Financial Reporting Quality, Turnover Risk, and Wage Differentials: Evidence from Worker-level Data Jung Ho Choi Brandon Gipper Sara Malik Stanford University Graduate School of Business Current draft: October 2019 -Preliminary- Please do not cite or circulate. Abstract We examine whether and how financial reporting quality influences employee turnover and wages using employer-employee matched data in the U.S. We hypothesize and find that low financial reporting quality is associated with high employee turnover risk due to over- and under- employment, so workers demand wage premiums to bear this risk. High corporate governance firms exhibit a weaker association between financial reporting quality and turnover rates, suggesting that corporate governance mitigates turnover risk related to low financial reporting quality. We further find that more educated and higher paid workers receive higher wage premiums associated with financial reporting risk although turnover rates are similar across these different groups of workers, consistent with sophisticated workers identifying financial reporting risk. With Sarbanes-Oxley-mandated reports of internal controls weaknesses as a research setting, we show that as a firm’s internal control system weakens, workers start to require wage premiums. Overall, these analyses indicate that low financial reporting quality firms compensate for higher turnover risk by paying higher wages to workers. Keywords: Financial Reporting Quality, Wage, Turnover JEL classifications: D83, J31, J63, M41, M51, M52 ____________________________ Contact: [email protected], [email protected], and [email protected]; We thank Matthew Bloomfield, Jessica Kim-Gina, Sheffield E Lesure, Hyungil Oh, and Sorabh Tomar for helpful comments. Any opinions and conclusions expressed herein are those of the author(s) and do not necessarily represent the views of the U.S. Census Bureau. This research was performed at a Federal Statistical Research Data Center under FSRDC Project Number FSRDC1668. All results have been reviewed to ensure that no confidential information is disclosed. This research uses data from the Census Bureau's Longitudinal Employer Household Dynamics Program, which was partially supported by the following National Science Foundation Grants SES-9978093, SES-0339191 and ITR-0427889; National Institute on Aging Grant AG018854; and grants from the Alfred P. Sloan Foundation. We gratefully acknowledge financial support from Stanford University Graduate School of Business.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Financial Reporting Quality, Turnover Risk, and Wage Differentials:
Evidence from Worker-level Data
Jung Ho Choi
Brandon Gipper
Sara Malik
Stanford University Graduate School of Business
Current draft: October 2019
-Preliminary-
Please do not cite or circulate.
Abstract
We examine whether and how financial reporting quality influences employee turnover and wages
using employer-employee matched data in the U.S. We hypothesize and find that low financial
reporting quality is associated with high employee turnover risk due to over- and under-
employment, so workers demand wage premiums to bear this risk. High corporate governance
firms exhibit a weaker association between financial reporting quality and turnover rates,
suggesting that corporate governance mitigates turnover risk related to low financial reporting
quality. We further find that more educated and higher paid workers receive higher wage premiums
associated with financial reporting risk although turnover rates are similar across these different
groups of workers, consistent with sophisticated workers identifying financial reporting risk. With
Sarbanes-Oxley-mandated reports of internal controls weaknesses as a research setting, we show
that as a firm’s internal control system weakens, workers start to require wage premiums. Overall,
these analyses indicate that low financial reporting quality firms compensate for higher turnover
Contact: [email protected], [email protected], and [email protected]; We thank Matthew Bloomfield, Jessica Kim-Gina, Sheffield E Lesure, Hyungil Oh, and Sorabh Tomar for helpful comments. Any opinions and
conclusions expressed herein are those of the author(s) and do not necessarily represent the views of the U.S. Census
Bureau. This research was performed at a Federal Statistical Research Data Center under FSRDC Project Number
FSRDC1668. All results have been reviewed to ensure that no confidential information is disclosed. This research
uses data from the Census Bureau's Longitudinal Employer Household Dynamics Program, which was partially
supported by the following National Science Foundation Grants SES-9978093, SES-0339191 and ITR-0427889;
National Institute on Aging Grant AG018854; and grants from the Alfred P. Sloan Foundation. We gratefully
acknowledge financial support from Stanford University Graduate School of Business.
