Phil Davies
Tippie College of Business
The University of Iowa
W334 Pappajohn Business Building
Iowa City, IA 52242-1994
(319) 248-9458 phil-davies@uiowa.edu
Bernadette A. Minton
Fisher College of Business
The Ohio State University
834 Fisher Hall
2100 Neil Avenue
Columbus, OH 43210
(614) 688-3125 minton_15@fisher.osu.edu
Catherine Schrand
The Wharton School
University of Pennsylvania
1316 SH-DH
Philadelphia, PA 19104
(215) 898-6798 schrand@wharton.upenn.edu
January 2010
The authors thank The Global Association of Risk Professionals (GARP) for funding. We are especially grateful to Brian Bushee for providing his data on institutional ownership classifications. We thank Paul
Zarowin and seminar participants at New York University, the London Business School, INSEAD, the
University of Rochester, and Southern Methodist University for helpful suggestions on an earlier version of the paper, and David Barker, Matt Billett, Brian Bushee, Eric Lie, Anand Vijh, and seminar participants at the University of Iowa for comments on this version. Minton acknowledges financial support from the Dice Center for Research in Financial Economics.

We find robust evidence of investor clienteles for industry factor-price exposure: Investor interest, measured using share turnover and the number of institutions that hold a firm’s stock, is positively associated with stocks’ industry exposure, and institutional investors systematically overweight (underweight) high (low) industry exposure stocks in their portfolios. Clientele effects are most pronounced in industries in which return correlation with the aggregate market is low, where the benefits from learning about industry risk and from substituting investment in high-exposure stocks for investment in the industry assets are greatest. Clientele effects are strongest among small, transient, institutional investors. Contrary to traditional views of risk management, our findings suggest that market frictions may create incentives for some firms to not
hedge to attract liquidity.

In this paper, we document that investors display preferences for stocks with industry factor-price exposure (hereafter “industry exposure”), measured as the sensitivity of a firm’s stock return to industry returns after controlling for market returns. During the period from 1984 to 2006, a stock with high industry exposure experiences 50% higher share turnover, 13% more institutional investors, and 14% more mutual fund owners than a stock with low industry exposure after controlling for other established determinants of these proxies for investor interest. Moreover, across all industries, institutions systematically overweight high industry exposure stocks and underweight low industry exposure stocks in their portfolios.
The preference for industry exposure is inconsistent with asset allocation models based on perfect and complete capital markets. In such markets, investors will not display preferences for individual stocks with particular characteristics. However, when the assumptions of perfect and complete capital markets are relaxed, investors may rationally exhibit preferences for stocks with particular characteristics. For example, investors display preferences for domestic stocks
(French and Poterba, 1991), particular industries (Kacperczyk, Sialm, and Zheng, 2005), and dividend payout policies (Graham and Kumar, 2006). This paper is the first to propose and document investor clienteles associated with industry factor-price exposure.
Understanding investors’ attraction to industry exposure is important because it helps to explain the determinants of investors’ portfolio allocation decisions. Specifically, we shed light on the importance of two market frictions, information acquisition costs and incomplete markets.
In addition, understanding investor preferences for industry exposure has implications for how we think about firms’ risk management decisions. Our analysis suggests that investor preferences may create incentives for some firms to not hedge and instead to remain exposed to the underlying assets of the industry in order to obtain the liquidity benefits associated with
1

greater investor interest. This prediction is noteworthy because it is opposite to the predictions from existing optimal hedging theories which suggest that market imperfections increase incentives for firms to manage risk.
1 Our consideration of the effect of investor preferences on corporate hedging decisions is new, but the general notion that firms might cater to investor preferences when making corporate decisions is well established. Examples include catering dividend policy decisions to dividend clienteles and catering cross-listing decisions to investors in other countries because of home bias. By characterizing the extent and nature of investor preferences for industry exposure, this study is the necessary first step in understanding firms’ decisions to cater hedging decisions to such preferences.
2
One class of existing models that provides an explanation for investor clienteles for industry exposure assumes that information acquisition is costly, specifically in the sense that learning is constrained. Investors optimally acquire information (i.e., learn) about a single underlying risk factor, such as industry risk, and apply that knowledge to invest in stocks that are exposed to the risk factor (Van Nieuwerburgh and Veldkamp, 2010). Investors trade off the benefits of returns to private information against the cost of under-diversification in their portfolio. This investor learning
hypothesis has been used to predict investor clienteles in asset
1 See, for example, Myers (1977), Smith and Stulz (1985), Froot, Scharfstein, and Stein (1993), and DeMarzo and
Duffie (1995). A notable exception is Adam, Dasgupta, and Titman (2007) who develop a theoretical model in which the propensity to hedge is a function of the competitiveness of an industry. Ceteris paribus, there will be
2 more heterogeneity in hedging policies in competitive industries.
The dividend clientele/dividend catering literature is similarly segregated. Some studies model investor preferences for dividends, analogous to our examination of investor preferences for exposure, based on the percentage of institutional ownership of stocks (Del Guercio, 1996; Brav and Heaton, 1998) or based on institutions’ portfolio allocations (Strickland, 1996; Hotchkiss and Lawrence, 2007). Other studies model dividend payout policy as a function of dividend clienteles (Li and Lie, 2006; Baker and Wurgler, 2004a, 2004b) which would be analogous to our suggested second step of examining firms’ hedging decisions, relying on the evidence about investors’ preferences to specify the model. Grinstein and Michaely (2005) examine institutional ownership of stocks and dividend policy simultaneously.
2

characteristics, including home bias (Van Nieuwerburgh and Veldkamp, 2009), but it also predicts investor clienteles for industry exposure.
Another explanation for investor clienteles for industry exposure is that high-exposure stocks are a substitute for investment in an industry’s underlying assets. Most investors do not hold the “assets” assumed in classic portfolio theory. Investment in certain assets, commodities for example, is constrained by physical storage costs, asset indivisibility, or regulation. In the absence of such market frictions, investors would include these assets in their optimal portfolio.
The market, however, is effectively incomplete with respect to these assets, which leads to a second-best weighting in an investor’s portfolio. A stock that maintains exposure to the underlying asset of an industry can act as a substitute for investment in the asset and effectively complete the market. This asset substitution hypothesis is a previously unexplored explanation for investor clienteles.
The main prediction of both the asset substitution
hypothesis and the investor learning hypothesis is that investor preferences for industry exposure will be greatest for stocks in industries for which the returns have a low correlation with the returns on the aggregate stock market. Under the asset substitution
hypothesis, if industry returns are highly correlated with market returns, an investor can simply invest in the market portfolio to obtain exposure to the underlying asset. In addition, the portfolio diversification benefits of asset substitution are minimal if the asset is highly correlated with the aggregate market. If, however, industry returns are not highly correlated with returns on the market portfolio, the incentives to use stocks as a substitute for the underlying asset are greater. Similarly, under the investor learning
hypothesis investors obtain few benefits from learning about industries with returns that are highly correlated with the market. A signal about one industry that is highly correlated with the market
3

is likely to contain information about other industries that are also highly correlated with the market, thereby reducing the value of learning. In contrast, for industries with a low correlation with the market, the returns to learning are greater. Thus, in equilibrium more investors will learn about and invest in high industry exposure stocks in industries that are not highly correlated with the market.
Using the average incremental adjusted R 2 from adding industry factor returns to a market model estimated within each of the 30 Fama-French industries, we identify five industries whose returns have a significantly lower correlation with the aggregate stock market than the others. We refer to the five industries (mining, coal, utilities, tobacco, and oil and gas) as “high specificity” industries. The industries we identify as “low specificity” industries, which have returns that are highly correlated with the stock market, include, for example, consumer goods, wholesale, and services. Consistent with the asset substitution
hypothesis and the investor learning
hypothesis, investor interest in stocks with high industry exposure is greatest in high specificity industries. For example, a stock with high industry exposure has 30% more institutional investors than a stock with low industry exposure stock within high specificity industries, ceteris paribus
. In contrast, this difference is just 5% in low specificity industries.
Across all industries, small institutional investors display stronger preferences for industry exposure than large institutional investors, where institution size is based on the total market value of an institution’s holdings. Within high specificity industries, institutions of all sizes display significant preferences for industry exposure, although investor preferences for industry exposure remain more pronounced for small institutions. These results are consistent with the asset substitution
hypothesis if smaller institutional investors are more wealth constrained and hence more sensitive to asset indivisibility and storage costs. However, stronger
4

preferences by smaller institutions are also consistent with the investor learning hypothesis if institution size is positively correlated with information endowments or negatively correlated with information acquisition costs. Under either scenario, smaller institutional investors have greater incentives to reduce information costs by learning about an industry factor and investing in stocks with high industry exposure.
Investment style also influences institutional investor preferences for industry exposure.
We use the Bushee (2001) investment style classifications. Across all industries we find that transient investors, who hold small stakes in many firms and trade frequently on publicly available information but who do not generally acquire private information, have the strongest preferences for industry exposure. Quasi-indexers, who tend not to rely heavily on private information and adopt a passive monitoring style, display the next greatest preferences. Finally, dedicated owners, who tend to gather private information, and have large, long-term holdings concentrated in a small number of firms, show no preference for high industry exposure stocks.
The differences in investor preferences across investment style categories – transient investors, quasi-indexers, and dedicated owners – are most pronounced in high specificity industries. For example, transient investors overweight high industry exposure stocks by 8% and underweight low industry exposure stocks by 12%, while dedicated owners overweight high industry exposure stocks by 3% and underweight low industry exposure stocks by 3%. In low specificity industries, no investment style partitions exhibit preferences for industry exposure.
The results across institutions classified by investment style provide direct evidence that supports the investor learning
explanation of preferences for industry exposure. The asset substitution explanation does not make any predictions about preferences being associated with investment style.
5

An alternative explanation of investor preferences for industry exposure is that investors are attracted to transparency of information about a firm’s fundamental value, and financial statement transparency is positively correlated with industry exposure. We find, however, that firms with fewer and less complicated intra-firm relations, which would likely have a more transparent presentation of risk but also more industry exposure, have similar levels of investor interest as more complicated and presumably less transparent firms. Thus, greater financial statement transparency does not appear to explain the results.
We also explore the possibility that investor preferences vary across institution type as a function of fiduciary standards. If banks, which are subject to the “prudent man” rule, expect that the courts will view high industry exposure stocks as imprudent (Del Guercio, 1996), then banks will forgo the benefits of asset substitution or investor learning because of expected litigation costs. When we examine the number of institutions that hold a firm’s stock by fiduciary type, we find only weak evidence that banks’ preferences for industry exposure are different from those of insurance companies, investment advisors, and pension funds and endowments. Analysis of their portfolio holdings, however, suggests that fiduciary standards do mitigate banks’ preferences for industry exposure. In low specificity industries, in which the benefits of investing in high exposure stocks are minimal, banks significantly underweight high exposure stocks relative to other institutions, consistent with the prudent man constraint. In high-specificity industries, however, banks do not underweight high exposure stocks, suggesting that the benefits from investing in these high exposure stocks at least partially offset the expected litigation costs within these industries.
Our results suggest that market frictions may create incentives for some firms to not hedge and to remain exposed to the underlying assets of the industry in order to retain (or attract)
6

investors. This finding may help to explain why the evidence that hedging increases firm value is weak, despite theoretical predictions that hedging is a value-maximizing activity in the presence of market imperfections. Jin and Jorion (2006), for example, find no evidence that hedging increases firm value for oil and gas exploration companies. However, returns in the oil and gas industry have a low correlation with the market, and our results show that the benefits of not hedging are likely to be greatest in such industries. Reductions in industry exposures associated with hedging may lead to lower investor interest, reduced liquidity, and potentially a higher cost of capital, which may offset the benefits more traditionally associated with risk management. We leave this question to future research.
The paper is organized as follows. Section 2 discusses our hypotheses for why investors may be attracted to industry exposure. Section 3 defines our measures of investor interest, industry exposure, and control variables, and documents a positive relation between investor interest and industry exposure. Section 4 examines cross-sectional variation in clientele effects derived from our proposed explanations for investor preferences for industry exposure. Section
5 examines whether institutional investors systematically overweight high industry exposures stocks and underweight low industry exposure stocks, and Section 6 concludes.
2. Explanations for investor preferences for industry exposure
In this section, we outline three explanations for the prediction that investors will display preferences for stocks within an industry that have higher levels of industry exposure: the asset substitution
hypothesis (Section 2.1), the investor learning
hypothesis (Section 2.2), and the transparency
hypothesis (Section 2.3).
7

