You are on page 1of 26

THE ACCOUNTING REVIEW American Accounting Association

Vol. 94, No. 3 DOI: 10.2308/accr-52202


May 2019
pp. 87–112

Financial Reporting Quality, Investment Horizon, and


Institutional Investor Trading Strategies
Brian J. Bushee
University of Pennsylvania

Theodore H. Goodman
Purdue University

Shyam V. Sunder
The University of Arizona
ABSTRACT: This paper provides evidence that financial reporting quality (FRQ) influences the holding costs of
trading strategies. While prior research has focused on the benefits of investment strategies based on poor FRQ
(i.e., larger returns due to a greater amount of private information), we examine whether poor FRQ imposes greater
holding costs on certain trading strategies. We show that poor FRQ motivates sophisticated investors with short-term
horizons to tilt their portfolios away from value stocks, whose returns are contingent on investors revising their beliefs
about firm fundamental value, and toward past winner stocks, whose future returns are realized more quickly. Poor
FRQ also increases the length of time that institutions maintain large positions in value stocks. Our results imply that
mis-valuations can be persistent when arbitrageurs perceive high holding costs from poor financial quality, even
when they can see through the opaque financial disclosures.
Keywords: institutional investors; financial reporting quality; trading strategies; arbitrage costs.

I. INTRODUCTION

T
his paper examines the potential negative consequences of poor financial reporting quality (FRQ) for the trading
strategies pursued by sophisticated investors. Prior work finds that poor FRQ enhances the returns from trading
strategies that rely on information discovery by increasing the value of private information. However, poor FRQ could
also increase the arbitrage risk of trading strategies by increasing ‘‘holding costs,’’ which are the costs associated with having to
hold a position in a mispriced security for a long duration until the mispricing corrects. If the price correction for a given trading
strategy relies on investors learning about an asset’s fundamental value from the public disclosure of financial statement
information, such as earnings, then poor FRQ could increase these holding costs. Poor FRQ results in a noisier earnings signal
that leads uninformed investors to require more quarters of earnings realizations before revising their expectations of
fundamental values, delaying when these values are impounded into prices. If the expected price correction period is long
enough, then the expected holding costs could outweigh the potential return benefits and deter an investor from pursuing a
strategy. Thus, we predict that holding costs related to poor FRQ reduce sophisticated investors’ willingness to implement
trading strategies that rely on investors using earnings to revise their beliefs about firm fundamental values for price correction.
Holding costs are particularly important for short-term-focused investors. Every quarter that a short-term investor has to
hold a position to wait for a price correction dampens the investor’s short-term excess portfolio returns and imposes opportunity
costs by preventing investments in other potential profit opportunities. Bushee (2001) argues that institutions may adopt a short-
term focus in response to competitive pressures to maximize near-term returns from their portfolio in order to attract new

We thank Mohan Venkatachalam (editor), two anonymous referees, John Campbell, Joseph Gerakos (FARS discussant), Mindy Kim, Santhosh
Ramalingegowda, Chengdong Yin, and workshop participants at The University of Georgia and the 2015 FARS Midyear Meeting for helpful comments
and suggestions. We also thank Wei Jiang for sharing her hedge fund data. We are grateful for the funding of this research by the Wharton School of the
University of Pennsylvania, Purdue University, and The University of Arizona.
Editor’s note: Accepted by Mohan Venkatachalam, under the Senior Editorship of Mark L. DeFond.
Submitted: October 2015
Accepted: July 2018
Published Online: July 2018
87
88 Bushee, Goodman, and Sunder

money to their funds. In addition, Shleifer and Vishny (1997) describe how arbitrageurs must consider the risk of the funds’
investors withdrawing their capital before a mispricing has been resolved. This risk would lead short-term investors to avoid
securities whose mispricing would not resolve quickly enough.
We examine the association between FRQ and holding costs by focusing on the trading decisions of ‘‘transient’’
institutional investors. Prior work indicates that transient institutions engage in strategies based on financial statement variables
and on stock price momentum (e.g., Collins, Gong, and Hribar 2003; Lev and Nissim 2006; Ke and Ramalingegowda 2005).
Given their use of different types of trading strategies and their short-term horizon, transient institutions are natural candidates
to examine whether holding costs due to poor FRQ affect how they implement these investment strategies. We expect that
trading strategies that require a longer holding period to exploit will impose significant holding costs on transient investors and
discourage them from investing in firms that are potentially mispriced.
We use two well-documented trading strategies—momentum and value—to provide a powerful research setting to
examine the impact of FRQ on holding costs. Poor FRQ could lead to greater uncertainty, increasing the potential future returns
that sophisticated investors could earn from both strategies (e.g., Ali, Hwang, and Trombley 2003; Zhang 2006). However,
future returns are generally earned more quickly in the momentum strategy than in the value strategy (Jegadeesh and Titman
1993; Ali et al. 2003). Price correction for value strategies tends to span multiple years as investors revise their beliefs about
firm fundamental value (La Porta, Lakonishok, Shleifer, and Vishny 1997). We predict that poor FRQ increases the length of
time needed for the price correction in value firms. As a result, poor FRQ will be associated with transient institutional investors
being less likely to invest in value stocks and more likely to invest in momentum stocks, whose returns are inherently more
short-term. Results consistent with our prediction would provide evidence on the role of FRQ in affecting not just the benefits
to trading strategies, but also the holding costs.
We provide evidence on the link between FRQ and an institutional investor’s willingness to pursue a trading strategy in
three ways. First, we test whether FRQ influences the extent to which transient institutional investors take large positions in
individual value and momentum stocks in their portfolio. We use the percent of large portfolio weights or ‘‘bets’’ in a particular
stock held by transient investors as a measure of whether short-term institutions are attempting to exploit a strategy (Bushee and
Goodman 2007). Second, we examine whether FRQ influences the length of time that institutional investors are observed to
hold large positions across the two strategies. We use the number of quarters that investors hold large bets in a particular stock
as the measure of the length of time needed to exploit the trading strategy. Third, we examine how short-term institutional
investors tilt their portfolio toward (away from) firms with poor FRQ when pursuing momentum (value) strategies.
We find support for our prediction that FRQ influences portfolio decisions through holding costs. Poor FRQ increases
(decreases) the likelihood that a transient institution has a large portfolio weight in an individual momentum (value) stock. In
addition, poor FRQ increases the amount of time an institution holds a position in a value stock, but does not increase the
amount of time that an investor holds a position in a momentum stock. At the portfolio level, we find that transient institutions
are more likely to tilt their portfolios toward (away from) firms with poor FRQ when pursuing a momentum (value) strategy.
This evidence is consistent with poor FRQ reducing the net benefits of a value strategy to a short-term investor by increasing
the holding costs of value stocks through a longer holding period. These results suggest that the willingness of an institutional
investor to take a large bet on an individual stock in a portfolio or to follow a particular strategy is not just a function of the
benefits provided by poor FRQ, but also by the expected holding costs inherent in strategies that rely on financial reporting for
price correction.
Our findings provide insight into how FRQ relates to the limits to arbitrage literature. Shleifer and Vishny (1997) provide a
model in which investors rationally focus on investments that will yield short-term returns even if there might be better long-
term investment opportunities. We show how FRQ plays a role in this process by influencing an investor’s estimated holding
period and size of position for a trading strategy.
Our results also contribute to recent work on the link between FRQ and informed trading by institutions. Using
international data, Maffett (2012) provides evidence that poor FRQ increases the association between changes in institutional
investor ownership and future returns. This evidence suggests that poor FRQ can enhance the returns from trading strategies
that rely on information discovery by increasing the value of private information. Our work complements Maffett’s (2012)
analysis by providing insight into the potential costs associated with FRQ for strategies that rely on investors revising their
beliefs about fundamental values.
Finally, our analysis contributes to research on the extent to which FRQ influences the timeliness of returns. Callen, Khan,
and Lu (2013) provide evidence that FRQ affects the speed with which investors incorporate information into prices. More
generally, Zhang (2006) finds that uncertainty due to reporting noise or to underlying fundamental volatility exacerbates
investor underreaction, increasing the profitability of momentum strategies. Our work complements this literature by showing
that FRQ influences institutional investor perceptions of holding costs from different strategies. Transient institutions perceive
that poor FRQ is likely to extend the holding period for a value strategy to a sufficient degree that they choose to not follow that
strategy, shifting toward a more short-term strategy (i.e., momentum).

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 89

The remainder of the paper is organized as follows. Section II provides the hypothesis development. Section III discusses
our variable measurement. Section IV details our empirical tests and results. Section V concludes.

II. HYPOTHESIS DEVELOPMENT


A rational investor deciding to pursue a particular trading strategy would consider the relevant benefits and costs of that
strategy.1 The benefits stem from the ability to identify mispriced securities that yield arbitrage opportunities within the
investor’s investment horizon. Among the costs, a rational investor would consider the total risk that would be added to the
investor’s portfolio by tilting it toward the mispriced securities. The total risk is a function of both the amount of risk that
cannot be fully hedged (e.g., idiosyncratic risk) and the number of months the investor must maintain an abnormal weight on
that security. Pontiff (2006) illustrates the importance of expected holding costs with an example where the holding costs are
the product of the cost per month and the number of months until the mispricing resolves. He notes that if the mispricing does
not correct sufficiently quickly, then the holding costs could exceed the return benefits, leading the investor to forgo the
arbitrage opportunity.
Prior work suggests that sophisticated institutional investors are able to trade off these benefits and costs in the presence of
poor FRQ. Maffett (2012) finds that institutional investors exhibit a higher degree of informed trading when FRQ is poor. In his
international setting, the benefits of opacity outweigh the costs, on average, for institutions that have decided to trade in those
markets.
We build on his analysis by exploring the effect of FRQ on the benefits and costs of value and momentum trading
strategies. These two strategies differ in the nature and timing of their price correction.2 Value strategies correct over a longer
period of time as investors learn about the firm’s fundamental value, while momentum strategies predict short-term returns.
This key difference allows us to examine whether poor FRQ imposes significant holding costs on certain strategies, which
would offset the benefits enough to make the strategies unprofitable for sophisticated investors to implement.

Financial Reporting Quality and Value Strategies


We examine value strategies following the assumption that investors use these strategies to identify firms whose prices
differ from fundamental values. Investors earn returns when the market values eventually approach fundamental values in the
future. The public release of future earnings is important to the value investor because the price correction occurs when
uninformed investors revise their estimates of firm value. Poor FRQ reduces the extent to which uninformed investors will rely
on financial information as an input in their valuation decisions. This lack of quality information influences the degree to which
the prices anticipate the eventual economic value that will be created by the firm.3
The link between FRQ and the decision to pursue a given strategy can be drawn from McNichols and Trueman (1994).
They provide a model where an investor considers the future beliefs of other investors when evaluating the net benefits of
gathering information about a firm’s fundamental value. Based on the ‘‘conventional Wall Street wisdom’’ that ‘‘a bargain that
remains a bargain is no bargain’’ (McNichols and Trueman 1994, 70), their model shows that if stock prices are slow to
recognize information about a firm’s fundamental value, then a short-term investor is less likely to take a large position in the
undervalued stock. In other words, low-quality public disclosure increases the risk that prices will not correct in the short run.
This leads to our first hypothesis:
H1: Poor financial reporting quality reduces the extent to which short-term investors follow value strategies because it
increases holding costs.
H1 also suggests that FRQ plays a role in limits to arbitrage models developed by Shleifer and Vishny (1997). Poor FRQ
increases the holding period of value strategies, making value stocks less appealing to professional investors who are concerned
that their investors might withdraw their capital.
While we expect that the holding costs related to poor financial reporting are likely to be material, we may not find support
for H1 if poor financial reporting also improves the expected returns from value strategies sufficiently to compensate short-term

1
We use ‘‘strategy’’ to refer to either an investor’s strategy with respect to an individual stock (i.e., do they place a large bet on that stock?) or an
investor’s strategy for their entire portfolio (i.e., do they tilt their portfolio toward a particular strategy?).
2
We focus on value and momentum strategies, rather than accruals and post-earnings announcement drift strategies, because we measure FRQ using
accrual quality (described in detail in Section III). As accruals are a component of both the accrual strategy and the drift strategy, there is a risk that poor
FRQ (extreme accruals) might mechanically coincide with more extreme long positions in these strategies (i.e., lower values or accruals, higher values
of unexpected earnings), which could confound our analysis.
3
While investors could offset the poor FRQ to some degree by gathering information about how to more effectively interpret these disclosures (e.g., Kim
and Verrecchia 1997), we assume that poor FRQ from mandatory accounting will, in general, lead to an overall information set with lower precision.

The Accounting Review


Volume 94, Number 3, 2019
90 Bushee, Goodman, and Sunder

investors for these holding costs. Poor FRQ could result in initial prices that are noisier signals about future fundamental values,
and such noise could create an opportunity for sophisticated investors to exploit their superior information gathering. For
example, the uncertainty created by poor FRQ could increase the degree to which market prices depart from fundamentals,
making value strategies more valuable in environments where FRQ is poor.