A large literature in labor economics has studied the substantial income dispersion across
workers. Recent papers emphasize that firm heterogeneity and employer-employee matching, in
addition to worker heterogeneity, mostly account for wage variation in recent decades (e.g., Card
et al., 2013; Song et al., 2019). These papers reinforce the importance of firm heterogeneity in
workers’ wage dispersion (such as firm’s industry, Krueger and Summers, 1988; or firm size,
Brown and Medoff, 1989). In other words, similar workers are paid differently because they work
for different firms. We conjecture that financial reporting quality could be among the firm’s
characteristics that shape the rank-and-file worker labor market and wage structure. In this paper,
we examine whether workers are compensated for poor quality financial reporting because
workers of those firms may experience higher turnover rates. To answer this question, we use
employer-employee matched data because the dataset allows us to estimate the wage premiums
while controlling for both firm and worker characteristics, which are a key challenge in the
literature (e.g., Abowd et al., 1999; Levetti and Shumutte, 2018).
We predict that workers will demand wage premiums for the assumption of turnover risk.
We also predict that employee turnover (measured in separations from and joins with the firm) is
associated with low financial reporting quality; this turnover is due to low financial reporting
quality’s association with both overemployment and underemployment (Biddle et al., 2009; Jung
et al., 2014) and investment inefficiency’s disruptive nature. 1 When firms over-employ and
workers join aggressively, subsequent corrections will be disruptive to the firm, including its
workers (Kedia and Philippon, 2008; Choi and Gipper, 2019). For example, if capital markets fail
to effectively monitor a firm, and the firm overinvests, then subsequently some of those
investments will fail disproportionately and associated workers will need to be reallocated or
separated from the firm.2 The turnover can be costly for the worker due to unemployment spells
1 Evidence from a series of papers suggests that higher financial reporting quality will cause an improvement in both
physical and human capital investment efficiency (e.g., Biddle and Hilary, 2006; Biddle et al., 2009; Chen et al.,
2011; Jung et al., 2014; Lara et al., 2016). Agency theory motivates these papers; higher financial reporting quality
causes improved allocation and monitoring of capital. 2 Because agency theory motivates this prediction, firm governance should moderate the relation between financial
reporting quality and employee turnover because improved monitoring can substitute for low financial reporting
quality.
2
and earnings losses (Jacobson et al., 1993; Couch and Placzek, 2010), and its prospect represents
a risk for which the firm may need to compensate (e.g., Baily, 1974; Rosen, 1986).3
Estimating this compensating wage differential is challenging because the relation between
financial reporting risk and wage premiums is conceptually identified for workers with the same
abilities and firms with the same characteristics except for financial reporting risk (Lavetti and
Schmutte, 2018).4 For example, consider two firms and two workers. Firm A has higher financial
reporting risk than Firm B. Worker A receives a high wage from Firm A. Worker B receives a low
wage from Firm B. These observations may not support a compensating wage differential for
financial reporting risk because worker and firm heterogeneity biases the estimate. This challenge
becomes further complicated because workers can move to different firms. If Firm B experiences
a shock to its financial reporting risk, Worker A can move to Firm B if he or she prefers high
financial reporting risk firms. Thus, we are unable to identify a compensating wage differential
unless we are also able to observe worker movements.
To tackle this empirical challenge, we use confidential employer-employee matched data
from the U.S. Census Bureau. This panel dataset contains detailed information on worker
characteristics, which measure various attributes of human capital.5 It allows us to control for
observable and time-invariant unobservable employee characteristics as well as employer
characteristics. Specifically, we prevent wage effects of employees’ inherent abilities and skills
and of employers’ average wage premiums from biasing our estimates. Abowd et al. (1999)
proposes an econometric model utilizing this employer-employee matched and demonstrates
quantitatively how the data set and model jointly tackle the empirical challenge. The ultimate
model which we estimate is from Abowd et al. (1999, herein referred to as “AKM”) and includes
both worker and firm fixed effects. Unobserved employee ability and how employees match to
employers (e.g., assortative matching) are sources of variation that are important to understand
3 Because we have predicted turnover, we believe that some portion of aggregate wage premiums will be a result of
worker composition and, therefore, be cross-sectionally related to worker characteristics. For example, turnover
risk arising from financial reporting quality can be hard to understand, so we expect more sophisticated workers
to receive compensating wage differentials, such as (already) high-wage or college-educated workers. 4 Financial reporting risk is the opposite of financial reporting quality. 5 Worker characteristics are important determinants of job choices and wages (e.g., Becker, 1993). Our baseline
turnover and wage regressions are comparable, showing the association between turnover and wage premiums and
(abnormal) accruals. With the wage premiums tests, we then more carefully expand the fixed effects to measure
the impact of employer-employee match composition.