We propose that there will be investor clientele effects in stocks that have high industry exposure because investors use the stocks as substitutes for investment in fundamental assets.
Most investors do not hold the “assets” assumed in classic portfolio theory. Physical storage costs, regulation, and asset indivisibility combined with wealth constraints make markets in certain assets effectively incomplete for many investors.
3 Stocks can act as substitutes for assets that are otherwise unavailable or costly to attain, but only when the stock retains exposure to the assets underlying a firm’s operations. Thus, investors will be attracted to stocks with high levels of industry exposure when markets are incomplete.
Market frictions, such as storage costs, create inaccessibility to an underlying asset
, not an industry, but we predict an attraction to industry exposure because we expect fundamental assets to be relatively homogeneous within industries. For example, consider the oil industry.
The underlying asset, common to all firms in the oil industry is oil. Although other industries such as transportation are likely to have some exposure to oil, oil is not the primary driver of cash flow volatility in those industries. If an investor wants to acquire exposure to oil, then the most effective way to gain exposure is to purchase shares in an oil firm that maintains its exposure to oil prices.
In addition to predicting that investor interest in a stock is positively related to industry exposure, the asset substitution hypothesis also generates two cross-sectional predictions. First, investor preferences for stocks with industry exposure will be more pronounced when the diversification benefits from exposure to the fundamental assets of an industry are large. The
3 Entrepreneurs can and do create new assets that overcome these frictions. Securitized loans and real estate investment trusts (REITs) are classic examples of investment vehicles created to make access to an underlying asset, mortgage loans or real estate, feasible investments for a greater number of investors. See Stiglitz (1972), however, for a discussion of market frictions that prevent the creation of new assets that would fully complete markets.
8

diversification benefits are negatively related to the correlation between industry returns and aggregate market returns. If a particular industry has returns that are highly correlated with market returns, then diversification benefits from investing in firms with high industry exposure within that industry are small. Purchasing the market portfolio would be just as effective. In contrast, the diversification benefits are larger if the industry is not highly correlated with the market. In Section 4.1 we create a measure of industry specificity to capture the diversification benefits of asset substitution.
Second, the asset substitution
hypothesis predicts that investor preferences for industry exposure will be higher for more wealth constrained investors, given asset indivisibility. In our cross-sectional analysis, we examine institutional investor preferences based on institution size, as a proxy for wealth constraints.
A second hypothesis that predicts investor clientele effects for stocks with high industry exposure is based on theories that assume private information and costly information acquisition.
As one example of this class of theories, Van Nieuwerburgh and Veldkamp (2010) develop a model in which agents make both an information processing choice and a portfolio choice. An important constraint is that agents are assumed to have limited mental processing ability such that learning about one asset (i.e., acquiring information) reduces the agent’s ability to learn about other assets.
4 Assuming that asset payoffs are correlated with underlying risk factors, Van
Nieuwerburgh and Veldkamp (2010) show that it is optimal for an investor to learn about a
4 Sims (2003) and Peng (2004) also model limited mental processing ability, but only Van Nieuwerburgh and
Veldkamp (2010) focus on the interaction between asset portfolio and information choices. Van Nieuwerburgh and
Veldkamp (2009) use a similar modeling approach to explain the home bias.
9

single underlying risk factor that is costly to learn about and that has increasing returns to information processing, and to forgo first-best portfolio diversification.
5
Based on their model, Van Nieuwerburgh and Veldkamp (2010), propose business cycle risk, idiosyncratic risk, and industry risk as potential candidates for investor learning and investment specialization. We specifically predict an attraction to industry exposure because empirical evidence suggests that analysts perceive significant benefits associated with learning about industry risk. O’Brien (1990), for example, documents that over 97% of analysts in her sample specialize in one industry. She speculates that focusing on one industry allows analysts to better understand the production and cost functions underlying the industry, which leads to better forecasting for all firms in the industry. Dunn and Nathan (2005) confirm this speculation by showing that the more business segments an analyst follows, and the greater the diversification within an industry, the less accurate the earnings forecast. In general, this class of models will predict investor preferences for industry exposure if the net benefits of costly information acquisition are positively correlated with industry risk.
The investor learning
hypothesis also generates cross-sectional predictions. The first two predictions are similar to those for the asset substitution hypothesis. First, the investor learning hypothesis predicts that investors will display stronger preferences for industry exposure in highspecificity industries.
The potential benefits from learning about high specificity industries are greater, so more investors will choose to learn about high specificity industries. Second, under the assumption that information acquisition costs per dollar invested are higher for small institutions than large institutions, the investor learning
hypothesis predicts that small institutions
5 There is considerable empirical evidence supporting the notions developed in Van Nieuwerburgh and Veldkamp
(2010) that investors will hold a specialized (undiversified) portfolio and a well diversified portfolio. See, for example, Polkovnichenko (2005), Goetzmann and Kumar (2008), and Massa and Simonov (2003), Kacperczyk,
Sialm, and Zheng (2005), and Van Nieuwerburgh and Veldkamp (2009).
10

are likely to have greater benefits from learning about an industry factor and investing in stocks that remain exposed to that factor.
A prediction unique to the investor learning
hypothesis is that investor interest in industry exposure will be a function of investment style. Investors who, ex post, are observed to have different investment styles that are associated with different levels of information acquisition will exhibit variation in preferences for exposure. In our cross-sectional analysis, we examine the preferences of transient investors, quasi-indexers, and dedicated investors for industry exposure (see Section 4.2).
2.3 Transparency
A third hypothesis that we consider, but which ultimately is not supported by the data, is that investors are attracted to industry exposure because 1) information asymmetry leads investors to be attracted to financial statement transparency, and 2) financial statement transparency is positively correlated with industry exposure. The first condition is supported by models of trade with heterogeneously informed investors that suggest that information asymmetries introduce adverse selection into securities transactions. Greater financial statement transparency that reduces information asymmetry will reduce the adverse selection costs (e.g.,
Glosten and Milgrom, 1985; Kyle, 1985; Amihud and Mendelson, 1986; Admati and Pfleiderer,
1988; Diamond and Verrecchia, 1991). Empirical evidence, using a variety of proxies for transparency, supports this prediction.
6
6 For example, greater liquidity is associated with greater transparency measured by more informative financial reporting (e.g., Bartov and Bodnar, 1996; Leuz and Verrecchia, 2000; Boone and Raman, 2001; and Eleswarapu,
Thompson, and Venkataraman, 2004); tracking stocks (e.g., Billett and Mauer, 2000); direct marketing to capital markets via investor relations (Bushee and Miller, 2007), or even advertising (Brennan and Tamarowski, 2000).
Institutions, in particular, are attracted to transparency (Healey, Hutton, and Palepu, 1999; Bushee and Noe, 2000;
11

The second condition – a positive correlation between financial statement transparency and industry exposure – is more debatable. If firms with greater industry exposure have fewer operating segments and less complex intra-firm relationships, then the positive correlation with financial statement transparency is consistent with empirical research that defines transparency as information precision about firm value (Baldwin, 1984; Bushman, Chen, Engel, and Smith,
2004). The assumption that firms with fewer operating segments and less complex intra-firm relationships have greater industry exposure seems plausible given evidence in Lamont (1997) that more concentrated firms have greater cash flow risk associated with the underlying assets.
However, it remains an empirical issue since highly concentrated firms may engage in financial hedging to help reduce their underlying exposure.
3. Investor interest and industry exposure: empirical findings
Throughout the paper, we estimate variants of the following reduced form model of investor interest:
INVESTOR INTEREST iy
   
INDEXP iy
  k
 k
CONTROL kiy
  iy
where we use three proxies for INVESTOR INTEREST iy
for firm i in year y ; INDEXP iy
is firm-
(1) year industry exposure; and
CONTROL kiy
is a matrix of firm-year control variables. We estimate equation (1) annually and report the average of the coefficient estimates. Standard errors are clustered by industry in the annual regressions. All models throughout the paper include the control variables, but results for the control variables are tabulated only in the first set of reported results.
Aggarwal, Klapper, and Wysocki, 2005; Ferreira and Matos, 2008). Finally, Graham, Harvey and Rajgopal (2005) also provide survey evidence that managers believe that transparency is associated with improved liquidity.
12

The and investor learning
explanations for investor interest both imply that individual investors are most likely to be attracted to stocks with high industry exposure.
Individual investors are likely to be wealth constrained and unable to invest in underlying physical assets such as gold and oil. In addition, individual investors have less access to information and lower information processing capabilities than institutional investors.
Measuring individual investor interest, however, is challenging. We use share turnover as an indirect proxy for individual investor interest in equation (1). TURNOVER is the natural logarithm of average monthly turnover (volume divided by shares outstanding), computed for each firm i in each year y . While evidence suggests that share turnover captures individual investor interest in part (Barber and Odean, 2008; Hou, Peng, and Xiong, 2006; and Loh, 2008), it also captures disagreement among investors (Garfinkel, 2009), and it could be driven by a handful of large institutional investors trading in large quantities, or many investors trading in smaller quantities.
We also use institutional ownership and mutual fund ownership as measures of investor interest.
LNUMGR
is the natural log of 1 + the number of institutions that hold stock i
at the end of year y.
Data on annual institutional ownership are from the Thomson Financial 13-F database.
The Thomson database is based on the universe of 13-F filings without any selection or removal of firms. The only potential selection issues are that holdings under $20,000 are not required to be reported on a 13-F filing, and institutions that exercise investment discretion over less than
$100 million in equity are not required to file a form 13-F. Since all of our firms are publicly traded, we assume that the firm has zero institutional investors if it is not included in the reported
13

holdings of any institutions on the Thomson Financial database.
LNUMFUNDS is the natural log of 1 + the number of mutual funds that hold stock i
at the end of year y
. Data on annual fund ownership are from the Thomson Financial Mutual Fund database.
7 We eliminate mutual funds that have an investment objective code (IOC) equal to 1, 5 or 6, which represent International funds, Municipal Bond funds, and Bond and Preferred Stock funds, respectively. We also eliminate funds that have less than three annual observations in which the market value of assets at the beginning of the year is less than $1 million.
The institutional ownership and mutual fund ownership proxies represent more direct measures of investor interest than share turnover. However, the disadvantage of these measures is that preferences for industry exposure may be less pronounced for institutional investors than for individual investors. However, Bushee and Goodman (2007) find that, with the exception of institutions that hold large blocks and take big portfolio bets, there is little evidence of private information trading by institutions.
Following Jorion (1990) and Tufano (1998), for each firm ( i
), we calculate industry exposure at the end of each year ( y
) by estimating an extended market model using the past 60 months ( t
) of return data: r
   i i mkt r
  i ind r
,
 
