards extremely concentrated industries (see the x’s in Figure 2). However, despite the large disparities between industry concentration across this period, the average annual LQ remained relatively constant. The large concentration of industries then diminished during the 1980’s and early 90’s, wherein the majority of industries remained within relative concentration of each other, causing the landscape to resemble a series of large metropolitan areas characteristic of having an amalgamation of industries. The trend of extreme concentration subsequently began to increase during the mid-90’s until the end of the sample period. An explanation for these trends could be that the United States goes

0 .5 1 1.5 2 2.5 Average LQ Retail Trade Agriculture Mining Transportation Finance Services Wholesale Trade Construction Public Manufacturing

through periods where firms within industries benefit from either specialization or diversification economies, which cause the concentration of industries to change depending on which type provides more benefit to firm

Figure 2. Average Location Quotient Across Time. Figure 2 plots the average annual

LQ over the sample time period (line) and the average annual LQ for each industry division (x’s).

4.2: PORTFOLIO DESCRIPTIVE STATISTICS

Table 2 and 3 report a relationship between firm j annual returns and the local industry annual returns. The average annual returns of both the firm and firms within the same industry and area are roughly the same (Table 2) and have a positive correlation between the two (Table 3). This relationship is also reflected in the correlation between the interaction term and annual returns, while all other correlations between firm j annual returns and portfolio returns and industry concentration are negative (Table 3). It is interesting to note that there is a negative correlation between firm j annual returns and the annual returns of firms which are located within its region but do not operate within the industry especially in the context of a very small negative correlation between firm annual returns and industry concentration. The negative correlation between firm annual returns and industry concentration is indicative of a spatial distance-decay effect, which has been previously shown to have a significant impact on knowledge transfer; however,

if spatial distance is the only hindrance within the economy, this correlation would be expected to diminish or be positive for firms once they enter the area. The negative correlation between firm annual returns and firms which are within the different industry-same area category indicate there is a social dimension to the distance-decay effect, which is larger than the spatial dimension if looking purely at correlations.

Table 2. Portfolio Statistics. This table reports summary statistics (mean, standard

deviation, minimum, 10th

, 50th

, and 90th

percentiles, and maximum) for firm annual returns, same industry-same area portfolio annual returns, different industry-same area portfolio annual returns, same industry-different area portfolio annual returns, industry concentration, and the interaction between industry concentration and the same

industry-same area portfolio of annual returns.

Mean Sd Min 10th ₅₀th ₉₀th _Max

Firm Returns 0.88 7.67 -19.11 -6.92 0.34 8.42 35.41 Same Industry-Same Area 0.89 2.45 -10.13 -1.89 0.82 3.68 19.72 Diff Industry- Same Area 0.27 1.96 -14.57 -2.05 0.39 2.35 10.71 Same Industry- Diff Area 0.30 2.10 -16.23 -2.04 0.44 2.47 11.27 LQ 1.32 5.76 0 0.007 0.04 3.80 229.47 Interaction 1.10 19.32 -481.09 -0.07 0.01 1.61 339.40

Table 3. Portfolio Correlation Matrix. This table reports the correlation matrix between

firm annual returns, same industry-same area annual returns portfolio, different industry-same area annual returns portfolio, same industry-different area annual returns portfolio, industry concentration and the interaction between industry concentration and the same industry-same area portfolio of annual returns.

  Firms Returns Same Industry-Same Area Diff Industry- Same Area Same Industry- Diff Area LQ Interaction Firms Returns 1.0 Same Industry-Same Area 0.48 1.0 Diff Industry- Same Area -0.09 -0.5 1.0 Same Industry- Diff Area -0.21 -0.6 0.7 1.0 LQ -0.006 -0.02 0.02 -0.02 1.0 Interaction 0.1 0.2 -0.1 -0.1 0.02 1.0   SECTION 5:RESULTS  

The results of the regression analysis show there is a persistent, positive significant relationship between firm j annual returns and annual returns in the portfolio same industry-same area across time, which is not related to industry concentration. Table 4 shows the results of the estimates across five different regressions. Within each regression, the effect of the firm annual returns on the same industry-same area portfolio (ria) remain consistent and significant indicating that, ceteris paribus, when the annual return of firm j increases by 1%, all other firms within the region and operating in the same industry will increase by 1.08% within the same year. This effect remains consistent even when taking into account previous years. The significance of the previous two years (as shown in regression 5) for the same industry-same area portfolio indicates that this effect remains in place over time. This effect remained highly significant throughout all regressions and the coefficient only decreased once the control for industry effects was added. It is interesting to note that to coefficient on the same industry-same area portfolio stays above 1, indicating that the volatility of the portfolio is higher than the firm. Because of the setup of this analysis, in that which firms are located in which portfolio is relative to the firm in the dependent variable, this increased volatility provides support to the argument that investors are allocating resources to firms within the same area and industry based only on the performance of a single firm.