3
when making inferences about labor outcomes.6 AKM-type regressions enable us to decompose
the influence of this variation when estimating compensating wage differentials. 7 Another
advantage of this data set is the ability to examine worker flows as opposed to only net changes in
total workers (Burgess, Lane, and Steven, 2000). In our regression analyses, we examine the
association between dependent variables such as establishment-level employee turnover,
separations, and joins with independent variables (i) lagged, absolute abnormal accruals (i.e.,
Dechow and Dichev, 2002; McNichols, 2002) and (ii) lagged, absolute total accruals, along with
comprehensive time-varying controls and fixed effects.8
We find that low financial reporting quality firms have higher employee turnover; when
examining worker wages, we find that high financial reporting quality firms pay lower wage
premiums to worker (i.e., high financial reporting quality firms have lower compensating wage
differentials (Rosen, 1986)). First, the turnover effect is observable in both separations and joins.
These turnover effects are also highly durable over at least a period of four years. This effect is
also present when examining within-firm financial reporting quality variation.9 We also find that
the turnover effect is attenuated for high governance firms. Next on the wage effect, the result is
robust even after controlling for worker characteristics, worker fixed effects (a measure for
unobserved worker ability), and firm fixed effects (a measure for unobserved average wage
premiums). Despite the persistence of the financial reporting quality-wage association with these
controls, it is clear that much of the wage premium comes from changes in worker composition.
6 Bertrand and Schoar (2003) highlight (in part) these concerns with respect to executive labor markets, and their
model has been adopted in the accounting literature, e.g., Dyreng et al. (2010) or Ge et al. (2011). In labor
economics, Abowd et al. (1999) has been the workhorse model for many papers with a much wider cross-section
of firms’ workers, such as Card et al. (2013), Lavetti and Schmutte (2018), and Song et al. (2019). This model has
two main identification assumptions, (i) the orthogonal matching condition—or the model controls for all attributes
that are correlated with our variable of interest and contribute to the employer-employee match, and (ii) our variable
of interest is a good measure of the theoretical construct (i.e., financial reporting quality in this case) after
controlling for other variables. We discuss the research design more in Section 4. 7 We also can speak to whether workers demand premiums conditional on worker ability or whether workers simply
reallocate themselves across firms due to the firm’s (time-varying) financial reporting quality risk profile. These
effects are particularly important to understand all costs of low financial reporting quality to firms and workers,
i.e., not just higher wages but also costs associated with worker reallocation, separations, and hiring, such as
severance pay or training costs. 8 We also examine the durability of disruption effects by estimating the association of financial reporting quality on
turnover, separations, and joins up to four years ahead. We additionally show subsample flows by workers’
characteristics, such as education and gender. 9 Note that one association is only partially consistent with our expectations, joins are negatively associated with
absolute total accruals. Moreover, if financial reporting quality is negatively associated with employment
disruption, we might expect that joins would be positively associated with absolute total accruals. However, it is
also possible that firms have more trouble attracting workers when financial reporting quality is low.
4
With worker fixed effects (the fully-specified AKM model), coefficient estimates equal between
one-fourth to one-eighth (about one-tenth) of the coefficient magnitude excluding these controls.
That is, workers appear to demand wage premiums but also reshuffle themselves among firms to
bear this turnover risk. We further document that wage premiums are larger for higher income
workers and more educated workers, indicating worker sophistication affects wage setting.
Turnover ratios are (weakly) consistent with the argument. Descriptively, female workers also
have slightly higher turnover and wage premiums; though the cause is not obvious, possible
explanations include discrimination or differential preferences.
As is common in accounting and finance papers, we are also subject to other sources of
endogeneity besides unobserved worker ability and worker-to-firm matching, such as
unobservable variation in the riskiness of a firm’s projects and the impact of that risk on proxies
for financial reporting quality and returns to labor (e.g., Roberts and Whited, 2013). In certain
specifications, we use firm fixed effects; however, there could also be endogenous variation within
firms. To reduce the effect of such concerns, we examine the wage premiums in the post-SOX era
when firms announce an internal control weakness over financial reporting (ICW). We limit the
firms to those that remediate ICWs to reduce the influence of financial distress on wage premiums
(e.g., Agrawal and Matsa, 2013; Graham et al., 2019). We find that workers respond to this time
series variation in financial reporting quality which is plausibly unrelated to the riskiness of a
firm’s projects. Tabulated in the Internet Appendix, we do not find that these ICWs are
significantly associated with time-series variation in employee turnover. On average, we find that
workers receive compensating wage differentials after ICWs are announced. Additionally, we find
that college-educated workers tend to demand some wage premiums before the announcement and
incrementally more after, while non-college workers only demand wage premiums after the ICW
announcements. This is consistent with college-educated workers being more sophisticated or
occupying jobs that are closer to / use financial reporting systems and identifying low financial
reporting quality ex ante while non-college workers only finding out about low financial reporting
quality ex post.