Financial Reporting Quality and Momentum Strategies


A large literature discusses the existence of momentum in security prices (e.g., Jegadeesh and Titman 1993; Asness,
Moskowitz, and Pedersen 2013). Momentum strategies tend to be short-term, with the price correction beginning in the next
month’s returns.4 Firms designated as past winners tend to have positive earnings announcement returns during the quarter
following portfolio formation. However, in contrast to value strategies, where abnormal returns at the earnings announcements
of value stocks persist for years into the future (e.g., La Porta et al. 1997), the majority of the earnings announcement returns
for past winners happen during the quarter after portfolio formation.5 This suggests that momentum has an earnings component
that is driven by investor responses to short-term earnings, as opposed to investors gradually learning about firm fundamental
values over multiple quarters.6
Given that momentum strategies are not necessarily tied to investor beliefs about long-term cash flows and fundamental
values, there should be no incremental holding costs associated with poor FRQ. Therefore, a short-term investor who considers
the holding costs associated with poor financial reporting strategy to be high might choose a momentum strategy where such
holding costs are expected to be lower.
For a short-term investor, poor FRQ could improve the appeal of momentum strategies. First, poor FRQ could signal stock
prices with greater underreaction (e.g., Callen et al. 2013), which would make momentum strategies more profitable (e.g.,
Zhang 2006). Second, poor FRQ could signal that returns are more continuous through the year (due to less pronounced
responses to earnings news during the year), which also predicts more profitable momentum strategies (Da, Gurun, and
Warachka 2014).
The above discussion suggests that poor FRQ is not likely to make a momentum strategy more costly due to holding costs.
In fact, poor FRQ could make a momentum strategy more profitable. As a result, we expect that momentum strategies are a
more appealing option for short-term investors when FRQ is poor. This leads to our second hypothesis:
H2: Poor financial reporting quality increases the extent to which short-term investors follow momentum strategies
because it increases the potential return benefits without the additional holding costs.

III. VARIABLE MEASUREMENT

Measurement of Financial Reporting Quality


Our FRQ construct refers to how precisely financial reports reveal information about the firm’s underlying fundamental
value. As the underlying value of the firm is a function of its cash flows, this is analogous to how precisely a firm’s accruals
provide information about cash flows. Building on Dechow and Dichev (2002), a large literature uses the volatility of
unexpected accruals to measure accrual quality and to identify cases where the precision of accruals for revealing cash flow
information is low (see Dechow, Ge, and Schrand [2010] for a review). We use accrual quality as a measure of a firm’s FRQ.
Following past research (e.g., Core, Guay, and Verdi 2008), we estimate accrual quality using the procedure described in
Appendix A. Larger values of AQ indicate poorer FRQ. We transform AQ into a decile rank [0,9], RAQ. We also define an
indicator variable, POOR_AQ, that equals 1 for firms in the top 30 percent of the AQ distribution, and 0 otherwise.7

4
For example, Jegadeesh and Titman (1993) provide evidence that the cumulative returns from a momentum strategy (based on the lagged six-month
returns) peak 12 months after portfolio formation, with negative monthly returns during the second year after portfolio formation.
5
In Chan, Jegadeesh, and Lakonishok (1996), the abnormal earnings announcement returns of winners exceed those of losers by 2.6 percent in the
quarter following formation, but this decreases to 1.0 percent, 0.3 percent, and 0.2 percent over next four quarters.
6
Chan et al. (1996) and Chordia and Shivakumar (2006) provide evidence that short-term earnings momentum is associated with price momentum.
7
Francis, LaFond, Olsson, and Schipper (2005) note that AQ can be decomposed into innate and discretionary components, where the innate component
is associated with economic determinants of accrual quality (firm size, volatility of cash flows, volatility of sales, operating cycle, and frequency of
negative earnings). We expect that investors attempting to judge the time required for a value trading strategy will mainly be concerned with the
persistent component of accrual quality, and we expect innate accrual quality to be more persistent. As a result, we repeat our firm-year and institution-
quarter analysis substituting the innate component for AQ. We continue to observe results consistent with H1 (poor innate AQ increases holding costs
for value) using both firm-year tests and the institution-quarter tests. We also generally observe support for H2 (poor innate AQ makes momentum more
appealing), although the interaction term in Equation (1) remains positive, but becomes insignificant.

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 91

Measurement of Value Strategy


While we are unable to observe the underlying value-to-price ratios that sophisticated investors use in their value strategies,
there is extensive evidence that high book-to-market firms have higher future stock returns (e.g., Fama and French 1992;
Lakonishok, Shleifer, and Vishny 1994). In addition, La Porta et al. (1997) document that a material portion of the returns from
the book-to-market strategy is related to future earnings realizations over an extended period of time.
Existing work has also examined the role idiosyncratic risk plays in the pricing of book-to-market securities. Ali et al.
(2003) find that the book-to-market strategy provides the greatest returns when idiosyncratic risk is high. Our analysis
complements this work by providing insight into the expected holding period component of holding costs and the extent to
which it drives investor decisions.8
We measure book-to-market (BEME) in the following way. The numerator is calculated on June 30 using the book value
of common equity (CEQ) from the fiscal year that ended in the previous calendar year. The denominator is the market value as
of the calendar quarter-end. We transform BEME into a decile rank [0,9], RBEME. As we are only able to observe long
positions by institutional investors, we define an indicator variable capturing observations that are more likely to be included in
those long positions, VALUE, that equals 1 for firms in the top 30 percent of the BEME distribution, and 0 otherwise.
While we treat book-to-market as an indicator of a value strategy opportunity, there are several papers that argue that it is a
risk factor (e.g., Fama and French 1992). In this context, holding costs continue to play a role, albeit with a different
interpretation. Instead of representing a limit to arbitrage, these costs would be part of the risk borne by investors that motivates
a larger expected return. From this perspective, short-term investors might not view the higher expected return on book-to-
market stocks as sufficient compensation for the risk associated with holding that stock, given their particular aversion to
holding securities for a long period of time.

Measurement of Momentum Strategy


We measure momentum using data on the firm’s returns over the previous year. Following Fama and French (2008) and
several other studies, we omit the most recent month and calculate buy-and-hold returns over the previous 11 months (PAST_
RET). Next, we transform PAST_RET into a decile rank [0,9], RPAST_RET. Again, as we are only able to observe long
positions by institutional investors, we define an indicator variable capturing observations that are more likely to be included in
those long positions, PAST_WINNER, that equals 1 for firms in the top 30 percent of the PAST_RET distribution, and 0
otherwise.

Measurement of Institutional Investor Bets


We use institutional investors as a proxy for the type of sophisticated, short-term investor that would attempt to exploit
trading strategies. While we cannot directly observe an institutional investor’s perception of expected returns and costs from
different trading strategies, we can observe the positions that institutional investors report on a quarterly basis. Theoretical work
suggests that an investor’s decision to purchase an abnormal stake in a particular firm is a function of the precision of the
investor’s private information (e.g., Kim and Verrecchia 1997). Bushee and Goodman (2007) provide evidence that the
institutional investors placing larger bets on individual stocks are more likely to execute trades that are consistent with private
information in those stocks (e.g., purchasing before good news has been announced and selling after good news has been
released).
For each institution firm-quarter holding in a stock, we calculate its portfolio weight (WGT). Next, we partition the
institutional owners in each firm into quintiles based on WGT, requiring that the firm-quarter have at least 50 institutional
owners. We label the institutions’ positions in the top quintile as ‘‘large bets.’’ Comparing a specific institution’s portfolio
weight with the weights of the other institutional owners at the same firm-quarter also removes any firm-quarter fixed effect that
might determine the weight for that firm in the market portfolio.

Which Institutions Create Bets?


To identify whether short-term investors are willing to take a bet in a particular stock, we calculate the percent of large bets
that are taken by transient institutions (TRA_BET). Transient classifications are calculated using the procedure outlined in
Bushee (2001). Prior work provides evidence that transient institutions take positions that are consistent with knowledge of

8
Poor FRQ could also be related to the higher per-month cost of maintaining an abnormal weight on a given firm. For example, Rajgopal and
Venkatachalam (2011) document that FRQ is associated with changes in idiosyncratic risk over time. To address this concern, we include idiosyncratic
risk as a control variable in all of our tests.

The Accounting Review


Volume 94, Number 3, 2019
92 Bushee, Goodman, and Sunder

unexpected future earnings (Ke and Petroni 2004), and with analyst revisions in long-term growth rates that will be reported in
the immediate future (Ke, Ramalingegowda, and Yu 2006). Prior research also suggests that transient institutions have a greater
understanding of the accrual anomaly (Collins et al. 2003), and they exploit strategies such as post-earnings announcement drift
(Ke and Ramalingegowda 2005).
If poor FRQ increases the holding costs associated with pursuing a value strategy, as predicted by H1, then we expect
fewer large positions by transient institutions in value firms with poor FRQ. Moreover, if poor FRQ makes momentum
strategies relatively more appealing, then we expect more large positions by transient institutions at firms that were past winners
when FRQ is poor.

How Long are Bets Held?


We complement the analysis of the percent of bets by transient investors with the realized time that bets are maintained at a
given firm. Transient investors should maintain bets for a shorter period of time, ceteris paribus, given their investing horizon.
However, the decision to create a bet is a trade-off between both the expected benefits and costs of creating a bet. In contrast,
realized time held provides a more direct measure of the holding cost that would be induced by a longer required holding
period.
We measure the amount of time that the average institution maintains an abnormally positive position (THELD) using the
following steps. First, we identify the institutions that have a bet in place (the top quintile of WGT). We count the number of
quarters that an institution maintains a weight above the middle quintile over the following eight quarters. As a result, THELD
has a range between 0 (reduces weight in the quarter tþ1) and 8 (above middle quintile in quarter tþ8). If an institutional
investor position is not available on the Thomson Financial Institutional Investor database during a future quarter, then we
assume that the institution has reduced their position below the middle quintile. To ensure that THELD is not distorted by
observations where it is not possible for an institution to hold the firm for eight quarters, we drop firms and institutions that are
not listed on the Thomson Financial Institutional Investor database for the eight quarters following quarter t with at least 50
owners in each of those quarters. To aggregate the holding periods of the multiple institutions that own a given firm to the firm-
quarter level, we take the average value across all institutions with large bets.

Control Variables
Our multivariate analysis controls for the fact that FRQ could be correlated with other firm characteristics. All continuous
control variables are transformed into decile ranks on a [0,9] scale to control for outliers.
First, we include multiple controls related to the riskiness of a security. We include two market-based measures of risk:
systematic risk (BETA), and idiosyncratic risk (IRISK). In addition, Zhang (2006) suggests firm size (MVE), analyst coverage
(COV), and firm age (AGE) as additional measures of uncertainty.
Second, prior work suggests that FRQ is correlated with a stock’s liquidity (Ng 2011), and liquidity could influence an
investor’s decision to pursue a large investment in a given stock. We include share turnover (TURN), where high turnover
could reflect either low transaction costs or an investor’s ability to mask their own private information among the trades of
others.
Third, we control for the total number of institutional owners (NUM_INST), which signals the presence of sophisticated
parties that could accelerate the process by which prices reflect information.
Fourth, some institutions are subject to prudent person restrictions, which can constrain their ability to take large positions
in certain stocks. This could result in institutions making smaller investments in firms with accounting ratios suggesting risk.
Thus, we include leverage (LEV) and the occurrence of negative earnings (NEARN) as additional risk controls.
Fifth, we control for other well-documented accounting-related strategies that might be employed by short-term investors.
We include the accruals from the most recent fiscal year (ACC). Prior work also suggests that transient investors also might
exploit post-earnings announcement drift (Ke and Ramalingegowda 2005) as part of their strategy. We include the lagged
unexpected earnings (SUE) from the most recent quarter before the measurement of bet positions. SUE is computed as
seasonally adjusted income divided by the standard deviation of seasonal earnings changes over the previous eight quarters.
Sixth, to control for general unobserved heterogeneity across industries and time periods, we include industry fixed effects
(based on Fama-French 48 industry codes) and quarterly fixed effects (based on the quarterly report date for the institutional
investor data).
Last, when TRA_BET is the dependent variable, we also include the percent of non-bet positions by transients (TRA_
NOBET) to control for the possibility that transient institutions prefer certain stocks in general. Including this control
variable helps focus our analysis on situations where transient investors create large positions, as opposed to smaller
positions.