(2)
7 Mutual funds (i.e., registered investment companies) are a class of investors in the Thomson Financial database of
13-F filers, but data on mutual fund holdings in the mutual fund database are different from the data for the
“Investment Companies” in the 13-F database. The 13-F database category of investment companies includes institutions that are not registered investment companies (i.e., not mutual funds) but that derive a significant portion of their business from the mutual fund business (determined by Thomson). In addition, holdings data on the
Thomson Mutual Fund database is compiled primarily from the funds’ required semi-annual reports to shareholders
(N-30D filings) rather than 13-F filings.
14

where r mkt
denotes the monthly return on the CRSP equally weighted market index, and r ind denotes the monthly return on the appropriate equally weighted industry portfolio.
8 We use the industry definitions provided on Kenneth French’s website to construct 30 industry portfolios.
9
For a firm-year observation to be included in the sample we require at least 24 monthly return observations to estimate the extended market model.
The estimated coefficient  ind (
INDEXP
) measures the sensitivity of stock i
’s return to a i one percent return on the underlying industry, after controlling for movements in the aggregate stock market that affect the returns on stock i independent of industry returns. Indicator variables identify high and low exposure firms (
BETAHIGH
and
BETALOW
, respectively).
BETAHIGH = 1 ( BETALOW = 1) if INDEXP iy
is above (below) the 70 th (30 th ) percentile exposure for its industry group for year y
. The ranking is done before requiring that the sample firms have non-missing Compustat data.
Table 1 reports descriptive statistics for the exposure measure by industry. The magnitudes of the industry exposure estimates vary considerably across the 29 industries. The average industry exposure ( INDEXP ) ranges from 0.51 for the Coal industry to in excess of 1.00 for the Retail and Business Equipment industries. Column 3 reports estimates of the average stock market  s across each industry.
MKTBETA2
is the estimate of  i mkt from the extended
(two-factor) market model specified in equation (2).
(Insert Table 1 here.)
9
8 We use equally weighted returns to ensure that the returns from a small number of large companies do not drive our measure of industry returns.
The 30 th industry includes firms that do not fall into industries 1 to 29; we discard the small number of firms assigned to the 30 th industry.
15

Two pieces of evidence indicate that INDEXP measured based on historical returns data is a reasonable proxy for expected future industry exposure, which is the relevant conceptual construct in the analysis. First, the industry exposure proxies based on historical data are fairly stable from year to year (untabulated). Of firms that are classified as high beta (medium beta)
{low beta} in year y - 1, approximately 76.4% (68.5%) {74.9%} are in the same category in year y . Of firms that are classified as high beta (medium beta) {low beta} in year y - 2, approximately
65.1% (58.6%) {63.6%} are in the same category in year y
.
Second, an out of sample experiment suggests that
INDEXP
predicts future exposure.
For each industry, at the start of year y + 1, we form an equally weighted portfolio that buys stocks classified as high exposure (
BETAHIGH
= 1) as of year-end y
and shorts low exposure stocks ( BETALOW = 1). The portfolio is rebalanced annually. We regress the returns from the portfolio strategy on the excess returns from the market portfolio and the relevant industry portfolio: r high exp
 r low exp m
 r mkt
 r f

  ind
 r ind
 r f

  .
(3)
If the
BETAHIGH
and
BETALOW
classifications capture meaningful differences in industry exposure, we expect to observe   0 . The final column of Table 1 reports that, for all ind industries other than Books and Food, there is robust evidence that  ind
 0 .
We draw control variables (
CONTROL
) from four papers that examine the determinants of institutional ownership: Del Guercio, 1996, Falkenstein, 1996, Gompers and Metrick, 2001, and Hong and Kacperczyk, 2009. Broadly speaking, these papers include various specifications
16

of proxies for the following constructs: Firm size, leverage, growth, share price, systematic risk, dividend yield, past returns, return volatility, and firm age. The control variables are particularly important when considering investor preferences for a particular stock characteristic. Investors often categorize stocks into broad classes such as large-capitalization, value, growth, and momentum, before deciding how to allocate funds across the classes (Barberis and Shleifer,
2003). By including proxies for size, value, and momentum as control variables we minimize the likelihood that our findings with respect to investor preferences for industry exposure are driven by a positive correlation between industry exposure and a common characteristic used by investors to assign stocks into a particular asset class. The Appendix provides a detailed description of our proxies for these constructs.
Table 2 reports means and medians for the control variables separately for the high exposure (
BETAHIGH
), low exposure (
BETALOW
), and the remaining medium exposure
(
BETAMED
) firms. There is a robust monotonic relation between industry exposure and turnover. High exposure firms also tend to be younger, have lower dividend yields and higher debt-to-equity ratios than low exposure firms. Market-to-book ratios, which prior literature has used as a proxy for growth opportunities, do not vary significantly across the exposure categories. Overall, Table 2 shows that there are substantial differences in firm characteristics across low, medium, and high industry exposure firms.
(Insert Table 2 here.)
3.4 Investor interest and industry exposure
Table 3 reports the results of regressions that measure the association between industry exposure and our three proxies for investor interest: share turnover (
TURNOVER
), the number of
17

institutional owners ( LNUMGR ), and the number of mutual fund owners ( LNUMFUNDS ). The results for the control variables are generally consistent with prior research. We find that investor interest is positively associated with firm size, market-to-book ratios, return volatility, stock market betas, past returns, inclusion in the S&P 500 index, and listing on NASDAQ; and negatively associated with the inverse of price, dividend yields, debt-to-equity ratios, and firm age.
(Insert Table 3 here.)
Overall, the results in Table 3 provide strong evidence of a positive relation between our proxies for investor interest and our proxy for industry exposure. As reported in Columns 1, 3, and 5, the coefficients on the continuous industry exposure proxy (
INDEXP
) are positive and significant, which is consistent with the prediction that higher industry exposure is associated with higher investor interest.
In Columns 2, 4, and 6, we use the indicator variables defined in the previous section,
BETAHIGH and BETALOW , to measure industry exposure. The results indicate that the positive associations between
INDEXP
and our proxies for investor interest come from both an attraction to high exposure stocks and an aversion to low exposure stocks. For example, c eteris paribus
, turnover in firms with high industry exposure is 33% higher than that of firms with medium levels of exposure, while turnover of low exposure firms is 19% lower (Column 2). Similarly, a change in a firm’s industry exposure from the 30 th percentile to the 70 th percentile is associated with a 13% increase in the number of institutions holding the stock (Column 4).
When the proxy for investor interest is the number of mutual funds that hold a firm’s stock (
LNUMFUNDS
), the results using indictor variables for industry exposure (Column 6) indicate only a significant aversion to low exposure stocks. In general, the relation between
18

industry exposure and mutual fund investors is weaker than the relation for institutional investors as a whole.
10
(Insert Table 4 here.)
We perform several additional analyses to test the robustness of the results reported in
Table 3. The results are presented in Table 4. In Panel A, we show that the results are robust across firm size quartiles (the smallest quartile, middle two quartiles and the largest quartile).
The primary model in Table 3 includes a continuous measure of firm size as a control variable, but the separate quartile regressions allow for non-linearities in the relation between firm size and investor attraction to industry exposure. Further, an analysis within size quartiles helps to address the possibility that our estimates of industry exposure for infrequently traded (i.e., low turnover) stocks are biased downward, which would induce a positive relation between industry exposures and turnover. If the results in Table 3 are driven by nonsynchronous trading, we would expect to observe a positive association between industry exposure and investor interest among the quartile of small firms where nonsynchronous trading is likely to be greatest, but no relation among large firms.
As shown in Table 4 Panel A, the investor interest measures with the exception of
LNUMFUNDS
are significantly positively associated with industry exposure across all size groups. For the smallest quartile of firms,
LNUMFUNDS
is not statistically related to measures of industry exposure. The coefficient estimates on BETAHIGH also are not statistically
10 Subsequent analysis uses the Thomson 13-F database classifications of investment advisors and registered investment companies as an alternative measure of fund ownership.
19

significant in any firm size group when investor interest is measured by LNUMGR and
LNUMFUNDS
.
11
In Panel B we find that the results in Table 3 are robust to alternative measures of industry exposure. The first alternative is a demeaned continuous measure of industry exposure.
The summary statistics in Table 1 indicate substantial differences in the average industry exposure coefficient across industries. The demeaned measure controls for the possibility that investors are attracted to particular industries rather than to within-industry exposure. The results using the demeaned measure are similar to the results for the continuous specification of
INDEXP in Table 3. The second alternative measure controls directly for the Fama-French factors, SMB and HML. When estimating industry exposure (equation (2)), we include not just market returns, but also returns on the Fama-French factors, SMB and HML as control variables.
Having obtained the estimates of industry exposure, we estimate a model identical to the specification reported in Table 3 but with the new estimates of industry exposure. The results are qualitatively similar to the results in Table 3. Finally, we also include the factor loadings on
SMB and HML when examining the determinants of investor interest. The inclusion of the factor loadings on SMB and HML does not affect investor preferences for industry exposure.
In addition, in untabulated regressions, we estimate the turnover regression separately for
NASDAQ and non-NASDAQ firms, recognizing that turnover is measured differently on the
11 To mitigate the concern that nonsynchronous trading causes a downward bias in the exposure estimates and hence a mechanical relation between exposure and turnover, we also estimate the turnover model for a sample that excludes observations with a stock price at the start of the year below $5. The results using either the continuous specification for industry exposure ( INDEXP ) or the indicator variables are similar to those reported in Table 3, both in terms of statistical and economic significance. In addition, if measurement error affects our results, we would expect to observe a negative relation between low industry exposure firms and turnover, but we would not necessarily expect to observe a positive relation between high industry exposure firms and turnover.
20