We make three main contributions. First, we contribute to a prominent set of papers that
associates financial reporting quality and cost of capital or capital investment efficiency (e.g.,
Bushman and Smith, 2001; Francis et al., 2004; Biddle and Hilary, 2006; Lambert et al., 2007; and
Biddle et al., 2009). Instead of focusing on costs in capital markets or physical capital, we examine
5
costs in labor markets. As Whited (2019) writes, “[Costs in labor markets are] of growing interest
because over the last several decades, many advanced economies have shifted employment away
from capital-intensive manufacturing industries and toward labor-intensive service industries.”
This paper documents costly consequences of low financial reporting quality for an important
production input. These costs arise because firms share risks with workers or offer them
compensating wage differentials when they cannot share risks, such as turnover (e.g., Baily, 1974;
Guiso et al., 2005; Guvenen et al., 2017). When the firm’s financial reporting quality does not
indicate that the firm can successfully share risks, costs of labor rise. Because workers cannot
easily diversify this risk, it is plausibly more costly to firms than capital costs, which can be
diversified by investors.
Second, we contribute to an emerging and well-populated group of papers in both finance
and accounting that associate labor market outcomes, such as wages, turnover, and job searches
with characteristics of the firm. For example, those published in finance journals and some
working papers show that employee turnover, demand wage premiums, or apply elsewhere when
the firm is distressed (e.g., John et al, 1992; Berk et al., 2010; Chodorow-Reich, 2014; Brown and
Matsa, 2016; Baghai et al., 2018). Working papers that examine firms’ accounting attributes, such
as real earnings management, earnings persistence, or accounting fraud, show associations with
worker mobility or wages (e.g., Bai et al., 2018; Hass et al., 2018; Baik et al., 2019; Gipper and
Choi, 2019, Makridis and Zhou, 2019). 10 Our paper documents worker-level responses and
heterogeneity to accounting characteristics of the firm. Using this micro data, we show that
workers are mobile and behave strategically (consistent with Matsa, 2018), so they may leave or
negotiate higher wages if there is concern about the firm’s ability to avoid employee turnover due
to inefficient investments that arise out of low financial reporting quality.
Third, besides scattered use of Bertrand and Schoar (2003) for executives, we apply and
discuss the implications for the use of the research design and the employer-employee matched
data from wage regressions in AKM (Abowd et al., 1999) in accounting research. Papers in
accounting widely recognize endogeneity concerns with respect to firm-level variation. However,
studying employment or wage (whether with aggregated data or with worker-level data) introduces
10 Other papers use non-wage, non-turnover worker attributes (e.g., unionization, unemployment insurance coverage)
to test that workers are stakeholders to whom managers cater when preparing financial reports; for example, see
Ng et al. (2015), Chung et al. (2016), and Hamm et al. (2018).
6
another set of endogeneity concerns that are carefully treated by labor economists (e.g., Card et
al., 2013; Song et al., 2019). We discuss the importance of using models like AKM and employee-
employer matched data for understanding the correlations in the analysis. This differentiates our
paper from two concurrent working papers that examine related questions (Bai et al., 2018, and
Baik et al., 2019); moreover, these papers regress aggregated firm (establishment)–level wages on
a firm’s financial reporting attributes. We find that workers reallocate themselves among firms
with different financial reporting quality, so worker-firm matching and worker characteristics are
important determinants of compensating wage differentials for financial reporting quality.
2. Conceptual Framework
2.1. Wage Differentials
A long strand of literature studies wage setting to compensate for varying job characteristics
(e.g., Smith, 1979).11 For example, Thaler and Rosen (1976) find that jobs with higher injury
incidence pay higher wages to workers to compensate for the physical risk. Abowd and Ashenfelter
(1981) document that workers are compensated for unemployment possibilities. Graham et al.