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 93

TABLE 1
Descriptive Statistics
Standard Lower Upper
n Mean Deviation Quartile Median Quartile
TRA_BET (%) 106,917 36.5178 18.8067 21.7391 35.0000 50.0000
TRA_NOBET (%) 106,917 28.5957 10.2501 21.1538 27.2727 34.3629
THELD 83,912 4.3874 1.0749 3.6538 4.4359 5.1628
ln(BEME) 106,917 0.8660 0.7877 1.2951 0.8052 0.3692
PAST_RET 106,917 0.1745 0.5771 0.0951 0.1096 0.3366
VALUE 106,917 0.1815 0.3854 0.0000 0.0000 0.0000
PAST_WINNER 106,917 0.3114 0.4630 0.0000 0.0000 1.0000
ln(AQ) 106,917 3.8080 0.7677 4.2874 3.7767 3.2953
ln(MVE) 106,917 7.2743 1.4282 6.2337 7.0907 8.1180
BETA 106,917 1.1321 0.6867 0.6713 1.0397 1.4502
ln(IRISK) 106,917 2.4051 0.4578 2.7437 2.4329 2.0984
ln(TURN) 106,917 0.0740 0.8693 0.5625 0.0658 0.7035
NUM_INST 106,917 193.1243 179.9394 87.0000 135.0000 228.0000
SUE 106,917 0.0674 3.3329 0.5474 0.0856 0.6850
ACC 106,917 0.0092 0.0593 0.0150 0.0057 0.0316
COV 106,917 10.9326 8.2263 4.5000 9.0833 15.8333
AGE 106,917 28.6371 18.3974 14.0932 23.5945 38.7781
LEV 106,917 0.2201 0.1580 0.0865 0.2178 0.3306
NEARN 106,917 0.1310 0.3373 0.0000 0.0000 0.0000
Table 1 presents descriptive statistics on variables used in the firm-quarter analysis.
Variable definitions are reported in Appendix A.

IV. EMPIRICAL TESTS AND RESULTS

Sample Selection Criteria for Firm-Quarter Analysis


We require data from three main data sources. First, we use the Thomson Financial Spectrum database between 1981 and
2013 to obtain institutional investor data. Second, we obtain data on FRQ and other accounting control variables from
Compustat. We remove firms in financial industries for which our accrual-based measure of FRQ might be difficult to
interpret.9 Third, we use CRSP to estimate price and volume variables for common stocks listed on the NYSE, AMEX, or
NASDAQ exchanges. For all of our institutional investor tests, we calculate ranks of our variables based on this sample.
For the firm-quarter-level analysis, we require that the firm have at least 50 institutional owners in a given firm-quarter to
compute TRA_BET, producing a sample of 106,917 firm-quarter observations from 1981 to 2013. The sample used to estimate
how long bets are maintained (THELD) is slightly smaller as it requires two years of future data; it contains 83,912 firm-quarter
observations from 1981 to 2011.

Descriptive Statistics for Firm-Quarter Analysis


Table 1 presents descriptive statistics for the dependent variables used in this study. The mean (median) percent of bets due
to transient investors (TRA_BET) is 36.52 percent (35.00 percent), with a lower quartile of 21.74 percent and an upper quartile
of 50.00 percent. For comparison, the mean (median) percent of non-bet positions held by transients (TRA_NOBET) is 28.60
percent (27.27 percent). This difference indicates that transients are more highly represented among large bet positions,
consistent with the transient classification capturing institutions that actively manage their portfolio and engage in trading
strategies that require large bets on certain stocks.
Table 1 also provides descriptive statistics on the length of time held (THELD). The mean (median) THELD is 4.39 (4.44)
quarters, with a lower quartile of 3.65 quarters and an upper quartile of 5.16 quarters. Table 1 also provides descriptive statistics
for the independent variables used in this study before these variables are transformed into decile ranks.

9
We remove the following Fama-French industries: 44 (Banking), 45 (Insurance), 46 (Real Estate), 47 (Trading), and 48 (Other).

The Accounting Review


Volume 94, Number 3, 2019
94 Bushee, Goodman, and Sunder

TABLE 2
Univariate Differences Across Sample Split Based on TRA_BET and THELD

Panel A: Sample Split Based on TRA_BET


TRA_BET ¼ High TRA_BET ¼ Low Difference
Mean Mean in Means t-statistic
POOR_AQ 0.2409 0.1275 0.1134*** 17.32
VALUE 0.1822 0.1808 0.0014 0.20
PAST_WINNER 0.4086 0.2138 0.1949*** 22.61
RMVE 6.3768 6.7792 0.4024*** 7.53
RBETA 5.1967 4.0496 1.1471*** 17.52
RIRISK 4.2514 2.4074 1.8440*** 29.23
RTURN 6.3340 4.6259 1.7081*** 33.56
RNUM_INST 6.4543 6.8229 0.3686*** 7.34
RSUE 4.7931 4.3924 0.4007*** 13.16
RACC 4.4554 4.5762 0.1209*** 3.99
RCOV 6.2229 6.4158 0.1929*** 3.27
RAGE 4.7271 5.7556 1.0285*** 13.12
RLEV 4.7242 4.5860 0.1382** 2.04
NEARN 0.1729 0.0889 0.0839*** 13.82

Panel B: Sample Split Based on THELD


THELD ¼ Low THELD ¼ High Difference
Mean Mean in Means t-statistic
POOR_AQ 0.2150 0.1109 0.1041*** 14.74
VALUE 0.1879 0.1362 0.0517*** 7.71
PAST_WINNER 0.3447 0.2812 0.0635*** 6.96
RMVE 6.6972 7.0690 0.3718*** 7.30
RBETA 5.1465 3.9587 1.1878*** 16.78
RIRISK 3.9277 2.2246 1.7031*** 26.27
RTURN 6.4214 4.5207 1.9006*** 31.67
RNUM_INST 6.8027 7.0570 0.2543*** 5.11
RSUE 4.6100 4.6520 0.0421 1.18
RACC 4.5053 4.5161 0.0108 0.31
RCOV 6.5473 6.5824 0.0352 0.55
RAGE 5.0197 5.8888 0.8690*** 10.22
RLEV 4.6340 4.6788 0.0448 0.63
NEARN 0.1568 0.0799 0.0769*** 11.91
*, **, *** Significantly different from zero at 0.10, 0.05, and 0.01, respectively, using a two-tailed test.
Table 2 presents univariate comparison for the sample split based on percent of bets by transient investors (TRA_BET) and the average holding period of a
large bet (THELD). Panel A (Panel B) splits the sample evenly into high and low groups based on TRA_BET (THELD). Both panels also report t-statistics
for the difference across groups, calculated using two-way clustering based on firm and quarter.
All variable definitions are reported in Appendix A.

Table 2 provides descriptive statistics on the types of firms where transient institutions place large bets and the extent to
which these bets vary with poor FRQ. Panel A partitions the sample into high and low groups based on TRA_BET. Firms with a
larger percent of transient bets have significantly poorer FRQ (RAQ). However, this raises the question as to what types of
trading strategies these institutions are following that lead them to create large long positions in firms with poor FRQ.
Panel A of Table 2 also indicates that firms with a greater percent of transient bets are more likely to be past winners
(PAST_WINNER). These firms also differ on other dimensions, such as greater uncertainty measures (larger values of RBETA
and RIRISK, smaller values of RAGE and RCOV), greater liquidity (RTURN), fewer institutional owners (RNUM_INST),
smaller market capitalization (RMVE), lower accruals (RACC), greater recent earnings surprises (RSUE), and more riskier firms
from a prudent person perspective (RLEV and NEARN).

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 95

TABLE 3
Univariate Results for Samples Partitioned Based on Financial Reporting Quality and Trading Strategy

Panel A: Average Values of TRA_BET Based on Strategy and FRQ


(3) (4)
(1) (2) PAST_ PAST_
VALUE VALUE WINNER WINNER
¼1 ¼0 (1)  (2) ¼1 ¼0 (3)  (4)
Poor AQ 39.7333 43.8528 4.1195*** 49.5117 39.8006 9.7111***
(5.12) (17.67)
Medium AQ 37.4491 36.8926 0.5565 42.8523 34.2368 8.6155***
(0.73) (19.53)
Good AQ 34.7771 32.6243 2.1528*** 37.8806 31.0132 6.8674***
(3.23) (12.55)
Poor  Good 4.9562*** 11.2285*** 6.2723*** 11.6311*** 8.7874*** 2.8437***
(5.83) (18.93) (7.05) (15.40) (15.59) (3.89)

Panel B: Average Values of THELD Based on Strategy and FRQ


(1) (2) (3) (4)
PAST_ PAST_
VALUE VALUE WINNER WINNER
¼1 ¼0 (1)  (2) ¼1 ¼0 (3)  (4)
Poor AQ 4.0921 3.9779
0.1142** 3.8116 4.0937 0.2821***
(2.33) (8.16)
Medium AQ 4.0695 4.3666 0.2971*** 4.2085 4.3756 0.1671***
(6.88) (5.75)
Good AQ 4.3257 4.6483 0.3226*** 4.4924 4.6343 0.1419***
(8.19) (4.56)
Poor  Good 0.2337*** 0.6704*** 0.4367*** 0.6808*** 0.5405*** 0.1402***
(4.38) (17.82) (7.56) (16.10) (14.54) (3.68)
*, **, *** Significantly different from zero at 0.10, 0.05, and 0.01, respectively, using a two-tailed test.
Table 3 presents univariate comparison for the sample split based on FRQ and on whether a firm would be included in the long position of a particular
strategy (value or momentum). Panel A provides analysis of TRA_BET split based on strategy (value or momentum) and FRQ. Panel B provides analysis
of THELD split based on strategy (value or momentum) and FRQ. The top and bottom 1 percent of all continuous variables are winsorized. Both panels
also report t-statistics for the difference across groups, calculated using two-way clustering based on firm and quarter.
All variable definitions are reported in Appendix A.

Panel B of Table 2 partitions the sample based on the realized holding period (THELD). We observe similar differences
with respect to FRQ, size, uncertainty, and liquidity. However, there is not a significant difference in RSUE, RACC, RCOV,
RLEV, or NEARN.
Table 3 provides univariate evidence on our hypotheses. For each of our dependent variables (TRA_BET and THELD), we
perform a difference-in-differences analysis to test our predictions that poor FRQ influences institutional investor positions
differently between value and momentum strategies due to holding costs. A difference-in-differences approach is helpful in our
context because Table 2 indicates that transient institutions prefer stocks with poor FRQ, possibly due to information
advantages in opaque settings (e.g., Maffett 2012).10 Thus, Table 3 allows us to understand whether the degree to which
transient institutions prefer poor FRQ depends on other firm characteristics that could be linked to holding costs.
Panel A of Table 3 reports average values of TRA_BET for samples partitioned based on FRQ and trading strategy.
Columns (1) and (2) indicate that when FRQ is poor, transient institutions are less likely to take bets in VALUE firms relative to

10
When analyzing differences in our dependent variables across groups, controlling for these main effects of both VALUE and PAST_WINNER is also
important, as both are correlated with future returns, which could itself influence the holding period. For example, all else equal, transient institutions
might prefer to hold firms with higher future expected returns. In addition, higher expected returns might correspond to a shorter value of THELD, as
prior work suggests that institutions with large bets sell following positive future returns (Bushee and Goodman 2007). Similarly, controlling for the
main effect of FRQ is important, as different types of institutions could have preferences for different firm disclosure choices (e.g., Bushee and Noe
2000).

The Accounting Review


Volume 94, Number 3, 2019
96 Bushee, Goodman, and Sunder

non-VALUE (39.73 percent versus 43.85 percent). However, when FRQ is good, transient institutions are more likely to take
bets in VALUE firms relative to non-VALUE firms (34.78 percent versus 32.62 percent). The difference-in-differences (6.27
percent) is also statistically significant (t ¼ 7.05) and economically meaningful, as it is 17.18 percent of the unconditional mean
of TRA_BET. These results can also be interpreted as evidence that transient institutions prefer poor FRQ by a larger degree
among firms that are non-value stocks (11.23 . 4.96), which is consistent with our hypothesis that poor FRQ increases holding
costs among value stocks and makes poor FRQ relatively less preferable to transient investors.
Columns (3) and (4) of Table 3, Panel A provide a parallel analysis for the momentum strategy. Transient institutions are
more likely to have large bets in PAST_WINNER relative to other firms (49.51 percent versus 39.80 percent) among firms with
poor FRQ. However, this difference is smaller among firms with good FRQ (37.88 percent versus 31.01 percent). Furthermore,
this difference-in-differences (2.84 percent) is statistically significant (t ¼ 3.89) and economically meaningful, as it is 7.79
percent of the unconditional mean of TRA_BET.
Panel B of Table 3 provides univariate evidence on the length of holding periods. Columns (1) and (2) indicate that among
firms with poor FRQ, the holding period is longer among VALUE firms relative to other firms (4.09 versus 3.98). However,
among firms with good FRQ, the holding period among VALUE firms is smaller than among other firms (4.33 versus 4.65). In
addition, the difference-in-differences (0.44) is statistically significant (t ¼ 7.56) and economically meaningful, as it is 9.95
percent of the unconditional mean of THELD. Columns (3) and (4) examine whether holding periods vary based on
momentum. When FRQ is poor, the holding period for PAST_WINNER is lower than for other firms (3.81 versus 4.09). For
firms with good FRQ, the gap in holding period between PAST_WINNER and other firms is still negative, but smaller (4.49
versus 4.63). The difference-in-differences is again statistically significant (t ¼ 3.68), representing 3.20 percent of the
unconditional mean of THELD.
Overall, the univariate analysis in Table 3 provides evidence that the combination of FRQ and trading strategy influences
whether a short-term institution takes a large position and how long that position is held.