NASDAQ exchange. The results are similar. The results are also similar if stocks of financial firms, which represent approximately 20% of the sample observations, are excluded.
Throughout the remainder of our analysis we run all regressions with and without financial firms. The results are not sensitive to the exclusion of financial firms.
3.5 Transparency and industry exposure
One explanation for investor attraction to industry exposure is that investors are attracted to financial statement transparency and firms with more industry exposure are more transparent.
This transparency explanation is different from the asset substitution explanation and the investor learning explanation in that investors are not attracted to industry exposure per se
.
Rather, they are attracted to firms that have more transparent information about their asset exposures.
To rule out this explanation, we sort firms into sub-samples based on measures of line-ofbusiness diversification. We expect line-of-business diversification to be negatively correlated with financial statement transparency and industry exposure. If investors are attracted to industry exposure because it is negatively correlated with line-of-business diversification, and therefore positively correlated with financial statement transparency, we expect to observe no evidence that investors are attracted to industry exposure within a sample of firms with similar levels of diversification.
(Insert Table 5 here.)
We use two proxies for line-of-business diversification. The first proxy (
MULTISEG
) is an indicator variable equal to one if the firm is a multi-segment firm (regardless of the number of segments) and equal to zero if the firm is a single-segment firm. The second proxy is one minus
21

the firm’s revenue-based concentration ratio ( DIVERSE ), which is computed following
Comment and Jarrell (1995). The minimum value of
DIVERSE
is zero for a single segment firm and it approaches one as diversity increases (concentration decreases).
Table 5 Panel A presents the results from estimating equation (1) separately within portfolios of single segment firms (
MULTISEG
= 0) and multi-segment firms (
MULTISEG
= 1).
Panel B sorts the firms into three portfolios based on the rank of the continuous measure
DIVERSE
: the bottom quartile, the middle two quartiles, and the upper quartile. Firms are ranked within industry by year. We only report the results for the model specification that includes the indicator variables BETAHIGH and BETALOW to measure industry exposure.
12
Results using the continuous variable (
INDEXP
) yield similar inferences.
For all three proxies of investor interest ( TURNOVER , LNUMGR , and LNUMFUNDS ), the coefficients on
BETAHIGH
and
BETALOW
within all the diversification-sorted sub-samples are similar in magnitude and significance levels to those presented in Table 3. Across all proxies for investor interest investors display a strong aversion to stocks with low industry exposure.
Furthermore, using
TURNOVER
as a proxy for investor interest, there is robust evidence that investors display a preference for stocks with high industry exposure across all the sub-samples.
Thus the evidence in Table 5 provides little support for the hypothesis that investors exhibit preferences for industry exposure because industry exposure is correlated with financial statement transparency.
12 Segment reporting requirements changed significantly during the sample period. Between 1984 and 1997, a firm’s operating segments roughly correspond to its distinct product market industries. For the remainder of the sample period, firms defined their segments based on the internal management structure of the firm, which may or may not be by industry-level product line. For example, a firm could report segment data for its wholesale and retail operations (i.e., by customer type). Our diversification proxies are more likely to capture the negative correlation between diversification and industry exposure in the earlier period when segments were primarily defined at the industry level. In results not reported, we re-estimate the regression models in Table 5 for the period 1984 to 1997 and the period 1998 to 2006. The results in both periods are similar to those reported.
22

4. Understanding investor preferences for industry exposure
The asset substitution explanation for investor attraction to industry exposure predicts that investor demand to substitute investment in the stock of a firm that maintains industry exposure for investment in the underlying asset will be greatest in industries with returns that are not highly correlated with the aggregate market. The diversification benefits are greatest in these industries. The investor learning
explanation also predicts stronger preferences for industry exposure among such industries. For industries with returns that are highly correlated with aggregate market returns, a signal about one industry is likely to contain information about other industries that are also highly correlated with the aggregate market, thereby reducing the value of learning.
We refer to the degree to which industry returns are correlated with aggregate market returns as industry specificity. Industry specificity for industry j in year y is measured using parameter estimates from estimation of a standard market model and an extended market model for each firm within industry j.
The extended market model includes the equally-weighted market returns and the appropriate equally-weighted industry returns. The models are estimated using monthly return data over the period January of year y
– 4 to December of year y
. For each firm, i, in year y, we compute the difference between the adjusted R 2 values of the two models.
We use the average difference in adjusted R 2 values within an industry as our measure of
23

industry specificity.
13 The result is 23 annual observations for the 29 Fama-French industry groups, which is 667 industry-year observations of industry specificity (
SPECIFICITY
).
14
(Insert Figure 1 here.)
Figure 1 presents boxplots of the average industry specificity metric over our sample period from 1984 to 2006 for each of the 29 Fama-French industries. The greater is the average difference in adjusted R 2 s, the greater is the industry specificity. Five industries stand out as having substantially higher specificity measures: Utilities, Mining, Tobacco, Crude Oil and
Natural Gas, and Coal.
15 These five industries are designated “high-specificity” industries in the empirical analysis. For these high-specificity industries, the adjusted R 2 in the two factor market model is 17% higher in absolute terms, on average, than the adjusted R 2 in the standard market model. The industries with the lowest average specificity are: Wholesale, Electrical Equipment,
Services, Games, and Consumer Goods. These five industries are considered low-specificity industries in the empirical analysis. For the low-specificity industries, the adjusted R 2 increases by just 1.3% in absolute terms following the addition of industry returns to the market model.
16
Our classification of industries as high, medium, or low specificity is static over the sample period, but the boxplots in Figure 1 illustrate variation in specificity during the sample period for many industries. Much of the variation in
SPECIFITY
within each industry, however,
13 An alternative metric to the difference in adjusted R 2 is the F-statistic associated with the test of whether industry returns improve model specification. The correlation between these two variables is 0.98.
14 Our measure of industry specificity is similar to the country-level synchronicity measures developed by Morck,
Li, Yang, and Yeung (2004).
15 It is worth noting that these industries are not necessarily the industries with the highest industry factor price exposure (see Table 1).
16 Estimates of specificity based on the Fama-French 3-factor model are highly correlated (97%) and the estimates identify the same industries as high or low specificity industries.
24

is due to an increasing trend in SPECIFICITY from 1984 through 2006 across all industries. The relative ranking of the industries generally remains stable throughout the sample period.
17
Our measure of industry specificity has two important features. First, it is estimated using market data, which is available monthly. While the covariances of industry cash flows with aggregate market cash flows might be a more conceptually appropriate construct for industry specificity, we would have only 23 annual observations of financial statement data to estimate the covariances. Second, our measure applies equally well across industries. Measures based on an ex ante
identifiable market frictions, such as asset indivisibility or storage costs, would, by necessity, be industry-specific. It is worth noting, however, that the industries that our market-based measure identifies as “high” specificity are those in which ex ante
identifiable market frictions, that make direct investment costly, are likely to be high.
(Insert Table 6 here.)
Table 6 reports results of regressions that measure the association between industry exposure and investor interest as a function of specificity. Panel A reports results of estimation of equation (1) with three additional regressors:
SPECIFICITY
and the interaction of
SPECIFICTY
with the indicator variables,
BETALOW
and
BETAHIGH
. If industry specificity increases investor attraction to industry exposure, we expect a positive coefficient on the interaction between
BETAHIGH
and
SPECIFICITY
and a negative coefficient on the interaction between BETALOW and SPECIFICITY .
The results in Panel A show that the coefficient on the interaction term between
BETALOW
and
SPECIFICITY
is negative and significantly different from zero for all three
17 The only exception is the specificity measure for the Coal industry. We classify the Coal industry as a highspecificity industry in all years, but it has the most significant time-series variation, which is not surprising given the small number of firms in the industry (Table 1). All of the results are robust to exclusion of the Coal industry observations from the regressions.
25

measures of investor interest, indicating that investors, ceteris paribus , seek to avoid low industry exposure stocks particularly in higher specificity industries. Moreover, the coefficient on the interaction term between
BETAHIGH
and
SPECIFICITY
is positive and significantly different from zero for two measures of investor interest ( LNUMGR and LNUMFUNDS ), indicating that higher specificity is associated with stronger preferences for industry exposure.
Panel B reports results for high-, medium-, and low-specificity industries, respectively.
We expect to observe stronger investor preferences for industry exposure within the high specificity sub-sample. In Column 1, share turnover is used as a proxy for investor interest.
Within high-specificity industries, there is robust evidence that investors are attracted to high industry exposure stocks and display an aversion to low industry exposure stocks. In medium- and low-specificity industries, investors also display an attraction to high industry exposure and an aversion to low industry exposure, but the coefficients are attenuated relative to the highspecificity industries. The results using the number of institutional owners and mutual funds as proxies for investor interest follow a similar pattern. The differences between the coefficients on
BETAHIGH
and
BETALOW
increase with industry specificity.
4.2 Institution characteristics and investor clienteles in industry exposure
In this section, we investigate which types of institutional investors display preferences for industry exposure. First, we classify institutions annually into three groups based on the total equity value of their holdings: small, medium, and large. Small (large) institutions are defined as those in the lower (upper) quartile. Second, we use the classification system in the Thomson
Financial database and numerous studies of institutional ownership: (1) bank trusts, (2) insurance companies, (3) investment companies, (4) investment advisors, and (5) other. The “other”
26

category includes pension and endowment funds.
18 We aggregate the investment companies and investment advisors into one institution type.
19 Third, we use the Bushee (2001) classifications of institutional investors. Bushee (2001) identifies three types of institutional investors:
Dedicated owners, quasi-indexers, and transient investors.
20 Dedicated owners have large, longterm holdings concentrated in a small number of firms, and are more likely to gather private information about a firm and directly monitor its managers. Quasi-indexers tend not to rely heavily on private information and adopt a passive monitoring style. Transient investors hold small stakes in many firms and trade frequently on publicly available information, but they do not generally acquire private information.
For each partitioning of the institutional investors, we examine preferences for exposure across all industries (Table 7) and separately within high and low specificity industries (Table 8).
We report the results only for the model specification that includes the indicator variables
BETAHIGH
and
BETALOW
to measure industry exposure. Results using the continuous variable ( INDEXP ) yield similar inferences.
(Insert Table 7 here.)
4.2.1 Attraction to industry exposure by institution size
18 We thank Brian Bushee for providing us with his institution classifications during the sample period. There is a coding error in the Thomson Financial 13-F database. Thomson reports that partway through 1998, and in subsequent years, many banks (Type 1) and independent investment advisors (Type 4) are misclassified as other institutions (Type 5). Bushee’s database provides a consistent classification of the institutions on the Thomson
Financial database.
19 The Investment Company category (Type 3) in the Thomson 13-F database includes investment advisors that
Thomson determines derive a “significant” portion of their advisory services from the mutual fund business.
20 The Bushee (2001) annual institution classifications are based on k-means clustering of standardized factor scores, which are created on an institution-year basis using the weighted average of firm-specific characteristics of an institution’s portfolio holdings. Approximately 4% of institution-year observations are dedicated owners, 60% are quasi-indexers, and 36% are transient investors. Brian Bushee generously provided us with his classifications of investors.
27