(2019) demonstrate a positive relation between firm leverage and worker wages, attributing a
compensating wage differential to bankruptcy risk. These papers usually model a wage differential
by using a hedonic pricing model with implicit contracts (Rosen, 1974). The estimate in the
hedonic pricing model reflects workers’ revealed preferences over job characteristics and
willingness to be compensated for job specific risks.12
11 One reason for wage differentials across firms is to compensate for firm characteristics. These wage differentials
can have important consequences, such as contributing to income inequality. For example, Song et al. (2019)
document that the 90th percentile worker earns 10.6 times more than the 10th percentile worker in 2013. Although
the phenomenon is well documented in the literature, more recently the role of firm in the income dispersion has
drawn the attention of the literature. Card et al. (2013) find that not only intra-firm income inequality but also inter-
firm income inequality are important to explain cross worker income dispersion in Germany. Song et al. (2019)
demonstrate that across firm income dispersion in the U.S. account for 42% of income dispersion of workers in
2013. Their analysis additionally indicates that another important factor to explain an increase in income inequality
in the U.S. is assortative matching, high paid workers match with high paying firms. Overall, these papers point
out that understanding wage differentials across firms are important to understand income dispersion across
worker. 12 One limitation of the hedonic pricing model to estimate this wage differential is unobserved worker and firm
characteristics (Hwang et al, 1992). In other words, it is difficult to observe an offer curve from the hedonic pricing
model for employees with the same ability from employers with the same characteristics except for financial
reporting risk. In Section 4, we build on Abowd et al. (1999) to improve our estimate of compensating wage
differentials for accounting characteristics to deal with these two side heterogeneities. In addition, an endogenous
movement of workers can influence the estimate in the hedonic pricing model when a compensating wage
differential is estimated even with both firm and worker fixed effects (Lavetti and Schmutte, 2018).
7
2.2. Financial Reporting Quality, Turnover, and Wage Differentials
Financial reporting risk can generate compensating wage differentials through turnover risks.
Prior studies demonstrate negative consequences of high turnover rates (e.g., Jacobson et al., 1993;
Couch and Placzek, 2010). The immediate costs of high involuntary turnover rates are
unemployment spells and earnings losses. Jacobson et al. (1993) find that displaced workers due
to factory closing suffer wage losses averaging 25% per year for the next six years. These displaced
workers take multiple months to find a new job. These findings indicate that a potential link
between financial reporting quality and turnover rates can explain why workers may require wage
premiums for financial reporting risk.
Low financial reporting quality can increase turnover rates though inefficient employment.
Financial reporting risk may lead to over- or under-employment and, thus, excessive turnover
rates. Prior studies document the effect of financial reporting quality on investment inefficiency
both in physical and human capital. 13 Inefficient employment increases the volatility of
employment unless it is negatively correlated with productivity shocks in general. Overemployed
workers (due to over-investment) are likely to be separated when firms receive negative
productivity shocks and can no longer afford to maintain the overemployed workers. Poor financial
reporting quality also limits firms’ abilities to choose employment decisions efficiently when firms
are under financial constraints. Falato and Liang (2016) find that firms contract their worker
headcount substantially when they violate loan covenants probably due to negative productivity
shocks. This effect can be larger for low financial reporting quality firms than for high financial
reporting quality firms. Moreover, all of these outcomes can be amplified, because poor financial
reporting quality limits shareholders ability to deter managers’ empire building behaviors (Jensen,
1986; Richardson, 2006). In this sense, we conjecture that low financial reporting quality also
increases employment volatility and, thus, employee turnover ratios. Nonetheless, this relation is
an empirical question due to the relation between inefficient employment and productivity shocks.
13 E.g., Biddle et al., 2009; Jung et al., 2014; Roychowdhury et al., 2019. Jung et al. (2014) find that low financial
reporting quality induces labor investment inefficiency via overemployment and financial constraints. They find
that the absolute value of abnormal net hires is larger as financial reporting quality is worse. McNichols and
Stubben (2008, page 1,599) document that discretionary revenues lead to excessive physical investments although
subsequent investment drops are not explicitly tested.
8
Collectively, these arguments suggest that workers employed by high financial reporting risk
firms are likely to face higher turnover risks and to be compensated by higher wages for those
risks.14 In this sense, our first hypotheses are as follows.15
correlations of Turnover, Separations, and Joins. Because these variables are constructed from the
same worker flow data, we expect and find that they are all strongly positively correlated with
each other. In fact, Turnover has a mechanical relation with the other two ratios. In addition, both
total accruals and abnormal accruals are positively correlated with Turnover, Separations, and
Joins. These univariate correlations are consistent with our expectations discussed in Section 2.
Other correlations also appear reasonable. For example, size is negatively correlated with worker
flows (Lane, Isaac, and Stevens, 1996); profitability (i.e., return on assets) and investment
opportunities / growth options (i.e., Tobin’s Q) are positively related to joins; economic volatility
(i.e., standard deviation of daily returns) is positively related to Turnover, Separations, and Joins.
20
Overall, these turnover correlations suggest that changes in workers (and/or worker composition)
are associated with financial reporting quality.