Multivariate Analysis of Determinants of Large Bets


To examine whether the poor FRQ influences the extent to which short-term investors follow value or momentum
strategies when placing large bets, we estimate the following model:
TRA BET ¼ h0 þ h1 POOR AQ  VALUE þ h2 POOR AQ  PAST WINNER þ h3 VALUE þ h4 PAST WINNER
þ h5 POOR AQ þ h6 CONTROLS þ h7 POOR AQ  CONTROLS þ Quarter Fixed Effect
þ Industry Fixed Effect þ Error ð1Þ
Table 4 presents estimated coefficients for Equation (1), with t-statistics based on standard errors clustered by firm and quarter.
The coefficient on the interaction between POOR_AQ and VALUE is negative and significant (t ¼3.18), indicating that poor
FRQ makes transient institutions relatively less likely to hold large bets in value stocks (i.e., pursue a value strategy). We also
observe a positive and significant coefficient on the interaction between POOR_AQ and PAST_WINNER (t ¼ 2.23), indicating
that that transients are more likely to hold large bets in past winners (i.e., follow a momentum strategy) for firms with poor
FRQ. These results are consistent with our holding cost framework, where short-term institutions perceive that poor FRQ
decreases (increases) the net benefits of a value (momentum) strategy by influencing the expected holding period.11 Thus, FRQ
influences the types of trading strategies pursued by short-term institutions.12

Multivariate Analysis of Holding Period for Bets


As the decision to place a bet could relate to both the expected benefits and costs, we attempt to isolate the effect related to
holding costs by examining determinants of expected holding period. We examine the holding period for bets by estimating the
following model:

11
We also estimated a version of Table 4 replacing the indicator variables (POOR_AQ, VALUE, and PAST_WINNER) with their decile ranks (RAQ,
RBEME, RPAST_RET). In this specification, we continue to observe support for our predictions related to value (coefficient on RAQ  RBEME is
negative and significant). The results related to momentum are no longer significantly positive (coefficient on RAQ  RPAST_RET is no longer
significant); however, as expected, the coefficient on RAQ  RPAST_RET is statistically greater than the coefficient on RAQ  RBEME.
12
While this analysis focuses on short-term institutions, we could also consider the percent of long-term or dedicated institutional investors taking large
positions in individual stocks. For this placebo analysis, we estimate a version of Equation (1) replacing TRA_BET and TRA_NOBET with DED_BET
(percent of bets by dedicated investors) and DED_NOBET (percent of non-bets by dedicated investors). We find an insignificant coefficient on POOR_
AQ  PAST_WINNER, and a negative and statistically significant coefficient on POOR_AQ  VALUE. However, a test of the differences in these
coefficients is not significant at conventional levels (t-statistic ¼1.38), suggesting that there are no clear results for this placebo test. To ensure that we
are not capturing some omitted variable that could influence both transient and dedicated bets at the firm level, we estimate a version of Equation (1)
that includes DED_BET and DED_NOBET. The results produce inferences regarding H1 and H2 that are the same as Table 4.

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 97

TABLE 4
Regression of Transient Investor Bets on Financial Reporting Quality and Trading Strategy Variables (Equation (1))
Predicted
Sign Estimate t-statistic
POOR_AQ  VALUE  (H1) 1.5594*** 3.18
POOR_AQ  PAST_WINNER þ (H2) 0.6761** 2.23
VALUE 0.1215 0.43
PAST_WINNER 4.2611*** 21.14
POOR_AQ 0.1613 0.12
RMVE 0.9459*** 5.32
RBETA 0.0785 1.44
RIRISK 1.0040*** 13.06
RTURN 1.7425*** 22.40
RNUM_INST 1.9742*** 10.51
RSUE 0.2113*** 9.26
RACC 0.2042*** 6.69
RLEV 0.3406*** 6.39
RCOV 0.2303*** 3.17
RAGE 0.0443 0.82
NEARN 1.1809*** 3.57
TRA_NOBET 0.6865*** 27.47
POOR_AQ  RMVE 0.1344 0.52
POOR_AQ  RBETA 0.1348 1.46
POOR_AQ  RIRISK 0.3213** 2.54
POOR_AQ  RTURN 0.2359* 1.95
POOR_AQ  RNUM_INST 0.6489** 2.18
POOR_AQ  RSUE 0.0574 1.25
POOR_AQ  RACC 0.0461 0.80
POOR_AQ  RLEV 0.2615** 3.35
POOR_AQ  RCOV 0.0941 0.77
POOR_AQ  RAGE 0.0633 0.76
POOR_AQ  NEARN 0.8554 1.55
POOR_AQ  TRA_NOBET 0.0028 0.14
Quarter Fixed Effects Yes
Industry Fixed Effects Yes
Two-way clustered standard errors (Firm, Quarter) Yes
n 106,917
*, **, *** Significantly different from zero at 0.10, 0.05, and 0.01, respectively, using a two-tailed test.
Table 4 presents estimates of a model predicting whether transient institutions created large bets in a particular stock as a function of both FRQ and
whether the firm would be included in the long position of a value or momentum strategy (Equation (1)). The top and bottom 1 percent of all continuous
variables are winsorized, and t-statistics are calculated using two-way clustering based on firm and quarter.
All variable definitions are reported in Appendix A.

THELD ¼ h0 þ h1 POOR AQ  VALUE þ h2 POOR AQ  PAST WINNER þ h3 VALUE þ h4 PAST WINNER


þ h5 POOR AQ þ h6 CONTROLS þ h7 POOR AQ  CONTROLS þ Quarter Fixed Effect
þ Industry Fixed Effect þ Error ð2Þ

Table 5 presents the estimated coefficients from Equation (2), with t-statistics based on standard errors clustered by firm and
quarter. Consistent with H1, we observe a positive and significant coefficient (t ¼ 2.67) on the interaction between POOR_AQ and
VALUE, indicating that poor FRQ leads to a longer holding period among large bets placed on value stocks. In contrast, we
observe a negative and significant coefficient (t ¼ 2.59) on the interaction between POOR_AQ and PAST_WINNER, indicating
that poor FRQ appears to result in shorter holding periods for past winners. However, our prediction was not that poor FRQ would
necessarily reduce the holding time for momentum, but rather be relatively less costly in terms of holding time for momentum

The Accounting Review


Volume 94, Number 3, 2019
98 Bushee, Goodman, and Sunder

TABLE 5
Regression of Time Held on Financial Reporting Quality and Trading Strategy Variables (Equation (2))
Predicted
Sign Estimate t-statistic
POOR_AQ  VALUE þ (H1) 0.1259*** 2.67
POOR_AQ  PAST_WINNER 0.0760** 2.59
VALUE 0.1272*** 5.61
PAST_WINNER 0.0539*** 3.22
POOR_AQ 0.1200 0.97
RMVE 0.0063 0.40
RBETA 0.0005 0.09
RIRISK 0.0805*** 11.37
RTURN 0.1873*** 25.37
RNUM_INST 0.0549*** 3.28
RSUE 0.0018 0.95
RACC 0.0066** 2.24
RLEV 0.0226*** 4.76
RCOV 0.0149** 2.42
RAGE 0.0006 0.14
NEARN 0.0661** 2.53
POOR_AQ  RMVE 0.0515* 1.80
POOR_AQ  RBETA 0.0135 1.59
POOR_AQ  RIRISK 0.0215* 1.91
POOR_AQ  RTURN 0.0101 0.90
POOR_AQ  RNUM_INST 0.0840** 2.61
POOR_AQ  RSUE 0.0123*** 3.18
POOR_AQ  RACC 0.0043 0.87
POOR_AQ  RLEV 0.0322*** 4.30
POOR_AQ  RCOV 0.0131 1.15
POOR_AQ  RAGE 0.0007 0.09
POOR_AQ  NEARN 0.0150 0.35
Quarter Fixed Effects Yes
Industry Fixed Effects Yes
n 83,912
Difference in Coefficients:
POOR_AQ  VALUE  POOR_AQ  PAST_WINNER þ (H2) 0.2019*** 4.06
*, **, *** Significantly different from zero at 0.10, 0.05, and 0.01, respectively, using a two-tailed test.
Table 5 presents models predicting the length of time an institution holds a large bet as a function of FRQ and whether the firm would be included in the
long position of a value or momentum strategy (Equation (2)). The top and bottom 1 percent of all continuous variables are winsorized. t-statistics are
calculated using two-way clustering based on firm and quarter.
All variable definitions are reported in Appendix A.

compared to value.13 Thus, we also compare these two interaction coefficients. As predicted, a t-test indicates that the coefficient
on the value interaction is significantly more positive than the coefficient on the momentum interaction (t ¼ 4.06).
Overall, this evidence is consistent with our prediction that poor FRQ extends the time held for value stocks, but does not
extend (and appears to even reduce) the holding time for past winners.14 This analysis further corroborates our holding cost

13
While we argue that poor reporting quality is not likely to increase the holding period for past winners, it is useful to describe conditions where it could
decrease the holding period for these positions. Following Zhang (2006), we expect that the uncertainty due to poor reporting quality might make a
momentum long position more profitable, and that those profits would likely be realized in the short-term given the inherent short-term nature of
momentum (this is also consistent with the evidence in Table 10). In addition, Bushee and Goodman (2007) indicate that investors with a large position
are more likely to sell following positive performance. Taken together, if past winners have greater short-term performance when FRQ is poor, then it
could lead to a shorter holding period.
14
One possible explanation for why poor FRQ reduces holding time for past winners is that poor FRQ motivates short-term investors to buy past winners
and, thus, herding toward these stocks over time increases prices enough that some of the institutions that created bets opt to close out their position.

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 99

interpretation of the results presented in Table 4: poor FRQ alters the expected holding period for value and momentum firms,
which influences whether transient investors are willing to create positions in these stocks.15
As an additional sensitivity analysis, we calculate THELD for both transient (THELD_TRA) and non-transient investors
(THELD_NONTRA) at each firm. This analysis imposes additional constraints on the sample, as it requires at least one transient
bet to calculate THELD_TRA.16 In addition, this analysis could be influenced by the self-selection of transient investors to take
bets only when the expected holding period is likely to be short, which would reduce the power of our tests.17 In this analysis
(untabulated), we observe that the degree to which poor FRQ extends the holding period of value positions is driven by the time
held by non-transient institutions, while the degree to which poor FRQ reduces the holding period for momentum stocks is
driven by the positions held by transient investors.

Institution-Quarter-Level Analysis
The preceding tests are based on a firm-quarter analysis of large bet positions. While this analysis provides insight into
individual positions, it is also useful to consider the how the aggregation of these individual decisions shapes an institution’s
portfolio. Thus, we examine an institution’s portfolio tilt. Specifically, we measure each institution’s decision to tilt their
portfolio toward poor FRQ firms as they pursue a given trading strategy each quarter, and whether these tilts depend on their
investment horizon.
Following Ali, Chen, Yao, and Yu (2008), we measure the extent to which an institution is tilted toward a particular firm
characteristic by ranking that characteristic across all firms and then calculating the weighted average of that characteristic
across the firms in the institution’s portfolio. We partition the institution’s portfolio into VALUE and other firms for each
sample, and then we calculate the weighted average of RAQ (which is on a [0,9] scale). We define the difference between these
weighted averages as a measure of the degree to which an institution has tilted toward poor FRQ firms in exploiting a value
strategy (AQ_TILTVALUE ). We follow a parallel process splitting the sample based on PAST_WINNER to calculate the degree to
which an institution has tilted toward poor FRQ firms in exploiting a momentum strategy (AQ_TILTPAST_WINNER). These
definitions are more formally stated in Appendix A.
For all control variables, we also calculate the degree to which an institution has tilted their overall portfolio toward a given
firm characteristic. For this analysis, we do not impose the limit of 50 institutions per firm-year; however, we do require the
ability to calculate AQ_TILTVALUE and AQ_TILTPAST_WINNER, and that an institution has at least ten portfolio positions with
data for our variables of interest and control variables during the calendar quarter where institution-quarter variables are
measured. An additional benefit of the institution-quarter analysis is that we can include controls for an institution’s exposure to
prudent person regulations (measured using fiduciary type), which could influence an institution’s preference for poor FRQ.18
Table 6 presents summary statistics on the various measures of portfolio tilt. Both AQ_TILTVALUE and
AQ_TILTPAST_WINNER have an interquartile range greater than 1.25, suggesting that there is considerable cross-sectional
variation for these measures. Table 6 also provides descriptive statistics on the frequency of transient and dedicated investors in
the sample, as well as the control variables. The average value of BEME_TILT (PAST_RET_TILT) is below (above) 4.50,
consistent with the average institution placing smaller (larger) weights on value (momentum) stocks. The other descriptive
statistics suggest that there are strong institutional preferences for larger stocks, for stocks held by other institutions, and for
stocks with greater share turnover, less idiosyncratic risk, and better FRQ.
When considering the data from an institution-quarter perspective, H1 and H2 predict that transient institutions are less
likely to tilt their portfolios toward poor FRQ stocks when implementing a value strategy (lower values of AQ_TILTVALUE ), and
are more likely to tilt their portfolios toward poor FRQ stocks when implementing a momentum strategy (higher values of

15
We also estimate a version of Table 5 where we replace the indicator variables (POOR_AQ, VALUE, and PAST_WINNER) with their decile ranks
(RAQ, RBEME, RPAST_RET). In this specification, we continue to observe support for our predictions related to value (coefficient on RAQ  RBEME is
positive and significant). The results also continue to support H2 (coefficient on RAQ  RPAST_RET is negative and significant, and the coefficient on
RAQ  RPAST_RET is statistically less than the coefficient on RAQ  RBEME).
16
This sample constraint both reduces the number of observations to calculate the average value of THELD_TRA, resulting in a noisier estimate, and
reduces the number of firm-years for our firm-year sample, which reduces the power to detect statistical significance of the coefficients on our variable
of interest.
17
This argument suggests that when holding periods are long, transient institutions might opt to make smaller portfolio positions in a given stock (i.e.,
non-bet positions). To explore this possibility, we also examine the length of time transient investors keep non-bet positions in their portfolio (i.e.,
number of quarters they hold a positive portfolio weight in the future). We find that the poor FRQ significantly reduces this measure of holding period
for past winners, as in Table 5. In addition, we find a positive coefficient on the POOR_AQ  VALUE interaction, whose t-statistic is 1.49 (two-tailed p-
value of 13.8 percent).
18
In this analysis, we also include PRUDENT as an indicator for institutions whose fiduciary type is listed as a bank trust or pension (either corporate/
private or public). These institutions might prefer to reduce exposure to firms with poor FRQ because these firms could be viewed as too risky given
their fiduciary obligations, rather than for short-term trading strategy reasons.