Table 7 Panel A shows that small institutions display preferences for high industry exposure stocks and an aversion to low industry exposure stocks. Medium sized institutions also display an aversion to low industry exposure stocks, but the aversion is less pronounced in terms of economic magnitude. Medium sized institutions show no attraction to high exposure stocks.
Among large institutions there is no evidence that industry exposure, either high or low, matters.
Within high specificity industries (Table 8 Panel A), all institutions, regardless of size, display robust preferences for high exposure stocks.
Ceteris paribus
, a high industry exposure stock will experience between 6% and 8% more institutional investors of all sizes than stocks with average levels of industry exposure. Small and medium sized institutions display considerable aversion to low industry exposure stocks, but large institutions do not.
(Insert Table 8 here.)
Among low specificity industries there is little evidence that funds of any size display robust preferences for industry exposure. Small institutions avoid stocks with low industry exposure but are not attracted to stocks with high industry exposure, while medium and large institutions display a small but statistically significant aversion to high exposure stocks.
The stronger attraction to industry exposure among the smaller institutions is consistent with both asset substitution and investor learning
. Large institutions are likely to have fewer wealth constraints and be better able to invest directly in underlying assets to which they seek exposure. This option is not likely to be available to smaller wealth constrained institutions. At the same time, large institutions tend to take larger dollar positions than small institutions when investing in stocks, resulting in lower information acquisition costs per dollar invested relative to small institutions. As such, small institutions are likely to have greater benefits from learning about an industry factor and investing in stocks that remain exposed to that factor.
28

4.2.2 Attraction to industry exposure by institutional fiduciary standards
Table 7 Panel B shows that, except for banks, all types of institutions display a significant attraction to high industry exposure stocks and a significant aversion to low exposure stocks.
Table 8 shows that the attraction to high exposure stocks and the aversion to low exposure stocks for the non-bank institutions is greatest in the high specificity industries, consistent with the results throughout the paper.
The results for banks, however, are different. Looking across all industries, banks exhibit only a significant aversion to low-exposure stocks (Table 7). For low specificity industries, banks show an aversion to high exposure stocks. It is only for the high-specificity industries that banks exhibit a strong positive attraction to high exposure stocks.
The bank results are consistent with the different fiduciary standards of banks relative to other institutions. Del Guercio (1996) finds that banks, which are subject to the prudent man rule, tend to avoid stocks that they expect the courts to view as imprudent.
21 The expected litigation costs appear to be a constraint on banks’ attraction to industry exposure in low specificity industries where the benefits of investing in high industry exposure stocks are smallest. However, in high specificity industries banks appear to trade off the benefits of asset substitution
/ investor learning
against the expected litigation costs associated with an imprudent investment in a stock with high industry exposure. While the results support this interpretation based on direction, we caution that there are no statistically significant differences in preferences across the institutional types.
21 When the courts consider whether an investment is prudent or not, they tend to focus on the characteristics of assets in isolation, rather than considering the role of the asset in the bank’s overall portfolio.
29

4.2.3 Attraction to industry exposure by institution investment style
Table 7 Panel C reports cross-sectional analysis of the association between industry exposure and investor interest as a function of the institutional investor’s style, which is related to the institution’s information acquisition practices (Bushee, 2001). Dedicated owners show an aversion to low industry exposure stocks, but they do not exhibit a preference for high industry exposure stocks. In contrast, quasi-indexers and transient investors show an aversion to low exposure stocks and a preference for high exposure stocks.
Ceteris paribus
, the number of transient investors investing in a stock is 15% higher for stocks with high industry exposure relative to stocks with low industry exposure, while the difference for dedicated owners is less than 4%. The differences in preferences between transient investors and dedicated owners are statistically significant in 17 of the 23 sample years.
Table 8 Panel C shows that the patterns across investment style categories are most pronounced in high specificity industries. All institutions, regardless of their investment style display significant preferences for industry exposure, but the difference between preferences for high and low industry exposure stocks is greatest for transient investors followed by quasiindexers, and dedicated owners. A change in a firm’s industry exposure from the 30 th percentile to the 70 th percentile is associated with a 37% (15.70%) increase in the number of transient investors (dedicated owners) ceteris paribus
. The difference between the preferences of transient investors and dedicated owners is statistically significant in 12 out of 23 years.
These results are consistent with the investor learning explanation for investor attraction to industry exposure. Transient investors invest in a large number of stocks and seek to minimize the per-stock information acquisition costs. One way to minimize information acquisition costs is to obtain a signal about an underlying risk factor and apply this information
30

to all stocks that are exposed to the risk factor. In contrast, dedicated investors, who invest in a handful of firms, will choose to bear the costs of acquiring unique firm-specific information.
The asset substitution explanation, in contrast, does not make any predictions about a relation between investment styles and preferences for industry exposure.
The presentation of the averages of the annual coefficient estimates in Tables 7 and 8 obscures a distinct time trend in the significance of the BETAHIGH coefficient estimate for transient and quasi-indexers investors (not tabulated). The coefficient estimates for the transient and quasi-indexers investors are not statistically different from zero in the 1980’s and early
1990’s, but they are consistently significantly positive in the last ten years of the sample period.
While the time trend does not affect our statistical analysis, it does suggest that preferences for industry exposure have increased for transient and quasi-index investors in the last decade.
This increased preference for industry exposure by transient investors coincides with the introduction and rapid growth of Exchange Traded Funds (ETFs). In 1993 the first ETF was traded on the American Stock Exchange (AMEX). The number of funds grew from one in 1993 to 359 by the end of 2006, and the assets invested in ETFs grew from approximately $1 billion to
$422 billion.
22 ETFs are designed to track returns in particular sectors or markets, providing investors with access to sector or market exposure at a lower cost than more traditional mutual funds. A cost effective way for ETFs to track returns in a particular sector, such as oil and gas, is to invest in the stocks of firms that have high industry exposure. ETFs typically hold a large number of stocks, so they are likely to be classified as either transient or quasi-index investors.
The rapid growth in ETFs may, at least in part, explain the increased preferences for industry exposure among transient and quasi-index investors over the last decade.
22 Source: Investment Company Institute (ICI) Fact Book, 2008.
31

5. Analysis of portfolio diversification
In addition to predicting that investors will be attracted to high industry exposure stocks, the asset substitution and investor learning explanations for investor demand for industry exposure also predict that investor portfolios will be overly concentrated in (i.e., underdiversified) in high industry exposure stocks. We test this prediction by examining the reported holdings of institutional investors in the Thomson 13-F database.
The percent of the portfolio that institutional owner ( j
) holds in stocks of exposure type e in each of the 29 Fama-French industries ( i ) in year y ( PHELD ) is:
PHELD e ijy
 s
S 
 1
MV s ijy f
F 
 1
MV ijy f
(4) where
MV
is the market value of the stocks.
S
denotes the number of firms within industry i
in year y
with industry exposure level e
(high, moderate, or low) that institution j
invests in, and
F denotes the total number of firms within industry i in year y that institution j invests in. The weight is computed for each of the 13-F institution types for each year ( y
) between 1984 and
2006.
An institution’s percent held is compared to a benchmark weight that reflects no preference for exposure. The benchmark weight ( w
) is computed as the value weighted percentage of stocks classified as having high, moderate, or low exposure ( e ) in our sample for each industry ( i
) in each year ( y
): w e iy
 g
G 
 1
MV iy g
N  n
 1
MV iy n
(5)
32

where G denotes the number of firms in industry i in year y with industry exposure level e , and N denotes the total number of firms in industry i
in year y
.
The excess (XS) weight in each exposure category equals the percent held minus the benchmark:
XS BETA e ijy

PHELD e ijy
 w e iy
.
(6)
The excess weights across the three exposure categories for each institution-industry-year observation sum to zero. The null hypothesis that institutions do not overweight (underweight) high (low) industry exposure stocks implies that the excess invested in the high (low) exposure category is zero: 0.
ijy

(Insert Table 9 here.)
Table 9 reports the average excess weights. Panel A shows that both small and large institutions systematically overweight high industry exposure stocks. These findings are most pronounced within high specificity industries. For example, large institutions overweight high industry exposure stocks by 5% and underweight low industry exposure stocks by 8%. Among low specificity industries, there is no robust evidence that either small or large institutions have preferences for industry exposure.
Panel B of Table 9 reports that banks underweight high industry exposure stocks. The underweighting is significant in low specificity industries. This result is consistent with the prudent man constraint on portfolio allocation decisions of banks (Del Guercio, 1996). In high specificity industries, however, banks do not underweight high industry exposure stocks, consistent with the benefits of investing in these industries at least partially offsetting the
33

fiduciary costs. In contrast, investment advisors and pensions and endowments significantly overweight high industry exposure stocks in the high-specificity industries and underweight low exposure stocks.
Results across investment style are reported in Panel C. Transient investors display the greatest preferences for high industry exposure stocks and the greatest aversion to low industry exposure stocks. For example, in high specificity industries, transient investors overweight high exposure stocks by 8% while underweighting low exposure stocks by 12%. In contrast, there is no robust evidence that quasi-indexers overweight stocks with high industry exposure. The dedicated owners do tend to overweight high industry exposure stocks and underweight low industry exposure stocks, but the magnitudes are much smaller.
6. Conclusion
This paper shows investor clientele effects in industry factor price exposure and characterizes sources of investor preferences for industry exposure in stocks. Industry exposure is positively associated with share turnover, the number of institutional investors that hold the firm’s stock, and the number of mutual funds that hold the firm’s stock. The positive association between industry exposure and investor interest comes from both an attraction to high exposure stocks and an aversion to low exposure stocks. Furthermore, institutional investors systematically overweight high industry exposure stocks in their portfolios while underweighting low industry exposure stocks.
The attraction to high exposure stocks and the aversion to low exposure stocks is most pronounced among industries for which the correlation of industry returns with aggregate market returns is low: Mining, coal, utilities, tobacco, and oil and gas. The diversification benefits of
34

substituting stocks with high exposure to the underlying assets for direct investment in the underlying assets are greatest in these industries, as are the benefits from learning about an industry factor. Investor preferences for exposure are greatest for small institutions and institutions that follow a transient investment style, meaning they hold a large number of stocks and rely mostly on public information. Preferences for exposure are least prevalent among large institutions and dedicated institutional investors who invest in only a small number of stocks.
35

References
Adam, T., S. Dasgupta, and S. Titman, 2007, Financial constraints, competition and hedging in industry equilibrium,
Journal of Finance
62(5), 2445-2473.
Admati, A. and P. Pfleiderer, 1988, A theory of intraday patterns: Volume and price variability,
Review of Financial Studies
1
,
3-40.
Aggarwal, R., L. Klapper, and P. Wysocki, 2005, Portfolio preferences of foreign institutional investors, Journal of Banking and Finance 29, 2919-2946.
Amihud, Y. and H. Mendelson, 1986, Asset pricing and the bid-ask spread,
Journal of Financial
Economics
17, 223-249.
Baldwin, B. A., 1984, Segment earnings disclosure and the ability of security analysts to forecast earnings per share,
The Accounting Review
59, 376-389.
Barber, B. M. and T. Odean, 2008, All that glitters: The effect of attention on the buying behavior of individual and institutional investors,
Review of Financial Studies
21, 785-818.
Barberis, N. and A. Shleifer, 2003, Style investing, Journal of Financial Economics 68, 161-199.
Baker, M. and J. Wurgler, 2004, A catering theory of dividends,
Journal of Finance
59, 1125-
1165.
Baker, M. and J. Wurgler, 2004, Appearing and disappearing dividends: The link of catering incentives,
Journal of Financial Economics
73, 271-288.
Bartov, E. and G. M. Bodnar, 1996, Alternative accounting methods, information asymmetry and liquidity: Theory and evidence,
Accounting Review
71, 397-418.
Billett, M. T. and D. Mauer, 2000, Diversification and the value of internal capital markets: The case of tracking stock,
Journal of Banking and Finance
24, 1457-1490.
Boone, J. P. and K. K. Raman, 2001, Off-balance sheet R&D assets and market liquidity,
Journal of Accounting and Public Policy 20, 97-128.
Boyle P., R. Uppal, and T. Wang, 2003, Ambiguity aversion and the puzzle of own-company stock in pension plans, Working Paper, London Business School.
Brav, A. and J.B. Heaton, 1998, Did ERISA’s prudent man rule change the pricing of dividend omitting firms, Working paper, Duke University.
Brennan, M. and C. Tamarowski, 2000, Investor relations, liquidity, and stock prices,
Journal of
Applied Corporate Finance 12, 26-37.
Bushee, B. J., 2001, Do institutional investors prefer near-term earnings over long-run value?,
Contemporary Accounting Research
18, 207-46.
36