Column (4) presents univariate correlations of the worker-level data. Real wages have a
slight negative correlation with total accruals, which is inconsistent with our predictions from
Section 2; however, abnormal accruals strongly positively correlate with real wages. As with
turnover, many other firm-level variable correlations with wages are generally intuitive. Size and
investment opportunities / growth options are positively correlated with worker wages (Troske,
1999). Leverage and tangibility (i.e., physical capital intensive firms) are negatively correlated
with wages; this finding could be surprising given the positive predicted relation between leverage
and worker wages (e.g., Berk et al., 2010). However, even in samples of bankrupt (and matched
control) firms, the observed correlation flips for larger firms (e.g., see Graham et al., 2019, Table
7). Also perhaps surprisingly, lagged profitability and lagged economic volatility are weakly,
negatively correlated with worker wages. All worker characteristics are correlated with wages as
predicted by widely understood notions of human capital (Becker, 1993) and the gender gap;
specifically, age, experience, education, and male gender are strongly, positively associated with
wages.22
5.1. Turnover Results
Table 3 presents our first main result. We find that abnormal accruals are positively
associated with turnover. Unobserved, local labor market and unobserved firm heterogeneity can
contribute to this relation. Incrementally including establishment-industry, year, and county fixed
effects and, separately, firm fixed effects, cause the effect to attenuate from 0.103 to 0.058 to 0.028
in columns (1) through (3), respectively. The magnitude (e.g., from column (1)) indicates that a
one standard deviation change in Abnormal Accruals is associated with a 3.2% change in Turnover
relative to the sample mean or about 37 (2.16 million) additional separating or joining workers per
firm-year (across the entire sample).23 This is consistent with our discussion in Section 2, financial
22 Untabulated results indicate that total accruals and abnormal accruals are correlated but different constructs. For
example, these variables have different correlations with leverage, tangibility, and the mergers and acquisitions
indicator. Also, total (abnormal) accruals are negatively (positively) related to worker age and education. Both
accruals measures are negatively related to worker experience and female gender. 23 Standard deviation of Abnormal Accruals (0.119) multiplied by the column (1) coefficient (0.103) divided by
Turnover mean (0.378) equals 3.2%. Standard deviation of Abnormal Accruals (0.119) multiplied by the column
(1) coefficient (0.103) multiplied by mean Firm Employee Size (3,039) equal to 37. This quantity multiplied by 58
thousand firm-year observations.
21
reporting quality is negatively associated with employee turnover, so plausibly workers are having
voluntary or involuntary separations or sorting themselves (e.g., increasing joins) for reasons
related to financial reporting quality. In columns (4) through (6), we find positive but insignificant
correlations between Total Accruals and turnover. While this insignificant correlation is not
consistent with our predictions, we are able to dig further into this result when individually
examining separations and joins in Panel B. Overall, Abnormal Accruals are associated with lower
investment efficiency (e.g., Biddle et al., 2009) and with employee turnover. These findings are
consistent with our predictions that workers experience financial reporting quality-related turnover
risk. Although we do not empirically distinguish voluntary turnover from involuntary turnover,
plausibly some turnover is workers voluntarily reallocating themselves to firms that constitute
better matches for their own (risk) preferences or “type,” and some turnover is workers experience
involuntary separations due to the consequences of low investment efficiency.
We also discuss the coefficient estimates on the control variables. Turnover and size
(measured by assets) are negatively correlated, except in the specification with firm fixed effects.
Moreover, larger firms have more worker stability; however, as firms change in size, they have
less worker stability, i.e., more turnover. Within firm investment opportunities (i.e., Tobin’s Q)
and leverage are also associated with more employee turnover; these variables tend not to be
significantly (or are only weakly) correlated with turnover without firm fixed effects. Profitability
(return on assets) are positively correlated with turnover across firms and negatively associated
within firm. Establishment size (measured by worker count) is negatively correlated with turnover
across and within firms. Finally, returns volatility is positively correlated with turnover without
firm fixed effects. This control variable is plausibly related to other types of operational or financial
risks that could be correlated to financial reporting quality. Interestingly, when firm fixed effects
are included in the turnover specification, the association between turnover and returns volatility
is no longer significant at conventional levels (and attenuates substantially in magnitude) while
the correlation between Abnormal Accruals and turnover persists.