The Accounting Review


Volume 94, Number 3, 2019
100 Bushee, Goodman, and Sunder

TABLE 6
Institution-Quarter Analysis of Portfolio Tilts
Standard Lower Upper
Mean Deviation Quartile Median Quartile
AQ_TILTVALUE 0.0150 1.6544 1.1190 0.1436 0.8996
AQ_TILTPAST_WINNER 0.2256 1.2439 0.5584 0.2110 0.9637
TRA 0.2991 0.4579 0.0000 0.0000 1.0000
DED 0.0435 0.2040 0.0000 0.0000 0.0000
AQ_TILT 2.9464 0.8621 2.3695 2.8012 3.3856
BEME_TILT 2.6008 1.0683 1.8764 2.5079 3.2044
PAST_RET_TILT 5.1678 1.0922 4.4017 5.1672 5.9189
MVE_TILT 8.1893 0.9105 7.9753 8.5166 8.7779
BETA_TILT 3.9018 1.2008 3.0206 3.8726 4.7010
IRISK_TILT 1.9887 1.2276 1.1125 1.6715 2.5486
TURN_TILT 5.1402 1.0818 4.3651 5.0706 5.8395
NUM_INST_TILT 8.2012 0.8999 7.9815 8.5277 8.7884
SUE_TILT 4.9049 0.8219 4.3772 4.9275 5.4408
ACC_TILT 4.2977 0.6731 3.8760 4.2600 4.6943
COV_TILT 7.8044 0.9258 7.4588 8.0847 8.4488
AGE_TILT 6.5216 1.2568 5.8560 6.8167 7.4355
NEARN_TILT 0.0646 0.0867 0.0082 0.0347 0.0840
LEV_TILT 4.5732 0.8991 4.0153 4.6454 5.1264
PRUDENT 0.1859 0.3890 0.0000 0.0000 0.0000
Table 6 presents descriptive statistics on the portfolio tilts for institution-quarters in our sample (n ¼ 153,340). The top and bottom 1 percent of all
continuous variables are winsorized.
All variable definitions are reported in Appendix A.

AQ_TILTPAST_WINNER). Table 7 provides univariate analysis comparing the mean values for transient and non-transient
institutions. The first row indicates that for transients, the mean RAQ for VALUE firms is lower than the mean for the other firms
in their portfolio (AQ_TILTVALUE ¼0.0669), while the opposite is true for non-transient investors AQ_TILTVALUE ¼ 0.0071).
The difference in these portfolio tilts across groups is significant (t ¼ 2.31). For AQ_TILTPAST_WINNER, we observe the
opposite pattern: transients are significantly more likely than non-transients to tilt toward poor FRQ stocks when buying
PAST_WINNER firms (t ¼ 2.75). Taken together, these univariate results are consistent with our predictions. Transient
institutions are significantly less likely to tilt toward poor FRQ stocks when pursuing a value strategy, but more likely to tilt
toward poor FRQ stocks when pursuing a momentum strategy. The table also summarizes a number of other differences in
portfolio tilts across transient and non-transient institutions that we include as control variables in our multivariate analysis.
As there are other differences across transient and non-transient institutions, we use the following regression model to examine
the association between an institution’s horizon and its decision to tilt toward poor FRQ stocks as it pursues a trading strategy:
AQ TILTSTRATEGY ¼ p0 þ p1 TRA þ p2 CONTROLS þ Quarter Fixed Effect þ Error
where: ð3Þ
AQ TILTSTRATEGY ¼ AQ TILTVALUE or AQ TILTPAST WINNER
In addition to the indicator for whether an institution is classified as a transient institution (TRA), Equation (3) also includes an
indicator for dedicated institutions (DED), classified based on Bushee (2001). The trading behavior of dedicated investors suggests
that they are not as sensitive to holding costs. In addition, Ramalingegowda (2014) provides evidence of dedicated investors selling
shares in a manner consistent with a private information advantage (i.e., selling before a bankruptcy). While it is possible that
dedicated investors might use value-based screens to identify companies, it is less likely that they would use momentum when
deciding to create a long-term position given the short-term nature of momentum.19
Table 8 provides evidence on the degree to which transients tilt toward poor FRQ stocks in their value and momentum
holdings. In column (1), where the dependent variable is AQ_TILTVALUE, the coefficient on TRA is negative and significant (t ¼

19
It is also possible that dedicated investors do not select firms based on value characteristics and instead seek to identify firms where they can increase
the value by reforming governance or adjusting corporate policies (e.g., changes in capital structure).

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 101

TABLE 7
Comparison of Portfolio Tilts: Transients versus Non-Transients
(1) (2)
n ¼ 45,859 n ¼ 107,481
TRA ¼ 1 TRA ¼ 0 (1)  (2) t-statistic
Mean Mean Difference Difference
AQ_TILTVALUE 0.0669 0.0071 0.0740** 2.31
AQ_TILTPAST_WINNER 0.2769 0.2037 0.0732*** 2.75
AQ_TILT 3.2303 2.8252 0.4051*** 16.32
BEME_TILT 2.6026 2.6001 0.0025 0.08
PAST_RET_TILT 5.3468 5.0915 0.2553*** 6.46
MVE_TILT 8.0196 8.2617 0.2421*** 9.27
BETA_TILT 4.3109 3.7273 0.5836*** 17.08
IRISK_TILT 2.5565 1.7464 0.8101*** 24.85
TURN_TILT 5.7811 4.8668 0.9143*** 32.39
NUM_INST_TILT 8.0273 8.2755 0.2482*** 9.68
SUE_TILT 4.8994 4.9073 0.0079 0.36
ACC_TILT 4.3165 4.2897 0.0268* 1.79
PRUDENT 0.0710 0.2349 0.1639*** 14.04
AGE_TILT 6.0012 6.7436 0.7424*** 21.19
COV_TILT 7.6593 7.8663 0.2070*** 8.31
NEARN_TILT 0.0913 0.0532 0.0381*** 12.81
LEV_TILT 4.6662 4.5335 0.1327*** 4.78
*, **, *** Significantly different from zero at 0.10, 0.05, and 0.01, respectively, using a two-tailed test.
Table 7 presents a univariate comparison of portfolio tilts by institutions classified as transients with those not classified as transients. The top and bottom 1
percent of all continuous variables are winsorized. t-statistics are calculated using two-way clustering based on institution and quarter.
All variable definitions are reported in Appendix A.

3.96). This negative coefficient suggests that transient institutions place a smaller weight on value firms with higher RAQ
scores in their portfolio. In contrast, the coefficient on DED is positive and significant (t ¼ 2.57), indicating that dedicated
investors are inclined to place larger weights on value firms with larger values of RAQ. This evidence is consistent with H1:
transient institutions are less likely to tilt their portfolios toward poor FRQ value stocks.
Column (2) of Table 8 provides a parallel analysis examining the degree to which institutions tilt toward poor FRQ stocks
when they pursue a momentum strategy. The coefficient on TRA is positive and significant (t ¼ 5.25), indicating that transient
institutions are more likely to tilt toward winners with poor FRQ. The coefficient on DED is negative and insignificant (t ¼
1.30), although this test likely lacks power given that dedicated investors would be unlikely to pursue a momentum strategy,
as that strategy would require frequent rebalancing.
An alternative measure of an institution’s sensitivity to holding costs is whether an institution is a hedge fund. Hedge funds
have greater restrictions on the ability of investors to withdraw their funds. These restrictions can help the hedge funds manage
illiquid investments (e.g., Aragon 2007), thereby reducing the tensions in the Shleifer and Vishny (1997) theoretical model that
lead to a short-term focus.20
To examine differences between hedge funds and mutual funds, we construct a subsample from our overall institutional
investor sample that is composed of only hedge funds and mutual funds. We create an indicator variable for institutions classified
as hedge funds (HF) based on Agarwal, Fos, and Jiang (2013).21 In Table 9, we repeat the Table 8 analysis with a comparison of
hedge funds to mutual funds. Column (1) reveals a positive and significant coefficient on HF when AQ_TILTVALUE is the
dependent variable (t ¼ 2.89), indicating that hedge funds are more likely to place larger weights on firms with poor FRQ when
selecting value firms. This evidence is consistent with share restrictions limiting the degree to which institutions avoid poor FRQ
value stocks. In column (2), where AQ_TILTPAST_WINNER is the dependent variable, the coefficient on HF is negative, but not

20
Recent work by Agarwal, Vashishtha, and Venkatachalam (2018) provides evidence that regulations that increase transparency regarding fund holdings
can increase the short-term focus of fund managers, which could have negative consequences for corporate innovation. Agarwal et al. (2018) note that
hedge funds are distinct from mutual funds due to the lack of required disclosure by most hedge funds.
21
We greatly appreciate Wei Jiang’s willingness to share hedge fund data. The data used in our analysis is a list that has been updated to include
observations through 2012.

The Accounting Review


Volume 94, Number 3, 2019
102 Bushee, Goodman, and Sunder

TABLE 8
Institutional Investor Preferences for Financial Reporting Quality within a Trading Strategy
(1) (2)
Dep. Var. ¼ Dep. Var. ¼
AQ_TILTVALUE AQ_TILTPAST_WINNER (1)  (2)
TRA 0.0758*** 0.0700*** 0.1457***
(3.96) (5.25) (5.83)
DED 0.1290** 0.0384 0.1675***
(2.57) (1.30) (2.75)
BEME_TILT 0.0765*** 0.0269 0.0496
(3.66) (1.51) (1.57)
PAST_RET_TILT 0.0793*** 0.0051 0.0844**
(4.67) (0.31) (3.33)
AQ_TILT 0.4476*** 0.0108 0.4583***
(18.18) (0.55) (13.15)
MVE_TILT 0.1976** 0.0223 0.1752*
(2.48) (0.37) (1.73)
BETA_TILT 0.0150 0.0364* 0.0514*
(0.75) (1.82) (1.69)
IRISK_TILT 0.1013*** 0.0288 0.0726*
(3.55) (0.96) (1.83)
TURN_TILT 0.0305 0.0395* 0.0090
(1.23) (1.71) (0.29)
NUM_INST_TILT 0.0067 0.0664 0.0597
(0.07) (1.01) (0.53)
SUE_TILT 0.0379*** 0.0141 0.0237
(3.02) (0.86) (1.13)
ACC_TILT 0.0065 0.0471*** 0.0535**
(0.36) (2.81) (2.12)
PRUDENT 0.1609*** 0.0040 0.1569***
(5.88) (0.26) (4.93)
AGE_TILT 0.0044 0.0133 0.0088
(0.23) (0.77) (0.30)
COV_TILT 0.0504 0.0213 0.0717*
(1.49) (0.94) (1.66)
LEV_TILT 0.0983*** 0.0028 0.1011***
(5.85) (0.19) (4.34)
NEARN_TILT 0.5113*** 0.0982 0.4131
(3.02) (0.64) (1.61)
Quarter Fixed Effects Yes Yes Yes
Clustering based on Institution Yes Yes Yes
and Quarter
n 153,340 153,340 153,340
*, **, *** Significantly different from zero at 0.10, 0.05, and 0.01, respectively, using a two-tailed test.
Table 8 presents estimates of a model predicting the degree to which institutions tilt their portfolios toward poor FRQ firms when pursuing a value or
momentum strategy as a function of institutional investor horizon (TRA) and control variables (Equation (3)). The top and bottom 1 percent of all
continuous variables are winsorized. Within each column are the estimated coefficients, with its t-statistic beneath it in parentheses. t-statistics are
calculated using two-way clustering based on institution and quarter.
All variable definitions are reported in Appendix A.

significant at conventional levels (t ¼ 1.50). In summary, our analysis contrasting hedge funds and mutual funds provides
evidence supporting our results related to short-term investors tilting away from value stocks with poor FRQ.