Bushee, B. J. and T. H. Goodman, 2007, Which institutional investors trade based on private information about earnings and returns?,
Journal of Accounting Research
45(2), 289-321.
Bushee, B. J. and G. Miller, 2007, Investor relations, firm visibility, and investor following,
Working Paper, University of Pennsylvania.
Bushee, B. J. and C. F. Noe, 2000, Corporate disclosure practices, institutional investors, and stock return volatility,
Journal of Accounting Research
38, 171-202.
Bushman, R., Q. Chen, E. Engel, and A. Smith, 2004, Financial accounting information, organizational complexity and corporate governance systems,
Journal of Accounting and
Economics
37, 167–201.
Chan, J., 2002, Persistence, uncertain liquidity, and expected returns, Working paper, UCLA
Anderson School.
Chordia, T., A. Subrahmanyam, and V. Anshuman, 2001, Trading activity and expected stock returns,
Journal of Financial Economics
59, 3-32.
Comment, R. and G. A. Jarrell, 1995, Corporate focus and stock returns, Journal of Financial
Economics
37, 67-87.
Daniel, K. and S. Titman, 1997, Evidence on the characteristics of cross-sectional variation in stock returns, Journal of Finance 52, 1-33.
Del Guercio, D., 1996, The distorting effect of the prudent man law of institutional equity investments,
Journal of Financial Economics
40, 31-62.
DeMarzo, P. and D. Duffie, 1995. Corporate incentives for hedging and hedge accounting,
The
Review of Financial Studies 8 , 743-771.
Diamond, D. W. and R. E. Verrecchia, 1991, Disclosure, liquidity, and the cost of capital,
Journal of Finance
46, 1325-1359.
Dunn, K. and S. Nathan, 2005, Analyst industry diversification and earnings forecast accuracy,
Journal of Investing
, Summer, 7-14.
Eleswarapu, V. R., R. W. Thompson, and K. Venkataraman, 2004, The impact of Regulation
Fair Disclosure: Trading costs and information asymmetry, Journal of Financial and
Quantitative Analysis
39, 209-225.
Falkenstein, E., 1996, Preferences for stock characteristics as revealed by mutual fund portfolio holdings,
Journal of Finance
51, 111-135.
Fama, E. and K. French, 1993, Common risk factors on the returns of stocks and bonds,
Journal of Financial Economics
33, 3-56.
Ferreira M. and P. Matos, 2008, The colors of investors’ money: The role of institutional investors around the world, Journal of Financial Economics 88, 499-533.
37

French, K. and J. Poterba, 1991, International diversification and international equity markets,
American Economic Review
81, 222-260.
Froot K. A., D. S. Scharfstein, and J. C. Stein, 1993, Risk management: Coordinating corporate investment and financing policies, Journal of Finance 48, 1629-1658.
Garfinkel, J. A., 2009, Measuring Investors' Opinion Divergence,
Journal of Accounting
Research
, 47, 1317-1348.
Glosten, L. R. and P. R. Milgrom, 1985, Bid, ask and transaction prices in a specialist market with heterogeneously informed traders,
Journal of Financial Economics
14, 71-100.
Goetzmann, W. N. and A. Kumar, 2008, Equity portfolio diversification,
Review of Finance
12,
433-463.
Gompers, P., and A. Metrick, 2001, Institutional investors and equity prices,
Quarterly Journal of Economics
118, 229-260.
Graham, J., C. Harvey, and S. Rajgopal, 2005, The economic implications of corporate financial reporting, Journal of Accounting and Economics 40, 3-73.
Graham, J. and A. Kumar, 2006, Do dividend clienteles exist? Evidence on dividend preferences of retail investors,
Journal of Finance
61, 1305-1361.
Grinstein, Y. and R. Michaely, 2005, Institutional holdings and payout policy, Journal of
Finance
60, 1389-1426.
Healy, P.M., A.P. Hutton, and K.G. Palepu, 1999, Stock performance and intermediation changes surrounding sustained increases in disclosure,
Contemporary Accounting Research
16, 485.
Hong, H. and M. Kacperczyk, 2009, The price of sin: The effects of social norms on markets,
Journal of Financial Economic s 93, 15-36.9.
Hotchkiss, E. and S. Lawrence, 2007, Empirical evidence on the existence of dividend clienteles,
Working paper, Boston College.
Hou, K., L. Peng, and W. Xiong, 2006, A tale of two anomalies: The implication of investor attention for pricing and earnings momentum, Working paper, The Ohio State University.
Investment Company Institute Fact Book, 2008. 48 th
Washington D.C.
edition, Investment Company Institute,
Jin, Y. and P. Jorion, 2006, Firm value and hedging: Evidence from U.S. oil and gas producers,
Journal of Finance
61, 893-919.
Jorion, P., 1990, The exchange-rate exposure of U.S. multinationals,
Journal of Business
63,
331-345.
Kacperczyk, M., C. Sialm, and L. Zheng, 2005, On the industry concentration of actively managed equity mutual funds, Journal of Finance 60, 1983-2011.
38

Kyle, A. S., 1985, Continuous auctions and insider trading, Econometrica 53, 1315-1335.
Lamont, O., 1997, Cash flow and investment: Evidence from internal capital markets,
Journal of
Finance
52, 83-109.
Li, W. and E. Lie, 2006, Dividend changes and catering incentives, Journal of Financial
Economics
80, 293-308.
Leuz, C. and R. E. Verrecchia, 2000, The economic consequences of increased disclosure,
Journal of Accounting Research 38, 91-124.
Loh, R., 2008, Investor inattention and the under-reaction to stock recommendations, Working paper, The Ohio State University.
Massa, M. and A. Simonov, 2006, Hedging, familiarity and portfolio choice,
The Review of
Financial Studies
19, 633-685.
Miller, M. H. and F. Modigliani, 1961, Dividend policy, growth, and the valuation of shares',
Journal of Business
34, 235-264.
Morck, R., K. Li, F. Yang, and B. Yeung, 2004, Firm-specific variation and openness in emerging markets,
Review of Economics and Statistics
86, 658-669.
Myers, S. C., 1977, Determinants of corporate borrowing,
Journal of Financial Economics 5,
147–175.
O'Brien, P., 1990, Forecast accuracy of individual analysts in nine industries
, Journal of
Accounting Research
28, 286-304.
Pastor, L. and R. Stambaugh, 2003, Liquidity risk and stock returns,
Journal of Political
Economy
111, 642-685.
Peng, L., 2004, Learning with information capacity constraints,
Journal of Financial and
Quantitative Analysis
40
,
307-329
.
Polkovnichenko, V., 2005, Household portfolio diversification: A case for rank dependent preferences,
Review of Financial Studies
18, 1467-1502.
Sims, C., 2003, Implications of rational inattention,
Journal of Monetary Economics
50 , 665-6 90.
Smith, C. and R. Stulz, 1985, The determinants of firms' hedging policies, The Journal of
Financial and Quantitative Analysis
28
,
391-405.
Stiglitz, J. E., 1972, On the optimality of the stock market allocation of investment,
Quarterly
Journal of Economics 86, 25-60.
Strickland, D., 1996, Determinants of institutional ownership: Implications for dividend clienteles, Working Paper.
39

Tufano, P., 1998, The determinants of stock price exposure: Financial engineering and the gold mining industry,
Journal of Finance
53, 1015-1052.
Van Nieuwerburgh, S. and L. Veldkamp, 2009, Information immobility and the home bias puzzle, Journal of Finance 64(3), 1187-1215
Van Nieuwerburgh, S. and L. Veldkamp, 2010, Information acquisition and portfolio underdiversification, forthcoming
Review of Economic Studies
.
40

Appendix: Summary of control variables
Summary of control variables used in the analysis. We draw the control variable constructs for the determinants of institutional ownership from four sources:
Del Guercio (1996), Falkenstein (1996), Gompers and Metrick (2001), and Hong and Kacperczyk (2009). All variables except firm age and the indicator variables for inclusion in the S&P 500 index and Nasdaq stocks are winsorized at the 1 st and 99 th percentiles.
Firm size
Market-to-book ratio
Share price
Systematic risk
LOGSIZE_MVE Natural log of the market value of equity (in $ thousands) at year end.
LOGMB
INVPRICE
MKTBETA2
Natural log of market value of equity divided by common book equity at year end.
Inverse of stock price at year end.
Market beta from estimation of the two-factor extended market model in equation (2).
Dividends
Leverage
Turnover
Returns
DIVYLD
DE RATIO
TURNOVER
AVGRET
Idiosyncratic return volatility RETVOL
Firm age FIRMAGE
Included in the S&P 500 index S&P500
Trades on NASDAQ exchange NASDAQ
Annual dividend yield.
Total long-term debt (including current portion) divided by total common equity at year end.
Natural log of average monthly turnover during the year.
Average monthly return during the year.
Standard deviation of daily firm returns during the year.
Natural log of the number of months from the CRSP start date to year end.
Indicator variable = 1 if the firm is in the S&P 500 index as of year end, and = 0 otherwise.
Indicator variable = 1 if the firm is traded on the NASDAQ exchange as of year end according to CRSP and = 0 if it is traded on the NYSE/AMEX.
41

Figure 1: Industry specificity
To measure industry specificity for industry j in year y , we estimate a standard market model and an extended market model for each firm i within industry j using monthly return data over the period January of year y – 4 to December of year y . The extended market model includes the appropriate equally-weighted industry returns and the equally-weighted market returns. For each firm i in year y, we compute the difference between the adjusted R
We use the average difference in adjusted R 2 specificity across our sample period of 1984 – 2006.
2 values of the two models.
values within an industry as our measure of industry specificity. The box plot below summarizes the industry
42