Throughout all regressions, the interaction between industry concentration and the same industry-same area portfolio (lqria) remain insignificant with the exception of the last regression where its 2 year lag shows significance. Given that all other regressions are insignificant and the coefficients are very small and change signs it is likely that this coefficient, while statistically significant, has little economic significance to it. The insignificance of this term suggests that the comovement among firms found in the same industry-same area portfolio are not due to an information advantage stemming from geographical proximity as argued by Coval and Moskowitz (2001). The insignificance of variable for industry concentration (lq) across all regressions support these findings in showing that geographic industry concentration does not play a role on firm j’s annual returns. This provides evidence that any agglomeration economies gained by the firm are either incorporated into the market through its fundamentals or immediately incorporated via public information. Taking into account Garcia and Norli’s (2012) findings that the spatial concentration of firms has an effect on stock returns, these findings indicate that investors are capable of incorporating information on an industry level but are unable to do so on a firm level.

The positive significance of the different industry-same area portfolio (ra_i) across all regressions indicates that the comovement of the same industry-same area portfolio returns have to do with a local effect rather than an industry effect. Like the same industry-same area portfolio, the different industry-same area portfolio returns remain positive and significant within the same year as firm j. This effect becomes insignificant in year 2 but remains significant and is interestingly higher with a 1-year lag. Because this variable takes into account all other firms within the same region as firm j but operate outside of its industry, and the industries were intentionally broad to capture any cross-industry relationships, this effect can be attributed to a local relationship rather than the industry. This supports previous literature showing that there is a spatial dimension investment decisions. These findings support those of Dougal, et al (2015) where strong

local effects were found in corporate investment strategies. In addition to the spatial dimension, the significance of the contemporaneous same industry-different area portfolio (ri_a) within regression 3 provides evidence that there is a social dimension to investment decisions; however, the succession of regressions adding the yearly lags for the same industry-different area portfolio indicate that this effect is incorporated into the market over the long-term. These effects, however, do not seem to be related to the density of the industry.

Table 4. Fixed-Effects Regression Estimates This table reports estimates of the following regression:

R!,!!,!=  α +   β!,!R!,!!!!,! ! !!! +   β!,!R!,!!!,!!! ! !!! + β!,!R!!,!!,!!! ! !!! +     β!,! ! !!! L!,!!!!+ β!,! ! !!! L!,!!!!R!,!!,!!!+  β!Controls!!,!+ ε!,!!,!  

where the dependent variable R!,!!,! is the annual return of firm j operating in industry i and area a during year t, R!,!!!!,! (ria) is an

equally-weighted portfolio p of annual returns for firms which fall into the category same industry-same area for the current year t and two previous years (t – 1, t – 2), R!,!!!,!!! (ra_i) is a portfolio of annual returns for firms which fall into the category different

industry-same area for the current year and previous two years, R!,!!!,!!! (ri_a) is a portfolio of annual returns for firms which fall into the

category same industry-different area, L!,!!!! (lq) is the measurement of industry concentration under which firm j is located and

operates, and L!,!!!!R!,!!!!,! (lqria) is the interaction term between industry concentration and the portfolio of returns for same

industry-same area firms. Five firm fixed effects regressions with clustered standard errors by firm were performed where contemporaneous portfolios, industry concentration, and the interaction term was compared then their lagged variables were added. t-statistics are in parentheses. ***p<0.001, **p<0.05, *p<0.10 (1) Return (2) Return (3) Return (4) Return (5) Return

Same industry-Same area

ria (contemp.) 1.39*** (40.13) 1.58*** (43.83) 1.61*** (43.74) 1.18*** (25.32) 1.08*** (19.37)

ria (1 year lag) 0.20***

(4.64)