Mutual Fund-Level Analysis


Next, we use mutual fund-level data to examine additional measures that provide insight into why an institution would
have a short-term horizon, albeit on a more limited set of intuitional investors than in the institution-level analysis. First, using

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 103

TABLE 9
Hedge Fund Preferences for Financial Reporting Quality within a Strategy Relative to Mutual Funds
(1) (2)
Dep. Var. ¼ Dep. Var. ¼
AQ_TILTVALUE AQ_TILTPAST_WINNER (1)  (2)
HF 0.0866*** 0.0261 0.1128***
(2.89) (1.50) (3.13)
BEME_TILT 0.0966*** 0.0132 0.0835***
(4.36) (0.71) (2.63)
PAST_RET_TILT 0.0892*** 0.0177 0.1068***
(5.17) (1.08) (4.27)
AQ_TILT 0.4431*** 0.0552** 0.4983***
(17.12) (2.58) (12.86)
MVE_TILT 0.1561* 0.0163 0.1724
(1.83) (0.24) (1.49)
BETA_TILT 0.0353* 0.0250 0.0603*
(1.70) (1.25) (1.93)
IRISK_TILT 0.0796** 0.0025 0.0821*
(2.47) (0.08) (1.91)
TURN_TILT 0.0463* 0.0245 0.0218
(1.77) (0.97) (0.69)
NUM_INST_TILT 0.0611 0.0170 0.0780
(0.63) (0.22) (0.61)
SUE_TILT 0.0343** 0.0224 0.0119
(2.46) (1.30) (0.53)
ACC_TILT 0.0169 0.0483*** 0.0652**
(0.85) (2.66) (2.34)
AGE_TILT 0.0128 0.0022 0.0150
(0.62) (0.12) (0.49)
COV_TILT 0.0517 0.0154 0.0671
(1.30) (0.60) (1.32)
LEV_TILT 0.1087*** 0.0018 0.1068***
(6.01) (0.12) (4.19)
NEARN_TILT 0.5467*** 0.0403 0.5064**
(3.09) (0.27) (2.07)
Quarter Fixed Effects Yes Yes Yes
Clustering based on Institution Yes Yes Yes
and Quarter
n 104,298 104,298 104,298
*, **, *** Significantly different from zero at 0.10, 0.05, and 0.01, respectively, using a two-tailed test.
Table 9 presents estimates of a model predicting the degree to which institutions tilt their portfolios toward poor FRQ firms when pursuing a value or
momentum strategy as a function of whether an institution is classified as a hedge fund (HF) and control variables (Equation (3)). The analysis is based on
a subsample of the firms in the institutional investor-level analysis: those classified as hedge funds and those classified as mutual funds. The top and
bottom 1 percent of all continuous variables are winsorized. Within each column are the estimated coefficients, with its t-statistic beneath it in parentheses.
t-statistics are calculated using two-way clustering based on institution and quarter.
All variable definitions are reported in Appendix A.

an approach similar to Sirri and Tufano (1998), we estimate the degree to which fund flows are sensitive to low performance,
assuming that fund managers respond to a greater sensitivity by choosing a shorter investment horizon.22 Second, we estimate
the age of a mutual fund, following the assumption that the chance of withdrawals when performance is poor is greater for fund

22
We estimate a regression of fund flow in a given month on fund performance, where the coefficient on performance is allowed to vary across high and
low performance. The regression also includes controls for fund size, past fund flow, and the average market-wide level of fund flow. We also consider
an alternative measure of the risk of large withdrawals by calculating the frequency of large withdrawals over the past year.

The Accounting Review


Volume 94, Number 3, 2019
104 Bushee, Goodman, and Sunder

managers with shorter track records, leading the fund manager to choose a short-term horizon.23 We collect data from the
Thomson Financial mutual fund holdings database, and fund returns and fund flow data from the CRSP mutual fund database.
We link these datasets using the Mutual Fund Links (MFLINKS) dataset. Using mutual fund holdings data, we calculate
mutual fund versions of the portfolio tilt measures in Equation (3).
We estimate whether fund flow sensitivity and fund age are associated with the degree to which a fund tilts toward poor
FRQ stocks when it undertakes value and momentum strategies (untabulated). We expect that the younger (older) funds and the
funds whose flows are more (less) sensitive to negative performance will act in a manner that parallels the transient (dedicated)
institutions. We find that institutions that are less sensitive to poor performance are more likely to choose poor FRQ stocks
when they follow a value strategy. However, we do not see any significant associations between this fund flow sensitivity
measure and a fund’s portfolio’s tilt toward poor FRQ firms when following momentum strategy. These results parallel the
results in our analysis of dedicated institutions in Table 8. We also find evidence that younger funds are less (more) likely to tilt
toward poor FRQ stocks when following a value (momentum) strategy. These results parallel the results for transient
institutions documented in Table 8.

Financial Reporting Quality and the Timing of Returns


Our work is motivated by the idea that poor FRQ might increase the benefits (expected returns) from value and momentum
strategies, but the timing of these benefits would differ across strategies because the benefits to the momentum strategy are realized
more quickly. To show how the timing of the benefits does differ across strategies, we estimate the following model in which the
association between future stock returns and value (RBEME) and momentum (RPAST_RET) signals is allowed to vary with FRQ:24,25
SARðmÞ ¼ d0 þ d1 RAQ  RBEME þ d2 RAQ  RPAST RET þ d3 RBEME þ d4 PAST RET þ d5 RAQ þ d6 RIRISK
þ d7 RIRISK  RBEME þ d8 RIRISK  RPAST RET þ d9 RMVE þ d10 RAQ  RMVE þ d11 RIRISK  RMVE
þ d12 RBETA þ d13 RAQ  RBETA þ d14 RIRISK  RBETA þ d15 RMVE  RBEME þ d16 RMVE  RPAST RET
þ Month Fixed Effect þ Industry Fixed Effect þ Error
ð4Þ
The dependent variable, SAR(m), is the buy-and-hold size-adjusted abnormal return over the following m months (m ranges from 1 to
24).26,27 We limit our sample to NYSE/AMEX/NASDAQ common stocks traded between 1973 and 2013.28 We also require non-
missing CRSP and Compustat data for all variables in our regression. The independent variables in Equation (4) are based on scaled
decile ranks, where rankings are calculated monthly.
Last, we drop financial services firms and observations where the beginning monthly price per share is less than $5. Panel
A and Panel B of Table 10 present estimates of Equation (4) across different horizons (m ¼ 3,. . .,24 by 3).29

23
We estimate the age of the fund using the length of time a fund is in the CRSP mutual fund dataset since 1980.
24
As existing research documents that the performance of value strategies varies with idiosyncratic risk (Ali et al. 2003), we also allow the association
between our independent variables to vary with idiosyncratic risk (RIRISK). In addition, we allow the coefficient on the value momentum signals to
vary with firm size (RMVE). Given this structure of interactions, the coefficient on the main effect of a trading strategy (RPAST_RET or RBEME)
variable reflects the return to that strategy when RIRISK and RAQ are equal to 0 (low volatility firms with good accounting quality), and when RMVE is
equal to 0 (small firms that tend to have higher risk and poorer accounting quality). Thus, the coefficients on the main effects are difficult to interpret.
25
Unlike the analysis using institutional investors, which uses indicator variables for VALUE and PAST_WINNER to reflect the fact that we only observe
institutions’ long positions, we use scaled decile ranks for the stock return tests. This approach is consistent with prior research on accounting signals that
predict stock returns (e.g., Abarbanell and Bushee 1998; Mashruwala, Rajgopal, and Shevlin 2006), and the finer partitioning in the ranks relative to an
indicator variable adds power to the analysis, which is valuable given the volatility of stock returns. When we replicate Table 10 replacing the scaled decile
ranks of RBEME, RPAST_RET, and RAQ with indicator variables, we lose significance for individual coefficients, consistent with a loss of power. However,
the trend in the coefficient with respect to the horizon remains the same: as the horizon of the dependent variable (m) increases, the coefficient on POOR_AQ
 VALUE increases, while the coefficient on POOR_AQ  PAST_WINNER peaks and then decreases, which is consistent with the motivation for these tests.
26
Following previous research (e.g., Abarbanell and Bushee 1998), we use the delisting return when available, but when it is missing, we set the delisting
return equal to 100 percent when the delisting return is missing and the delisting code is between 400 and 700. For the size adjustment, we use size
breakpoints based on the distribution of NYSE firms and apply them to all common stocks on the NYSE, AMEX, and NASDAQ to calculate value-
weighted size portfolios for use as expected returns.
27
We also estimate a parallel version of Equation (4) where we substitute other uncertainty measures from the literature (e.g., Zhang 2006), such as firm
age or analyst coverage, for IRISK in Equation (4) and obtain similar results.
28
If we constrain our sample to start in 1981, when the institutional investor data become available, then the results become weaker for the momentum
effect (t-statistic ¼ 1.79 for the coefficient on RAQ  RPAST_RET). However, the correlation between the estimated coefficients of interest and the
horizon of the test remains the same: the return horizon (m) is positively correlated with the coefficient on RAQ  RBEME, but insignificantly correlated
with the coefficient on RAQ  RPAST_RET.
29
For consistency with the other tests in the paper, the standard errors for each model are estimated using two-way clustering based on firm and time.
However, these results are qualitatively similar if we remove the month fixed effect and estimate the regressions using monthly Fama-MacBeth
regressions with a Newey-West adjusted standard error.

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 105

TABLE 10
Future Stock Returns by Financial Reporting Quality and Trading Strategy

Panel A: Analysis of Buy-and-Hold Returns Over Different Horizons Using Monthly Returns (m ¼ 3–12)
m¼3 m¼6 m¼9 m ¼ 12
RAQ  RBEME 0.0043 0.0080 0.0216 0.0357
(0.79) (0.77) (1.35) (1.65)
RAQ  RPAST_RET 0.0121** 0.0149 0.0207 0.0246
(2.37) (1.58) (1.53) (1.42)
RBEME 0.0055 0.0104 0.0207 0.0298
(0.84) (0.85) (1.17) (1.25)
RPAST_RET 0.0190*** 0.0388*** 0.0438*** 0.0493**
(2.77) (3.35) (2.67) (2.28)
RAQ 0.0100 0.0140 0.0267 0.0412
(1.60) (1.17) (1.49) (1.73)
RMVE 0.0113 0.0253** 0.0493*** 0.0723***
(1.48) (1.97) (2.75) (3.00)
RBETA 0.0050 0.0113 0.019* 0.0232*
(1.19) (1.51) (1.79) (1.66)
RAQ  RMVE 0.0026 0.0052 0.0155 0.0252
(0.51) (0.51) (0.99) (1.18)
RAQ  RBETA 0.0099** 0.0221** 0.0376** 0.0476**
(1.98) (2.22) (2.51) (2.35)
RIRISK 0.0278*** 0.0434*** 0.0477** 0.0503
(3.18) (2.88) (2.13) (1.63)
RIRISK  RMVE 0.0137** 0.0275** 0.0366** 0.0496**
(2.28) (2.33) (2.05) (2.05)
RIRISK  RBEME 0.0323*** 0.0630*** 0.0900*** 0.1178***
(4.18) (4.54) (4.49) (4.35)
RIRISK  RPAST_RET 0.0146** 0.0129 0.0023 0.0230
(2.10) (1.07) (0.13) (1.00)
RIRISK  RBETA 0.0051 0.0124 0.0203 0.0282
(0.77) (0.97) (1.06) (1.10)
RMVE  RBEME 0.0163 0.0280** 0.0396** 0.0498**
(2.38) (2.26) (2.20) (2.05)
RMVE  RPAST_RET 0.0159** 0.0250** 0.0213 0.0211
(2.17) (2.09) (1.25) (0.98)
Month Fixed Effect Yes Yes Yes Yes
Industry Fixed Effect Yes Yes Yes Yes
n (firm-months) 734,182 734,182 734,182 734,182
(continued on next page)

Panels A and B of Table 10 provide insight into how financial reporting quality influences the way value and momentum
signals map into returns and how that mapping changes over different horizons. We observe that the coefficient on RAQ 
RBEME increases monotonically over time, achieving significance at the 5 percent level only when the horizon reaches 15
months. In contrast, the coefficient on RAQ  RPAST_RET is only positive and significant during the first three-month window.
This result shows that the timing of the benefits of poor FRQ for returns to the value strategy are spread out over the two-year
range, while the benefits of poor FRQ for the momentum strategy are realized much more quickly.30
Figure 1 provides a graph of the coefficients of interest across these different horizons (m ¼ 1,. . .,24). There is a clear
increasing trend in the coefficients on RAQ  RBEME. However, the coefficient on RAQ  RPAST_RET plateaus quickly and, in

30
It is also noteworthy that the coefficients are not significantly negative over any of the horizons in Panels A and B. This alleviates the concern that poor
accounting quality might reduce the returns from momentum or value strategies, which could make the strategy less desirable for any institutional
investor regardless of their horizon.