Table 1. Summary of exposure measures by industry
INDEXP is the mean of the firm-specific estimates of the monthly industry factor betas. MKTBETA2 is the mean of the firm-specific estimates of stock market betas from the two-factor market model. δ ind
is the estimated industry factor exposure in year y+1 for a portfolio that buys high industry exposure firms and shorts low industry exposure firms, where exposure is measured using historical data (see equation 3). δ ind
is hypothesized to be greater than zero if the BETAHIGH and BETALOW classifications capture meaningful differences in industry factor price exposure.
(*){**}[***] indicate statistical significance at the (10%) {5%} [1%] level.
Industry
18 Coal
N
105
INDEXP
0.5051
MKTBETA2
0.3570
δ ind
0.1478*
21 Telecommunications
2 Beer
1,743
264
0.5378
0.5540
0.3679
0.2606
0.1175**
0.5398***
4 Games
26 Wholesale
3 Smoke
2,049
3,564
102
1,304
0.5623
0.5648
0.6081
0.6361
0.4628
0.4066
0.3936***
0.1870*
0.1596 0.3732***
0.3794 0.2120*
6 Household
5 Books
10 Textiles
1 Food
25 Transportation
22 Services
15 Autos
28 Meals
1,812
1,204
648
1,938
2,189
8,523
1,247
1,473
0.6917
0.7056
0.7375
0.7677
0.7694
0.7907
0.7963
0.7983
0.2537 0.1851**
0.1650 -0.0804
0.2881 0.2124**
0.1646 0.1287
0.1434
0.2317
0.1939
0.1665
0.6189***
0.7209***
0.4708***
0.1421**
11 Construction
12 Steel
16 Carry
9 Chemicals
13 FabPr
7 Clothes
17 Mines
29 Financial
24 Paper
8 Health
19 Oil
20 Utilities
27 Retail
3,203
1,396
626
1,556
3,817
1,307
1,153
18,254
1,821
7,578
3,454
3,666
4,492
11,287
0.8001
0.8084
0.8314
0.8672
0.8807
0.8810
0.8916
0.9027
0.9507
0.9557
0.9582
0.9778
1.0135
1.0317
0.1842
0.1446
-0.0127
-0.0634
0.4825***
0.3523***
0.1056 0.4064***
0.1044 0.3294***
0.1593 0.5408***
0.1272 0.3090***
0.0693
0.0416
0.5389***
0.5770***
0.0491 0.6032***
-0.0076 0.7174***
0.0231
-0.0002
0.5498***
0.6856***
0.4196***
0.5985***
43

Table 2. Descriptive characteristics of sample firms
Means and medians (in parentheses) of industry factor price exposure ( INDEXP) , log monthly turnover
( TURNOVER) , and regression control variables across low exposure ( BETALOW ), medium exposure ( BETAMED ), and high exposure ( BETAHIGH ) firm-year observations. The control variables include the natural logarithm of the market value of equity, inverse price ratio, natural logarithm of the market-to-book ratio, dividend yield, debt-equity ratio, idiosyncratic return volatility, average monthly firm return, stock market beta, firm age, and indicator variables for stocks included in the S&P 500 and for NASDAQ listed stocks. The Appendix provides variable measurement details. A firm is considered high exposure (low exposure) if its industry factor price exposure is greater (less) than the 70 th (30 th ) percentile exposure, respectively. The percentiles are recalculated for each industry for each calendar year. The means and medians are measured for the sample across the years 1984 - 2006.
(*){**}[***] indicate statistical significance at the (10%) {5%} [1%] level. Significance levels are based on a Zstatistic associated with the annual t-statistics.
23
BETALOW
(n = 27,847)
BETAMED
(n = 36,084)
BETAHIGH
(n = 27,844)
HIGH vs
MED
HIGH vs.
LOW
MED vs.
LOW
Industry exposure
INDEXP
Log monthly turnover
TURNOVER
-0.2931
(-0.1029)
-3.1560
(-3.0972)
0.7959
(0.8116)
-2.9844
(-2.9584)
2.1168
(1.8283)
-2.6325
(-2.6146)
*** *** ***
*** *** ***
Firm size (Log MV of equity)
LOGSIZE_MVE
Inverse share price
INVPRICE
Log market-to-book ratio
LOGMB
Dividend yield
DIVYLD
Leverage (Debt/equity ratio)
DE RATIO
4.8457
(4.6323)
0.2363
(0.0759)
0.5828
(0.5275)
0.0172
(0.0000)
1.2623
(0.6333)
5.4391
(5.3593)
0.1744
(0.0580)
0.5848
(0.5329)
0.0173
(0.0055)
1.1713
(0.6311)
5.1677
(5.1532)
0.2652
(0.0792)
0.5912
(0.5187)
0.0116
(0.0000)
1.4832
(0.6874)
*** *** ***
*** *** ***
*** ***
*** *** ***
Daily return volatility
RETVOL
Systematic risk
MKTBETA2
Average monthly return
AVGRET
Log firm age in months
FIRMAGE
S&P 500 index inclusion dummy
S&P500
NASDAQ firm indicator
NASDAQ
0.0322
(0.0267)
1.2048
(0.9577)
0.0128
(0.0118)
4.8338
(4.8675)
0.0773
(0.0000)
0.5378
(1.0000)
0.0297
(0.0247)
0.1029
(0.0147)
0.0130
(0.0124)
4.9548
(4.9767)
0.1258
(0.0000)
0.5043
(1.0000)
0.0370
(0.0319)
-0.9896
(-0.7295)
0.0177
(0.0146)
4.7490
(4.7362)
0.0943
(0.0000)
0.5931
(1.0000)
*** *** ***
*** *** ***
** *
*** *** ***
*** ***
*** * **
23
Z
 t

N
 1
  
where t j
is the t -statistic for year j, N is the number of years, and t
and  ( t ) are the mean and standard deviation, respectively, of the
N
realizations of t j
.
Z
has a t
distribution with
N
− 1 degrees of freedom.
44

Table 3. Determinants of investor interest
Models of industry factor price exposure as a determinant of investor interest in a firm’s stock. Proxies for investor interest include the natural logarithm of average monthly turnover (
TURNOVER
), institutional investor interest
( LNUMGR ), and mutual fund interest ( LNUMFUNDS ). Factor price exposure is measured by the continuous variable INDEXP and by indicator variables that equal 1 if a firm’s industry factor price exposure is greater (less) than the 70 th (30 th ) percentile exposure ( BETAHIGH and BETALOW ). The percentiles are recalculated for each industry for each calendar year. Control variables measured at or for the year ended t-1 include: the natural logarithm of the market value of equity (LOGSIZE_MVE), the inverse price ratio (INVPRICE), the natural logarithm of the market-to-book ratio (LOGMB), dividend yield (DIVYLD), debt equity ratio (DE RATIO), idiosyncratic return volatility (RETVOL), average monthly firm returns (AVGRET), and turnover (TURNOVER, except in TURNOVER regressions). Control variables measured at or for the year ended t include: stock market betas (MKTBETA2), firm age (FIRMAGE), and indicator variables for S&P 500 stocks (S&P500) and NASDAQ
(NASDAQ) listed stocks. The models are estimated annually from 1984 through 2006. The coefficient estimates, adjusted R 2 s, and number of observations (N) are the averages of the annual estimates. (*){**}[***] indicate statistical significance at the (10%) {5%} [1%] level. Significance levels are based on a Z-statistic associated with the annual t-statistics (see Table 2). Parenthetical amounts represent the number of annual test statistics that are significant at the 10% level in the 23 annual regressions .
LOGMB
DIVYLD
TURNOVER LNUMGR LNUMFUNDS
Intercept -4.6089*** -4.2382*** 1.2015*** 1.3267*** 0.5761*** 0.6912***
LOGSIZE_MVE 0.1879*** 0.1987*** 0.4693*** 0.4689*** 0.4810*** 0.4806***
INVPRICE -0.2937***
0.0791*** 0.0736*** -0.0300** -0.0313** -0.0623** -0.0635**
-3.8037*** -3.9146*** -1.5618*** -1.4895*** -5.8242*** -5.7147***
-0.0210*** -0.0151*** -0.0078*** -0.0066*** -0.0091*** -0.0081***
TURNOVER
RETVOL 13.4460***
MKTBETA2
AVGRET
FIRMAGE
0.2387*** 0.2442*** 0.2470*** 0.2513***
-6.9094*** -6.5498*** -10.0423*** -9.6792***
0.3729*** 0.0882*** 0.0472*** 0.0013
1.2600*** 1.5147*** 0.0259 0.0457
0.0368** -0.0024
0.9553*** 0.9478***
-0.0643*** -0.0787*** 0.1549*** 0.1526*** 0.1372*** 0.1357***
S&P500
NASDAQ
INDEXP
0.1544*** 0.1151*** 0.3500*** 0.3424*** 0.6624*** 0.6559***
0.2280*** 0.2135*** -0.1452*** -0.1486*** -0.2218*** -0.2246***
0.4966*** 0.0962*** 0.0917***
BETALOW
BETAHIGH
-0.1869***
0.3273***
-0.0994***
0.0346***
-0.1073***
0.0328
Difference
# of annual diffs sig
0.5142
(23/23)
0.1340
(19/23)
0.1401
(14/23)
Average N 2,441 2,441 2,441 2,441 2,441 2,441
Average annual Adj R 2 29.62% 26.69% 83.16% 83.14% 75.35% 75.33%
45

Table 4. Robustness tests for Table 3
This table reports additional analyses to validate the robustness of the main results reported in Table 3. Panel A reports the results across size quartiles of the sample firms. The quartiles are recalculated for each calendar year.
Panel B reports the results using alternative measures of industry exposure. The first alternative is a standardized
(demeaned) measure of the continuous variable INDEXP within industry/year. The second alternative measure is the estimate of industry exposure using an extended 3-factor Fama-French model ( INDEXP3 ). We report results of estimation of identical models to those specified in Table 3, but using the INDEXP3 estimates and indicator variables based on the extended 3-factor model estimates ( BETAHIGH3 and BETALOW3 ) . We also report results using these alternative measures and augmenting the models in Table 3 to include the coefficient estimates on SMB and HML. All models include the control variables defined in Table 3; coefficient estimates for the control variables are not tabulated. The models are estimated annually from 1984 through 2006. (*){**}[***] indicate statistical significance at the (10%) {5%} [1%] level. Significance levels are based on a Z-statistic associated with the annual t-statistics (see Table 2).
Smallest quartile
INDEXP
BETALOW
BETAHI
Middle two quartiles
INDEXP
BETALOW
BETAHI
Largest quartile
INDEXP
BETALOW
BETAHIGH
0.3921***
Panel A: By sample firm size quartiles
-0.1575***
0.3202***
0.0515***
-0.0598***
0.0183
0.0269
-0.0436
0.0034
0.4965***
-0.1760***
0.3269***
0.0722***
-0.0637***
0.0156
0.0744***
-0.0816***
0.0261
0.3254*** 0.0963***
-0.0802***
0.1421***
0.1508*** 0.0128 0.0072
INDEXP
Panel B: Alternate measures of industry exposure
Demeaned continuous measures
0.3089*** 0.0787*** 0.0594***
Estimates from extended 3-factor model
INDEXP3
BETALOW3
BETAHIGH3
0.3502***
-0.1523***
0.0564***
-0.0845***
0.0526***
-0.0895***
0.2643*** 0.0252** 0.0119
Estimates from extended 3-factor model with factor betas included in the regressions
INDEXP3 0.4612***
BETALOW3
BETAHIGH3
-0.1619***
0.2676***
0.0886***
-0.0920***
0.0320**
0.0807***
-0.0958***
0.0175
46