In Table 3 Panel B, we decompose turnover into Separations and Joins. In columns (1) and
(2), Abnormal Accruals are positively associated with both separations and joins. The separations
(joins) magnitude can be interpreted as follows: a one standard deviation change in Abnormal
Accruals is associated with about 18 (21) additional separating (joining) workers per firm-year (or
about 0.5% - 0.7% of the average firm’s workforce). Total Accruals have a split correlation
22
between separations and joins; moreover, Total Accruals is positively (negatively) associated with
separations (joins). This suggests that high (low) financial reporting quality firms, using this
measure, tend to have more joins (separations). These directional differences in separations and
joins are likely driving the statistically weak correlation between Total Accruals and turnover. We
find that other regressors generally have similar correlations as with Turnover, except that Tobin’s
Q splits between negative (positive) associations with separations (joins) and vice versa for
establishment-level worker size. In general, these results are consistent with the inferences that we
draw in Panel A. Financial reporting quality is negatively correlated with separations; workers
plausibly face higher voluntary and involuntary turnover risk when working for these low financial
reporting quality firms. The split correlation for joins is more challenging to interpret. Abnormal
Accruals seems also to be correlated with, at least, volatility of worker employment, even if
(somewhat) beneficial for workers who do join these firms. Alternatively, our measures also
capture between establishment worker transitions, so low financial reporting quality could be
associated with job location instability. If investment efficiency decreases, abandoning some
investment projects for others could result in workers being moved around, a disruption for the
worker. The negative association between joins and Total Accruals could indicate that workers
identify and avoid joining firms with low financial reporting quality. Overall, the evidence from
Table 3 suggests that workers face turnover risk on average; this risk can manifest as voluntary or
involuntary separations or disruptive / avoided joins for workers.
Figure 2 shows coefficient estimates along with 95% confidence interval bars for future
turnover, separations, and joins regressed on Abnormal Accruals.24 We use the specifications with
interacted establishment industry, year, and county effects (but excluding firm effects). We extend
the dependent variables up to four years after the independent variables, showing the estimates,
which range between 0.032 and 0.058 across the three panels; the coefficients are generally
significant at conventional levels, except year t+3 for turnover and separations. These results
indicate that the financial reporting quality-related turnover risk is highly durable, lasting at least
4 years. This can be consistent with continued voluntary and involuntary turnover arising out of
lower investment efficiency as a result of low financial reporting quality.
24 As with Table 3, using Total Accruals as a proxy for financial reporting quality or with firm fixed effects in the
specification estimating future years of turnover, separations, and joins has attenuated coefficient estimates.
Indeed, few coefficients are significant at conventional levels using these alternative designs as in Panel A of Table
3.
23
5.2. Wage Differential Results
Table 4 contains our results showing associations between financial reporting quality and
wages. Panel A contains baseline wage regressions that are comparable to the turnover regressions
from Table 3. We include specifications that follow Equations (2) and (3), so the estimates include
firm industry and year in columns (1) and (3) and establishment industry, year, and county effects
interacted plus worker characteristics. We again use Abnormal Accruals and Total Accruals as
proxies for low financial reporting quality. We find significant positive correlations between
proxies for low financial reporting quality and wages across all tests, consistent with workers
receiving positive compensating wage differentials for financial reporting risk, which from
evidence in Table 3, could be from a channel such as workers experiencing turnover risk.
Coefficient estimates for financial reporting quality proxies vary between 0.140 and 0.242; to
understand the magnitude, a one standard deviation increase in Abnormal Accruals would net the
average worker approximately a 1.7% increase in wages or $978 per year.25 Even in Panel A, we
note the importance of local labor markets and worker characteristics, coefficient estimates
decrease by 19% and 32% for Abnormal Accruals and Total Accruals, respectively, when
including these variables as controls.
In this table, some control variables have expected correlations. For example, we see the
strong, positive firm size wage effect (Brown and Medoff, 1989). Tobin’s Q is also positively
related to wages, so firms with growth options or higher-valued investment opportunities pay
workers more. Returns volatility is also positively associated with wages, so workers bearing
fundamental risk is also compensated (e.g., Baily, 1974). Worker characteristics have correlations
with wages consistent with accumulating human capital or the gender wage gap; i.e., more
experienced (but with diminishing returns), more educated, and male workers have higher wages.
As with the univariate correlations, surprisingly (e.g., Berk et al., 2010) wages are negatively
correlated with leverage. Perhaps these firms are not at risk of bankruptcy or have more physical
capital intensity, which is also negatively correlated with wages in these multivariate regressions.
Finally, and also surprisingly, cross-sectional profitability is negatively related to wage levels.
However, this coefficient and coefficients for asset tangibility and returns volatility, meaningfully
attenuate when including controls for local labor markets and worker characteristics. So,
25 For this calculation, we use the column (2) coefficient estimate (0.140) multiplied by standard deviation in
Abnormal Accruals (0.119) equals about 0.017. This approximate percentage increase multiplied by the average
wage ($58.7 thousand) equals $978.