0.11** (2.05)

ria (2 year lag) 0.13 **

(2.35) Diff industry-Same area

ra_i (contemp.) 0.35*** (11.88) 0.31*** (8.94) 0.14 ** (3.06) 0.11 * (1.97)

ra_i (1 year lag) 0.20***

(3.94)

0.20** (3.01)

ra_i (2 year lag) 0.05

(0.74) Same industry-Diff area

ri_a (contemp.) 0.09** (2.45) -0.03 (-0.59) -0.02 (-0.29)

ri_a (1 year lag) 0.04

(0.72)

-0.009 (-0.16)

ri_a (2 year lag) 0.06

(0.89) Industry Concentration lq (contemp.) -0.002 (-0.11) -0.003 (-0.18) -0.002 (-0.14) -0.003 (-0.09) -0.02 (-0.44) lq (1 year lag) 0.02 (0.54) 0.04 (0.59) lq (2 year lag) 0.03 (0.84) Interaction lqria (contemp.) 0.001 (0.44) -0.0002 (-0.08) -0.0005 (-0.19) 0.004 (0.71) -0.002 (-0.40)

lqria (1 year lag) -0.006 (-1.05)

0.006 (0.86)

lqria (2 year lag) -0.01**

(-1.98) Constant 0.17*** (6.21) -0.05 (-1.50) -0.09** (-2.47) 0.008 (0.11) -0.04 (-0.29)

Firm fixed effects Yes Yes Yes Yes Yes

Observations 15227 14804 14804 6390 3423

R2 _{0.242 0.245 0.246 0.251 0.262}

SECTION 6:CONCLUSION

Multiple studies have documented a local-bias amongst investors; however the mechanism through which this operates is debated. Many researchers suggest investors benefit from an information advantage. This notion is supported by parallel research within economic geography showing a distinct distance decay effect present within the transfer of knowledge and suggests that information is geographically bounded. Barberis, et al. (2005) create a theoretical framework under which information diffusion can cause frictions in the market creating non-fundamental comovement within firm returns. By comparing the comovement of firms with geographic industry concentration, this thesis attempts to test the argument that the local-bias of investors is due to an information advantage.

Looking at the United States between 1969 and 2000 and comparing firm annual stock returns with both industry concentration and a portfolio of firms based on their relative location and industry, I found evidence of strong local effects within the non-fundamental comovement of stock returns but could not attribute this to an information flow or advantage. These findings support the findings of Dougal, et al. (2015) where they found significant comovement of investment among local firms. The lack of significance in the effect of industry concentration on the comovement of firm annual stock returns disputes Coval and Moskowitz’s (2001) argument that geographical proximity plays a role in investor recognition. This study further suggests that investors are capable of incorporating knowledge on an industry level but are unable to do so on a firm level when compared with the findings of Garcia and Norli (2012).

This study contributes to the literature surrounding by analyzing local investor bias on an industry level. It incorporates the theories surrounding agglomeration economies and questions whether the spatial concentration of industries contributes to local investor bias. It adds to the literature surrounding non-fundamental comovement of firms and the effects of geography within finance. One shortcoming of this research is that the data does not extend to the present, making it difficult to project any findings into the present-day financial market, which itself has undergone a number of transformations due to the financial crisis and increase in communications technology which brought on high frequency trading. This research can be extended by including those dates, and examining whether trend found in the concentration of industries within the United States is related to financial shocks. Another limitation is that the data only takes into account public

firms which have different financing needs than private firms, and may be affected by geography more acutely than those public firms. This study is also limited in that there is no concrete theoretical framework under which to analyze the results.

Further research opportunities include analyzing whether the comovement of stock prices found in this study creates investment opportunities that are not yet exploited and whether there are other types of comovements in firm finance and investment decisions. Researchers can also examine whether this comovement is due to a herding effect or if investors are taking into account some other type of proximity effect. This study highlights a need for a concrete theoretical framework which describes local investor-bias, thus an interesting area of research would be to develop concrete measurements for the dimensions of proximity described by Boschma (2005) and look at how these play a role in home-bias.

REFERENCES

Addoum J., Kumar, A., Law, K. (2013). Geographic diffusion of information and stock returns. Working paper, University of Miami.