Table 5. Determinants of investor interest for diversification-sorted portfolios
Average annual coefficient estimates on the proxies for industry factor price exposure from models of the determinants of investor interest estimated as a function of diversification. Proxies for investor interest include the natural logarithm of average monthly turnover ( TURNOVER ), institutional investor interest ( LNUMGR ), and mutual fund investor interest (
LNUMFUNDS
). Industry factor price exposure is measured by indicator variables that equal
1 if a firm’s industry factor price exposure is greater (less) than the 70 th (30 th ) percentile exposure ( BETAHIGH and
BETALOW ). The percentiles are recalculated for each industry for each calendar year. Panel A reports results for single segment firms vs. multi-segment firms. Panel B reports results for firms in the lower quartile, middle two quartiles, and upper quartile of the variable DIVERSE , which is 1 - a revenue-based concentration ratio such that lower values represent greater concentration. Firms are ranked within industry by year. All models include the control variables defined in Table 3; the coefficient estimates for the control variables are not tabulated. The models are estimated annually from 1984 through 2006. (*){**}[***] indicate statistical significance at the (10%) {5%}
[1%] level. Significance levels are based on a Z-statistic associated with the annual t-statistics (see Table 2).
Parenthetical amounts represent the number of annual test statistics that are significant at the 10% level in the 23 annual regressions .
Panel A: Single Segment firms vs. multi-segment firms
Single segment firms
Intercept -4.1859*** 0.4683***
BETALOW -0.1984*** -0.0868*** -0.0920***
BETAHIGH
Difference 0.5273 0.0909 0.0874
Multi-segment firms
Intercept -4.1259*** 0.9248***
BETALOW -0.2199*** -0.1011*** -0.1106***
BETAHIGH
Difference 0.5360 0.1209 0.1147
Panel B: By quartiles of DIVERSE
Bottom quartile DIVERSE
Intercept -4.1824*** 0.4536***
-0.1976*** -0.0867*** -0.0926*** BETALOW
BETAHIGH
Difference 0.5260 0.0915 0.0888
Middle quartiles DIVERSE
Intercept -4.3384*** 1.0264***
BETALOW
Difference
BETAHIGH
-0.2469*** -0.0833*** -0.0977***
0.5781 0.0950 0.1036
Upper quartile DIVERSE
Intercept -3.9604*** 0.8957***
-0.1871*** -0.1155*** -0.1182*** BETALOW
BETAHIGH
Difference 0.5124 0.1419 0.1275
47

Table 6. Industry specificity and investor attraction to exposure
Average annual coefficient estimates on proxies for industry factor price exposure from models of the determinants of investor interest. Proxies for investor interest include the natural logarithm of average monthly turnover
( TURNOVER ), institutional investor interest ( LNUMGR ), and mutual fund interest ( LNUMFUNDS ). Industry factor price exposure is measured by indicator variables that equal 1 if a firm’s industry factor price exposure is greater
(less) than the 70 th (30 th ) percentile exposure ( BETAHIGH and BETALOW ). Panel A reports results for models that include a continuous measure of industry specificity (
SPECIFICITY
) and interaction terms of
SPECIFICITY
with
BETAHIGH and BETALOW . Panel B reports coefficient estimates on BETAHIGH and BETALOW for firms that operate in high-specificity, medium-specificity, and low-specificity industries. All models include the control variables defined in Table 3; the coefficient estimates for the control variables are not tabulated. The models are estimated annually from 1984 through 2006. The coefficient estimates presented are the averages of the annual estimates. (*){**}[***] indicate statistical significance at the (10%) {5%} [1%] level. Significance levels are based on a Z-statistic associated with the annual t-statistics (see Table 2). Standard errors are clustered by industry within each annual regression.
Panel A: Interaction variables for SPECIFICITY
-4.2687***
SPECIFICITY 0.2461***
BETALOW
BETAHIGH*SPECIFICITY
-0.5082**
BETAHIGH
-0.2461 0.3593***
0.0025
0.3445***
Average annual N
Average annual adjusted R 2
2,441 2,441 2,158
26.85% 83.20% 74.91%
High Specificity Industries
Panel B: By quartiles of SPECIFICITY
-4.2322***
BETALOW
BETAHIGH
Difference
Average annual N
Average annual adjusted R 2
0.6335 0.2726 0.2907
253 253 253
27.07% 83.15% 75.20%
Medium Specificity Industries
Intercept -4.2226*** 0.5363***
Average annual N
Average annual adjusted R 2
Low Specificity Industries
-0.1679***
BETAHIGH
Difference 0.4768 0.1185
1,613 1,613 1,613
28.09% 83.72% 76.12%
0.1174
-4.3897***
BETALOW
BETAHIGH
Difference
Average annual N
Average annual adjusted R 2
0.5593 0.0262 0.0573
575 575 575
26.26% 81.73% 74.09%
48

Table 7. Institutional investor attraction to exposure in the full sample of industries by investor type
Average annual coefficient estimates on the industry factor price exposure proxies from multivariate models of the determinants of the log of 1 + the number of institutions of a given type that hold a firm’s stock. Industry factor price exposure is measured by indicator variables that equal 1 if a firm’s industry factor price exposure is greater
(less) than the 70 th (30 th ) percentile exposure ( BETAHIGH and BETALOW ). The percentiles are recalculated for each industry for each calendar year. All models include the control variables defined in Table 3; the coefficient estimates for the control variables are not tabulated. The models are estimated annually from 1984 through 2006.
(*){**}[***] indicate statistical significance at the (10%) {5%} [1%] level. Significance levels are based on a Zstatistic associated with the annual t-statistics (see Table 2). Parenthetical amounts represent the number of annual test statistics that are significant at the 10% level in the 23 annual regressions .
Panel A: Institutional investors by size (market cap)
Small Medium Large
Difference 0.1461 0.0712 0.0041
# of annual differences that are significant (20/23) (14 (4/23)
Test vs. Small
Test vs. Medium
(11/23) (21/23)
(14/23)
BETALOW
Panel B: Institutional investors by fiduciary standards
Insurance
Companies
Investment Pensions/
Banks Advisors Endowments
-0.0731*** -0.0722*** -0.0880*** -0.0745***
Difference 0.0889 0.1184 0.1211 0.1149
# of annual differences that are significant
Test vs. Banks (4/23) (10/23) (8/23)
BETALOW
Panel C: Institutional investors by investment style
Dedicated
Owners
Quasiindexers
Transient
Investors
-0.0425*** -0.0922*** -0.0907***
Difference 0.0367 0.1193 0.1521
# of annual differences that are significant
Test vs. Dedicated Owners (14/23) (17/23)
Test vs. Quasi-indexers (7/23)
49

Table 8. Institutional investor attraction to exposure in high and low specificity industries by investor type
Average annual coefficient estimates for industry factor price exposure proxies from multivariate models of the determinants of the log of 1 + the number of institutions of a given type that hold a firm’s stock estimated separately for high and low specificity industries. Factor price exposure is measured by indicator variables that equal 1 if a firm’s industry factor price exposure is greater (less) than the 70 th (30 th ) percentile exposure ( BETAHIGH and BETALOW ). The percentiles are recalculated for each industry for each calendar year. All models include the control variables defined in Table 3; the coefficient estimates for the control variables are not tabulated. The models are estimated annually from 1984 through 2006. The coefficient estimates presented are the averages of the annual estimates. (*){**}[***] indicate statistical significance at the (10%) {5%} [1%] level. Significance levels are based on a Z-statistic associated with the annual t-statistics (see Table 2). Parenthetical amounts represent the number of annual test statistics that are significant at the 10% level in the 23 annual regressions .
BETALOW
BETAHIGH
Difference
# of annual sig differences
Test vs. Small
Test vs. Medium
HIGH SPECIFICTY INDUSTRIES LOW SPECIFICTY INDUSTRIES
Panel A: Institutional investors by fund size (market cap)
Small Medium Large Small Medium Large
-0.2062*** -0.1609*** -0.0648 -0.0515*** -0.0046 0.0245**
0.0784*** 0.0833*** 0.0634***
0.2846 0.2442 0.1282
(18/23) (17/23)
(4/23)
(9/23)
(11/23)
(7/23)
0.0265 -0.0440 -0.0656
(5/23) (4/23)
(8/23)
(7/23)
(8/23)
(1/23)
BETALOW
BETAHIGH
Difference
# of annual sig differences
Test
Panel B: Institutional investors by fiduciary standards
Insurance Investment Pensions/ Insurance Investment Pensions/
Banks Companies Advisors Endowments Banks Companies Advisors Endowments
-0.1228*** -0.2032*** -0.1779*** -0.1899*** -0.0295* -0.0327*** -0.0351** -0.0255*
0.0771*** 0.0928*** 0.0699*** 0.0942*** -0.0408*** -0.0046 -0.0254 -0.0328*
0.1999 0.2960 0.2478 0.2841 -0.0113 0.0281 0.0097 -0.0073
(16/23) (18/23) (14/23) (17/23) (5/23) (5/23) (3/23) (3/23)
(6/23) (7/23) (8/23) (3/23) (4/23) (1/23)
Panel C: Institutional investors by investment style
Dedicated QuasiTransient Transient
Owners indexers Investors Owners indexers Investors
BETALOW -0.1059*** -0.1429*** -0.2608***
BETAHIGH 0.0511**
-0.0099 -0.0466***
Difference
# of annual sig differences
Test vs. Dedicated Owners vs.
0.1570 0.2125 0.3746
(14/23) (17/23)
(3/23)
(21/23)
(12/23)
(13/23)
-0.0262 0.0183 0.0207
(3/23) (4/23)
(2/23)
(4/23)
(4/23)
(4/23)
50

Table 9. Institutional investor portfolio composition and industry exposure
Average annual excess weight placed on high and low exposure stocks in institutional investor portfolios over the period from 1984 to 2006. Each year we calculate the average weight invested in high and low industry exposure stocks. XS-BETAHIGH and XS-BETALOW represent the excess weight calculated relative to the value-weighted percentage of stocks with high and low industry exposure, respectively, within each industry each year. The null hypotheses of no investor preference for industry exposure is that XS-BETAHIGH = 0. (*){**}[***] indicate statistical significance at the (10%) {5%} [1%] level. Significance levels are based on a Z-statistic associated with the annual t-statistics (see Table 2).
All industries High specificity industries Low specificity industries
XS-BETAHIGH XS-BETALOW XS-BETAHIGH XS-BETALOW XS-BETAHIGH XS-BETALOW
0.0107*** -0.0165*** 0.0373** -0.0534** 0.0015 -0.0174***
0.0127*** -0.0056 0.0485** -0.0835*** -0.0008 -0.0076
Panel B: Fiduciary Standard
Banks -0.0273***
Insurance
0.0108
Companies 0.0060* -0.0145*** 0.0191 -0.0526*** -0.0076** -0.0099*
Pensions/Endowments
Panel C: Investment Style
0.0174*** -0.0124*** 0.0478** -0.0759*** 0.0098*** -0.0151***
0.0048 -0.0128*** 0.0498** -0.0790*** -0.0093** -0.0082**
Dedicated Owners
0.0178*** -0.0027 0.0277** -0.0266** 0.0241*** -0.0205*
Quasi-indexers -0.0092***
Transient Investors
-0.0061*
0.0425*** -0.0301*** 0.0754*** -0.1216*** 0.0230*** -0.0239***
51