24
profitable, tangible asset intensive, risky firms have a geographic and labor pool mix that drive
some of these associations. Control variable coefficient estimates in Panel B shows additional
evidence of this. The signs of coefficients flip for Return on Assets and Returns Volatility and
attenuate and are no longer significant for Tobin’s Q and Tangible Assets.
Table 4 Panel B includes the incremental fixed effects, progressing through to the AKM
model (Abowd et al., 1999). First, we include firm fixed effects in columns (1) and (4) to control
for average wage premiums of firms. A substantial positive bias in the cross-sectional results are
the result of firm heterogeneity (and unobserved worker effects matched to these firms that cannot
be disentangled empirically). We see attenuation and reduced statistical significance for these
specifications (even the sign flips, though the coefficient is very close to zero, in column (4) for
Total Accruals). Within firm wage differentials for financial reporting risk appears to be small;
however, this is before we account for unobserved worker heterogeneity and job matching. In
columns (2) and (5), we include worker fixed effects; again we see the coefficient attenuate (though
is still highly significant in both specifications). Worker effects alone account for worker
heterogeneity and some unobserved firm heterogeneity because employer-employee matches can
be sticky. As one would expect, worker effects substantially increases the explanatory power of
the model; r-squared increases from about 40% (Panel A) to about 93%. Finally, in columns (3)
and (6), we include both firm and worker fixed effects (i.e., AKM). With these specifications, we
get coefficient estimates with similar magnitudes, though the statistical significance is higher for
Abnormal Accruals. Also, for Abnormal Accruals, we find some incremental attenuation compared
with both the firm and, separately, worker fixed effect specification, suggesting that job-matching
causes incremental positive bias when estimating wage differentials for financial reporting risk.
To understand the magnitude, a one standard deviation increase in Abnormal Accruals would net
the average worker approximately a 0.2% increase in wages or $112 per year. Referencing back
to Figure 1, job switches appear to move workers along a positively-slopped expansion path. This
bias indicates that workers prefer to “consume” more of these risks as their income increases. This
could be consistent with financial reporting risk being associated with outcomes that could benefit
workers somehow, such as higher wages in the future.26
26 Alternatively, with Total Accruals, the bias between column (4) and (6) indicates that workers treat financial
reporting risk like a normal “bad” (i.e., opposite of normal good), where workers prefer less financial reporting
risk as their income increases. These different types of bias again indicate that Abnormal Accruals and Total
Accruals likely reflect different measures of financial reporting quality and can be consistent with Abnormal
25
5.3. Cross-sectional Results
We examine cross-sectional variation in turnover and wage associations with financial
reporting quality. We begin by examining variation in worker sophistication. Financial reporting
quality-related turnover risk is a complex causal chain; first, a (potential) worker has to identify
that a firm has low financial reporting or information quality and that this can cause managers to
make poor investments. Second, the worker needs to understand that poor investments will
increase the proportion of workers turning over in the future, i.e., she will have some financial
reporting quality-related turnover risk. Third and finally, this worker should reallocate herself to a
firm that matches her preferences or receive wage premiums to bear this type of risk. We use
worker education, non-college and college, and worker wage levels, below median and above
median, as proxies for sophistication. College workers and above median wage workers are
sophisticated and are likely to respond more to this complex causal chain. For turnover and
separations, we are limited to within establishment flow measures available in the LEHD data so
can (cannot) measure turnover and separations based on education (wage levels).
Table 5 Panel A presents associations from turnover and separation regressions for these
worker (though not establishment) subsamples. Columns (1) and (2) show results for non-college
workers; the magnitudes are slightly smaller than for college workers in columns (3) and (4).
Plausibly these turnovers that occur shortly after the revelation of financial reporting quality (i.e.,
in t+1 relative to the Abnormal Accruals) more likely measure voluntary (rather than involuntary)
realized turnover or separations. Instead, college workers could face very little (incremental)
turnover risk, resorting to wage premiums. In Panel B, we present these wage premium results.
For columns (2) and (4)—college workers—versus columns (1) and (3)—non-college workers, we
do measure larger increases in pay for college workers at 0.172 versus 0.117 and 0.188 versus
0.134 for Abnormal Accruals and Total Accruals, respectively. When examining worker splits by
wage levels in Panel C, we observe similar coefficient magnitude differences. The above median
wage worker subsample has coefficients of 0.140 and 0.118 while below median wage worker
subsample has coefficients of 0.031 and 0.018, for Abnormal Accruals and Total Accruals,