Ambos, TC., Ambos B. (2009). The impact of distance on knowledge transfer

effectiveness in multinational corporations. Journal of International Management

15. 1-14. Doi: 10.1016/j.intman.2008.02.002

Audretsch, DB., Feldman, MP. (1996). R&D Spillovers and the Geography of Innovation and Production. The American Economic Review 86(3). 630-640.

Barberis, N., Shleifer A., Wurgler, J. (2005). Comovement. Journal of Financial

Economics 75. 283-317. Doi: 10.1016/j.fineco.2004.04.003

Berry, T., Gamble, KJ. (2013). Informed local trading prior to earnings announcements.

Journal of Financial Markets 16(3). 505-525. Doi: 10.1016/j.finmar.2012.07.001

Beaudry, C., Schiffauerova, A. (2009). Who’s right, Marshall or Jacobs? The localization versus urbanization debate. Research Policy 38(2). 318-337. Doi:

10.1016/j.respol.2008.11.010

Boschma, R. (2005). Proximity and Innovation: A Critical Assessment. Regional Studies

39(1). 67-74. Doi: 10.1080/0034340052000320887

Chen, K., Chu, T., Billota, R. (2011). A spatial investigation of venture capital

investment in the US Biotechnology industry, 1995-2008. GeoJournal 76(3). 267-282. Doi: 10.1007/s10708-010-9344-z

Coval, JD., Moskowitz, T.J. (2001). The Geography of Investment: Informed Trading and Asset Prices. Journal of Political Economy, Vol 109(4). 811-841. Doi: 10.1086/322088

Combined Statistical Area. (2012). In United States Census Bureau online. Retrieved from https://www.census.gov/geo/reference/gtc/gtc_cbsa.html

Dougal, C., Parsons, CA., Titman, S., (2015). Urban Vibrancy and Corporate Growth.

Journal of Finance 70(1). 163-210. Doi: 10.1111/jofi. 12215

Ehrlich, I., Shin, JK., Yong Y. (2010). Human capital, endogenous information

acquisition, and home bias in financial markets. Working Paper, National Bureau

of Economic Research.

Francis, BB., Waisman, M., Hasan, I. (2007). Does Geography Matter to Bondholders? Working paper, FRB of Atlanta.

Garcia, D., Norli, Ø. (2012). Geographic dispersion and stock returns. Journal of

Financial Economics 106. 547-565. Doi: 10.1016/j.jfineco.2012.06.007

Garcia, D., Strobl, G. (2011). Relative Wealth Concerns and Complementarities in Information Acquisition. The Review of Financial Studies 24(1). 169-207. Graham, JR., Harvey, CR., Huang, H. (2009). Investor Confidence, Trading Frequency,

and Home Bias. Management Science 55(7). 1094-1106. Doi: 10.1287/mnsc.1090.1009

Howells, JRL. (2002). Tacit Knowledge, Innovation and Economic Geography. Urban

Studies 49. 871-884. Doi: 10.1080/0042098022012835

Ivković, Z., Weisbenner, S. (2005). Local does as local is: information content of the geography of individual investors' common stock investments. Journal of

Finance, 60. 267-306. Doi: 10.1111/j.1540-6261.2005.00730.x

Karlsson, A., Nordén, L. (2007). Home sweet home: Home bias and international diversification among individual investors. Journal of Banking & Finance 31(2). 317-333. Doi: 10.1016/j.jbankfin.2006.04.005

Krugman, P. (1991). Increasing returns and economic geography. Journal of Political

Economy 99. 483-499.

Lau, ST., Ng, L., Zhang, B. (2010). The world price of home-bias. Journal of Financial

Economics 97(2). 191-217. Doi: 10.1016/j.jfineco.2010.04.002

Location Quotient. (nd). In United States Bureau of Labor Statistics online. Retrieved from: http://www.bls.gov/help/def/lq.htm

OMB Bulletin No. 13-1. (2013). Retrieved online from

https://www.whitehouse.gov/sites/default/files/omb/bulletins/2013/b13-01.pdf Pirinsky, C., Wang, Q. (2006). Does Corporate Headquarters Location Matter for Stock

Returns? The Journal of Finance 61(4). 1991-2015. Doi: 10.1111/j.1540-6261.2006.00895.x

Geography 3. 429-452.

Wolf, HC. (2000) Intranational Home Bias in Trade. The Review of Economics and