The Accounting Review


Volume 94, Number 3, 2019
106 Bushee, Goodman, and Sunder

TABLE 10 (continued)
Panel B: Analysis of Buy-and-Hold Returns Over Different Horizons Using Monthly Returns (m ¼ 1524)
m ¼ 15 m ¼ 18 m ¼ 21 m ¼ 24
RAQ  RBEME 0.0526* 0.0718** 0.0865** 0.1054**
(1.92) (2.17) (2.23) (2.35)
RAQ  RPAST_RET 0.0274 0.0296 0.0208 0.0068
(1.31) (1.21) (0.74) (0.22)
RBEME 0.0421 0.0458 0.0448 0.0457
(1.40) (1.27) (1.06) (0.94)
RPAST_RET 0.0602** 0.0843*** 0.1174*** 0.1409***
(2.35) (2.82) (3.30) (3.61)
RAQ 0.0557* 0.0761** 0.0877** 0.1005
(1.86) (2.11) (2.09) (2.09)
RMVE 0.095*** 0.1065*** 0.1129*** 0.1239**
(3.13) (2.95) (2.71) (2.61)
RBETA 0.0257 0.0248 0.0250 0.0276
(1.48) (1.18) (1.02) (0.99)
RAQ  RMVE 0.0328 0.0443 0.0517 0.0628
(1.22) (1.35) (1.35) (1.41)
RAQ  RBETA 0.0568** 0.0573* 0.0540 0.0499
(2.25) (1.88) (1.52) (1.23)
RIRISK 0.0626 0.0629 0.0708 0.0853
(1.61) (1.34) (1.35) (1.44)
RIRISK  RMVE 0.0693** 0.0876** 0.1148** 0.1397***
(2.24) (2.30) (2.55) (2.69)
RIRISK  RBEME 0.1529*** 0.1819*** 0.2088*** 0.2348***
(4.48) (4.46) (4.45) (4.34)
RIRISK  RPAST_RET 0.0423 0.0716** 0.0921** 0.0967**
(1.51) (2.10) (2.33) (2.22)
RIRISK  RBETA 0.0322 0.0390 0.0555 0.0755
(1.00) (1.01) (1.26) (1.53)
RMVE  RBEME 0.0608** 0.0594 0.0503 0.0430
(1.98) (1.62) (1.16) (0.86)
RMVE  RPAST_RET 0.0245 0.0400 0.0687** 0.0921**
(0.97) (1.37) (2.01) (2.44)
Month Fixed Effect Yes Yes Yes Yes
Industry Fixed Effect Yes Yes Yes Yes
n (firm-months) 734,182 734,182 734,182 734,182
(continued on next page)

fact, decreases after 18 months.31 Overall, this evidence is consistent with our expectation that poor financial reporting quality
is related to both value and momentum strategies; however, the benefits to the value strategy accrue over a longer horizon.32
To complement our analysis of long-window returns, we also examine whether FRQ influences the link between
momentum and value strategies and the returns at subsequent earnings announcements. For this analysis, we estimate a version

31
While the coefficient increases up to 18 months, it is only significant up to three months. This is likely due to the increase in the scale of the dependent
variable over time due to compounding. For example, if a momentum strategy earns positive SAR during the first three months and earns zero SAR over
subsequent months, then the SAR over the full window could still be larger than the return in the first months, as the profits from those initial returns are
amplified by the expected returns due to the compounding. To investigate this empirically, we estimate a version of Table 10 that uses returns based on
non-overlapping three-month windows. In this specification, the coefficient on RAQ  RPAST_RET is only positive and significant for the shortest
horizon.
32
The regression analysis provides insight into how FRQ influences the mapping of value and momentum signals into returns, holding all other variables
constant. Another approach could be to estimate a regression predicting SAR(m) as a function of RBEME, RPAST_RET, RMVE, and RBETA by groups
split based on RAQ. We perform this supplemental analysis that confirms findings in Table 10: the association between RBEME and SAR(m) increases
over time more for poor FRQ firms relative to good FRQ firms, and the association between RPAST_RET and SAR(m) is more positive over shorter
windows for poor FRQ firms relative to good FRQ firms, although this difference converges to zero as the horizon increases.

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 107

TABLE 10 (continued)
Panel C: Analysis of Earnings Announcement Returns
CARðqÞ ¼ d0 þ d1 RAQ  RBEME þ d2 RAQ  RPAST RET þ d3 RBEME þ d4 PAST RET þ d5 RAQ þ d6 RIRISK
þ d7 RIRISK  RBEME þ d8 RIRISK  RPAST RET þ d9 RMVE þ d10 RAQ  RMVE þ d11 RIRISK  RMVE
þ d12 RBETA þ d13 RAQ  RBETA þ d14 RIRISK  RBETA þ d15 RMVE  RBEME þ d16 RMVE  RPAST RET
þ Month Fixed Effect þ Industry Fixed Effect þ Error

CAR(þ1) CAR(þ2) CAR(þ3) CAR(þ4)


RAQ  RBEME 0.0042** 0.0044** 0.0052** 0.0047**
(1.99) (2.01) (2.35) (2.10)
RAQ  RPAST_RET 0.0040** 0.0022 0.0006 0.0012
(2.03) (1.11) (0.30) (0.61)
RBEME 0.0049** 0.0042* 0.0033 0.0035
(2.11) (1.84) (1.41) (1.46)
RPAST_RET 0.0096*** 0.0026 0.0015 0.0035
(4.53) (1.29) (0.72) (1.59)
RAQ 0.0041 0.0053** 0.0051* 0.0053**
(1.62) (2.04) (1.95) (2.00)
RMVE 0.0033 0.0032 0.0065*** 0.0057**
(1.38) (1.38) (2.77) (2.38)
RBETA 0.0036*** 0.0025* 0.0036*** 0.0035**
(2.78) (1.89) (2.63) (2.53)
RAQ  RMVE 0.0010 0.0020 0.0028 0.0017
(0.48) (0.91) (1.28) (0.79)
RAQ  RBETA 0.0009 0.0008 0.0001 0.0010
(0.48) (0.39) (0.04) (0.50)
RIRISK 0.0028 0.0004 0.0028 0.0016
(1.07) (0.14) (1.01) (0.57)
RIRISK  RMVE 0.0083*** 0.0067*** 0.0081*** 0.0059**
(3.79) (2.99) (3.66) (2.56)
RIRISK  RBEME 0.0021 0.0011 0.0025 0.0028
(0.88) (0.45) (1.02) (1.08)
RIRISK  RPAST_RET 0.0027 0.0032 0.0020 0.0020
(1.19) (1.44) (0.87) (0.84)
RIRISK  RBETA 0.0056*** 0.0062*** 0.0067*** 0.0068***
(2.69) (2.91) (2.97) (3.11)
RMVE  RBEME 0.0082*** 0.0071*** 0.0066*** 0.0067***
(3.51) (3.08) (2.82) (2.84)
RMVE  RPAST_RET 0.0083*** 0.0032 0.0076*** 0.0086***
(3.72) (1.44) (3.46) (3.92)
Month Fixed Effect Yes Yes Yes Yes
Industry Fixed Effect Yes Yes Yes Yes
n (firm-months) 613,697 613,697 613,697 613,697
*, **, *** Significantly different from zero at 0.10, 0.05, and 0.01, respectively, using a two-tailed test.
Table 10 presents evidence on how the association between future stock returns and value and momentum signals varies with FRQ (Equation (4)). Panels
A and B provide an analysis of buy-and-hold returns calculated over different horizons. Each column presents a version of equation over a given horizon
(e.g., m ¼ 3 implies a three-month size horizon used to calculate size-adjusted returns). Panel C provides an analysis of the cumulative abnormal returns at
future earnings announcements. Each column uses the earning announcement returns that follow the portfolio formation period (e.g., q ¼ 1 is the return at
the first earnings announcement to follow portfolio formation, used for the dependent variable). For the stock return tests, ranks are calculated each month.
In addition, all ranked variables are transformed into scaled decile ranks ([0,1]) to facilitate their interpretation as hedge returns. Within each column are
the estimated coefficients, with its t-statistic beneath it in parentheses. t-statistics are calculated using two-way clustering based on firm and quarter.
All independent variable definitions are reported in Appendix A.

The Accounting Review


Volume 94, Number 3, 2019
108 Bushee, Goodman, and Sunder

FIGURE 1
Coefficients on the Interaction Terms in Equation (4) and the Return Measurement Window

Figure 1 plots the average interaction coefficients from Equation (4). The X-axis is the number of months (m) of the dependent variable in the model
(SAR(m)). SAR(m) is equal to size-adjusted abnormal return over the following m months.
All other variable definitions are reported in Appendix A.

of Equation (4) where the dependent variable is the cumulative abnormal return at an earnings announcement after portfolio
formation.33 Panel C of Table 10 presents these results. The coefficient on RAQ  RBEME is positively associated with the
announcement returns during each of the subsequent four quarters.34 The coefficient on RAQ  RPAST_RET is also positive,
but only significantly different from zero for the announcement return that immediately follows portfolio formation. These
results suggest that investors purchasing value stocks with poor accounting quality would need to hold their positions for
multiple quarters to earn the full returns from that strategy, while investors purchasing past winners with poor accounting

33
The daily abnormal return is a firm’s raw return less the return on the value-weighted market return. These daily abnormal returns are then accumulated
over the (2,1) window.
34
Comparing the results from Panels A, B, and C of Table 10 indicates that the coefficient on RAQ  RBEME in the regression predicting CAR(þ1) is
similar in magnitude to the coefficient on RAQ  RBEME in the regression predicting SAR(3). This implies that that a large proportion of the benefits
from investing in value firms with poor accounting quality over the first three months is realized at the earnings announcement. However, the
coefficient on RAQ  RPAST_WINNER in the regression predicting CAR(þ1) is noticeably smaller than the coefficient on RAQ  RPAST in the
regression predicting SAR(3), suggesting that a smaller portion of the benefits from poor accounting quality arises at earnings announcements.

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 109

quality would not need to hold their position longer than one quarter.35 Similar to Table 10, these supplemental results further
support our prediction that the benefits of poor FRQ are realized quickly for a momentum strategy, while the benefits for a
value strategy occur over a longer period of time.

V. CONCLUSIONS
We examine whether poor FRQ creates holding costs that influence which institutions create large positions in particular
stocks and how long those positions are held. We find that FRQ is associated with whether transient institutions create large
positions in certain stocks. Specifically, when FRQ is poor, transient institutions are less likely to place a large weight on value
stocks and more likely to place a large weight on momentum stocks. The decision by short-term institutions to shift away from
value and toward momentum among poor FRQ firms suggests that expected holding costs play a role in how these short-term
institutions construct their portfolio. To further explore holding costs, we examine how long large positions are held and find
that poor FRQ increases the holding period for positions in value stocks, but does not increase the holding period for positions
in momentum stocks.
Our analysis provides insight into the importance of holding costs, as longer expected holding periods can discourage
short-term institutions from following particular strategies. Thus, while poor FRQ can increase the potential returns to a trading
strategy, these benefits can be mitigated by higher holdings costs in those strategies that rely on financial reporting numbers for
price corrections. This evidence highlights another cost of poor FRQ in that it can influence the willingness of sophisticated,
short-term investors to take large positions and correct mispricing.

REFERENCES
Abarbanell, J., and B. Bushee. 1998. Abnormal returns to a fundamental analysis strategy. The Accounting Review 73 (1): 19–45.
Agarwal, V., V. Fos, and W. Jiang. 2013. Inferring reporting biases in hedge fund databases from hedge fund equity holdings.
Management Science 59 (6): 1271–1289. https://doi.org/10.1287/mnsc.1120.1647
Agarwal, V., R. Vashishtha, and M. Venkatachalam. 2018. Mutual fund transparency and corporate myopia. Review of Financial Studies
31 (5): 1966–2003. https://doi.org/10.1093/rfs/hhx125
Ali, A., L. Hwang, and M. Trombley. 2003. Arbitrage risk and the book-to-market anomaly. Journal of Financial Economics 69 (2): 355–
373. https://doi.org/10.1016/S0304-405X(03)00116-8
Ali, A., X. Chen, T. Yao, and T. Yu. 2008. Do mutual funds profit from the accruals anomaly? Journal of Accounting Research 46 (1): 1–
26. https://doi.org/10.1111/j.1475-679X.2007.00263.x
Aragon, G. 2007. Share restrictions and asset pricing: Evidence from the hedge fund industry. Journal of Financial Economics 83 (1): 33–
58. https://doi.org/10.1016/j.jfineco.2005.11.001
Asness, C., T. Moskowitz, and L. Pedersen. 2013. Value and momentum everywhere. Journal of Finance 68 (3): 929–985. https://doi.
org/10.1111/jofi.12021
Bushee, B. 2001. Do institutional investors prefer near-term earnings over long-run value? Contemporary Accounting Research 18 (2):
207–246. https://doi.org/10.1506/J4GU-BHWH-8HME-LE0X
Bushee, B., and T. Goodman. 2007. Which institutions trade based on private information about earnings and returns? Journal of
Accounting Research 45 (2): 289–321. https://doi.org/10.1111/j.1475-679X.2007.00234.x
Bushee, B., and C. Noe. 2000. Corporate disclosure practices, institutional investors, and stock return volatility. Journal of Accounting
Research 38 (Supplement): 171–202. https://doi.org/10.2307/2672914
Callen, J., M. Khan, and H. Lu. 2013. Accounting quality, stock price delay, and future stock returns. Contemporary Accounting Research
30 (1): 269–295. https://doi.org/10.1111/j.1911-3846.2011.01154.x
Chan, L., N. Jegadeesh, and J. Lakonishok. 1996. Momentum strategies. Journal of Finance 51 (5): 1681–1713. https://doi.org/10.1111/j.
1540-6261.1996.tb05222.x
Chordia, T., and L. Shivakumar. 2006. Earnings and price momentum. Journal of Financial Economics 80 (3): 627–656. https://doi.org/
10.1016/j.jfineco.2005.05.005
Collins, D., G. Gong, and P. Hribar. 2003. Investor sophistication and the mispricing of accruals. Review of Accounting Studies 8 (2/3):
251–276. https://doi.org/10.1023/A:1024417513085
Core, J., W. Guay, and R. Verdi. 2008. Is accruals quality a priced risk factor? Journal of Accounting and Economics 46 (1): 2–22. https://
doi.org/10.1016/j.jacceco.2007.08.001

35
While the magnitude of the coefficient on the interaction terms RAQ  RBEME and RAQ  RPAST_RET in models predicting the next earnings
announcement are similar (0.0042 versus 0.0040), the coefficient related to the value strategy represents a smaller proportion of the strategy’s total
returns at future earnings announcements. This proportion would be relevant for assessing holding costs if a short-term institution chooses to pursue a
strategy with the intention of capturing the full returns from that strategy. An institution is most likely to attempt to plan on earning the total strategy
return when that is necessary to offset the costs associated with the strategy.

The Accounting Review


Volume 94, Number 3, 2019
110 Bushee, Goodman, and Sunder

Da, Z., U. Gurun, and M. Warachka. 2014. Frog in the pan: Continuous information and momentum. Review of Financial Studies 27 (7):
2171–2218. https://doi.org/10.1093/rfs/hhu003
Dechow, P., and I. Dichev. 2002. The quality of accruals and earnings: The role of accrual estimation errors. The Accounting Review 77
(Supplement): 35–59. https://doi.org/10.2308/accr.2002.77.s-1.35
Dechow, P., W. Ge, and C. Schrand. 2010. Understanding earnings quality: A review of the proxies, their determinants and their
consequences. Journal of Accounting and Economics 50 (2/3): 344–401. https://doi.org/10.1016/j.jacceco.2010.09.001
Fama, E., and K. French. 1992. The cross-section of expected stock returns. Journal of Finance 47 (2): 427–465. https://doi.org/10.1111/
j.1540-6261.1992.tb04398.x
Fama, E., and K. French. 2008. Dissecting anomalies. Journal of Finance 63 (4): 1653–1678. https://doi.org/10.1111/j.1540-6261.2008.
01371.x
Francis, J., R. LaFond, P. Olsson, and K. Schipper. 2005. The market pricing of accruals quality. Journal of Accounting and Economics
39 (2): 295–327. https://doi.org/10.1016/j.jacceco.2004.06.003
Jegadeesh, N., and S. Titman. 1993. Returns to buying winners and selling losers: Implications for stock market efficiency. Journal of
Finance 48 (1): 65–91. https://doi.org/10.1111/j.1540-6261.1993.tb04702.x
Ke, B., and K. Petroni. 2004. How informed are actively trading institutional investors? Evidence from their trading behavior before a
break in a string of consecutive earnings increases. Journal of Accounting Research 42 (5): 895–927. https://doi.org/10.1111/j.
1475-679X.2004.00160.x
Ke, B., and S. Ramalingegowda. 2005. Do institutional investors exploit the post-earnings announcement drift? Journal of Accounting
and Economics 39 (1): 25–53. https://doi.org/10.1016/j.jacceco.2004.02.002
Ke, B., S. Ramalingegowda, and Y. Yu. 2006. The Effect of Investment Horizon on Institutional Investors’ Incentives to Acquire Private
Information on Long-Term Earnings. Working paper, The Pennsylvania State University, The University of Georgia, and The
University of Texas at Austin.
Kim, O., and R. Verrecchia. 1997. Pre-announcement and event-period private information. Journal of Accounting and Economics 24 (3):
395–419. https://doi.org/10.1016/S0165-4101(98)00013-5
Lakonishok, J., A. Shleifer, and R. Vishny. 1994. Contrarian investment, extrapolation, and risk. Journal of Finance 49 (5): 1541–1578.
https://doi.org/10.1111/j.1540-6261.1994.tb04772.x
La Porta, R., J. Lakonishok, A. Shleifer, and R. Vishny. 1997. Good news for value stocks: Further evidence on market efficiency.
Journal of Finance 52 (2): 859–874. https://doi.org/10.1111/j.1540-6261.1997.tb04825.x
Lev, B., and D. Nissim. 2006. The persistence of the accruals anomaly. Contemporary Accounting Research 23 (1): 193–226. https://doi.
org/10.1506/C6WA-Y05N-0038-CXTB
Maffett, M. 2012. Financial reporting opacity and informed trading by international institutional investors. Journal of Accounting and
Economics 54 (2/3): 201–220. https://doi.org/10.1016/j.jacceco.2012.09.002
Mashruwala, C., S. Rajgopal, and T. Shevlin. 2006. Why is the accrual anomaly not arbitraged away? The role of idiosyncratic risk and
transaction costs. Journal of Accounting and Economics 42 (1/2): 3–33. https://doi.org/10.1016/j.jacceco.2006.04.004
McNichols, M., and B. Trueman. 1994. Public disclosure, private information collection, and short-term trading. Journal of Accounting
and Economics 17 (1/2): 69–94. https://doi.org/10.1016/0165-4101(94)90005-1
Ng, J. 2011. The effect of information quality on liquidity risk. Journal of Accounting and Economics 52 (2/3): 126–143. https://doi.org/
10.1016/j.jacceco.2011.03.004
Pontiff, J. 2006. Costly arbitrage and the myth of idiosyncratic risk. Journal of Accounting and Economics 42 (1/2): 35–52. https://doi.
org/10.1016/j.jacceco.2006.04.002
Rajgopal, S., and M. Venkatachalam. 2011. Financial reporting quality and idiosyncratic return volatility. Journal of Accounting and
Economics 51 (1/2): 1–20. https://doi.org/10.1016/j.jacceco.2010.06.001
Ramalingegowda, S. 2014. Evidence from impending bankrupt firms that long horizon institutional investors are informed about future
firm value. Review of Accounting Studies 19 (2): 1009–1045. https://doi.org/10.1007/s11142-013-9271-6
Shleifer, A., and R. Vishny. 1997. The limits of arbitrage. Journal of Finance 52 (1): 35–55. https://doi.org/10.1111/j.1540-6261.1997.
tb03807.x
Sirri, E., and P. Tufano. 1998. Costly search and mutual fund flows. Journal of Finance 53 (5): 1589–1622. https://doi.org/10.1111/0022-
1082.00066
Zhang, X. F. 2006. Information uncertainty and stock returns. Journal of Finance 61 (1): 105–137. https://doi.org/10.1111/j.1540-6261.
2006.00831.x

The Accounting Review


Volume 94, Number 3, 2019
Financial Reporting Quality, Investment Horizon, and Institutional Investor Trading Strategies 111

APPENDIX A
Variable Definitions
Variable Definition
Financial Reporting Quality Variables
AQ ¼ the standard deviation of the residuals from Equation (5) for a firm calculated over the period from year t
1 through year t5. Larger values of AQ indicate poorer FRQ.
ACCt ¼ d0 þ d1CFt1 þ d2CFt þ d3CFtþ1 þ d4DREVt þ d5PPEt þ Error
where:
ACCt ¼ accruals during year t divided by average assets during year t;
CFt1 ¼ cash flow from operations during year t1 divided by average assets during year t;
CFt ¼ cash flow from operations during year t divided by average assets during year t;
CFtþ1 ¼ cash flow from operations during year tþ1 divided by average assets during year t;
DREVt ¼ change in sales divided by average assets during year t; and
PPEt ¼ gross PP&E divided by average assets during year t.
We estimate Equation (5) for each Fama-French 48 industry using all firms with the necessary Compustat
and CRSP data.
RAQ ¼ decile rank of AQ on a [0,9] scale.
POOR_AQ ¼ an indicator that a firm is in the top 30 percent of AQ.
Trading Strategy Variables
BEMEi,t ¼ firm i’s book-to-market ratio calculated annually on June 30 each year. The numerator (book value of
equity) is from the fiscal year that ended in the previous calendar year, and the denominator is the
market value from the most recent calendar quarter.
PAST_RET ¼ calculated based on stock returns over the previous year before the calendar quarter-end. Of those 12
months, we omit the most recent month and calculate buy-and-hold returns over the previous 11 months.
VALUE ¼ an indicator that a firm is in the top 30 percent of BEME at a given calendar quarter.
PAST_WINNER ¼ an indicator that a firm is in the top 30 percent of PAST_RET at a given calendar quarter.
Proxies for Different Types of Institutions
TRA ¼ an indicator that an institution is classified as a Transient based on Bushee (2001).
DED ¼ an indicator that an institution is classified as a Dedicated based on Bushee (2001).
HF ¼ an indicator that an institution is classified as a Hedge Fund based on Agarwal et al. (2013).
Dependent Variables in Firm-Quarter Analysis
TRA_BET ¼ the percent of bets in a given firm-quarter by transient institutions. A bet is defined as a position where an
institution’s portfolio weight in a given stock (WGT) is in the top quintile of portfolio weights across all
other institutions that hold that stock.
THELD ¼ the average number of quarters an institutional investor with a large bet (top quintile) in quarter t will wait
until their portfolio weight is reduced to the middle quintile calculated over the following eight quarters
([0,8] scale).
AQ_TILTVALUE ¼ weighted average of RAQ calculated over an institution’s value holdings only (VALUE ¼ 1) less weighted
average of RAQ calculated over that institution’s non-value (VALUE ¼ 0) holdings.
AQ_TILTPAST_WINNER ¼ weighted average of RAQ calculated over an institution’s winner holdings only (PAST_WINNER ¼ 1) less
weighted average of RAQ calculated over that institution’s non-winner (PAST_WINNER ¼ 0) holdings.
Control Variables
MVE ¼ market value of equity.
BETA ¼ sensitivity to the market portfolio calculated over the prior 60 months (minimum 24 months required).
IRISK ¼ idiosyncratic risk calculated over the prior 60 months (minimum 24 months required).
TURN ¼ average monthly share turnover (volume divided by shares outstanding) over the prior 12 months.
NUM_INST ¼ number of institutional owners listed for a firm at a given calendar quarter.
SUE ¼ unexpected earnings for the most recent quarter reported before the measurement of institutional investor
data, where unexpected earnings is seasonally adjusted change in earnings (EPSPXQ) divided by the
standard deviation of seasonal earnings changes over the previous eight quarters.
ACC ¼ total accruals divided by average assets.
AGE ¼ number of years a firm has been listed on the CRSP MSF file.
LEV ¼ total debt (DLC þ DLTT) divided by total assets (AT).
NEARN ¼ an indicator that income before extraordinary items (IB) was negative in the previous fiscal year.
TRA_NOBET ¼ the percent of non-large bet institutions (bottom four quintile of WGT) in a given firm-quarter by transient
institutions.
(continued on next page)

The Accounting Review


Volume 94, Number 3, 2019
112 Bushee, Goodman, and Sunder

APPENDIX AVariable Definitions


Variable Definition
PRUDENT ¼ institutions whose fiduciary type is listed as a bank trust or pension (either corporate/private or public).
R(X) ¼ decile rank of variable (X) on a [0,9] scale. For example, RMVE is the decile rank of MVE.
X_TILT ¼ weighted average of the rank of variable X for each institution where weights are based on the institution’s
portfolio ranks. The ranks are calculated each quarter for all firms listed on the Thomson Financial
Institutional Investor database with non-missing variables for variables of interest and control variables
described in Section III.

The Accounting Review


Volume 94, Number 3, 2019

You might also like