Asymmetric Information between Employers
[Job Market Paper]
Abstract
Employer learning about workers abilities plays a key role in determining how workers
sort into jobs and are compensated. This study explores whether learning is symmetric or
asymmetric, i.e., whether potential employers have the same information about worker
ability as the incumbent rm. I develop a model of asymmetric learning that nests the
symmetric learning case and allows the degree of asymmetry to vary. I derive testable
implications for the prevalence of asymmetric learning involving a new dependent variable:
the variance in pay changes. Using the NLSY, I employ three distinct identi cation strategies
to test di⁄erent predictions of the model. I rst test whether laid-o⁄ workers appear
negatively selected compared to workers who lost jobs in plant closings, by comparing the
variances in pay changes at their new jobs. I next exploit the fact that groups of workers
di⁄er in their variances in ability based on economic conditions at time of entry into a rm to
show that incumbent wages track ability more closely than do outside rm wages. Finally, I
provide additional evidence using the fact that learning about ability is more symmetric for
some occupations than for others. All three cases favor the asymmetric learning model and
suggest that the e⁄ect on wage setting is signi cant both statistically and in terms of
economic magnitudes.
,1 Introduction
The sorting of workers into rms is an important aspect of the labor market. A signi cant part of
this sorting involves both worker and rm learning about the quality of workers over time. For
example, many patterns observed in the job mobility literature that young workers change jobs
often (Topel and Ward 1992) yet long-term employment relationships are common and the
probability of remaining at a rm rises with tenure (Farber 1999) are consistent with employer
learning over time about overall or match-speci c worker quality. However, the extent to which
information about workers spreads across the market is less clear. Asymmetric information
exists when incumbent employers have more information about worker quality than potential
employers do. Consider a team of research assistants working for a professor. The professor can
see exactly which worker performed each task and who contributed important ideas, yet the
outside world observes only what appears on the bottom of an academic paper: the professor is
grateful for the excellent research assistance
of....
Under asymmetric information, ine¢ ciencies can arise in both allocating workers to jobs and
investing in worker human. Waldman s 1984 model assumes that outside rms can observe
promotions but not wages and implies that a worker s outside option changes sharply upon
promotion.1 The incumbent rm must o⁄er a large wage increase to keep the worker so there are
fewer promotions than would be optimal. 2 Turning to human capital, several theoretical papers
1 Other papers focusing on the promotion-as-signal hypothesis include Bernhardt (1995) which contains an
extended version of the Waldman (1984) model with more empirical predictions and Zabojnik and Bernhardt
(2001) which brings tournament theory and general human capital investment into the model.
2 Similarly, Milgrom and Oster (1987) hypothesize that skills of minority workers are di¢ cult for outside rms to
discern, but promotions allay this problem by increasing observability. Firms thus underpromote minority workers
to retain their informational advantage.
1
,point out that under asymmetric information, workers could underinvest in general skills. 3
When outside rms cannot observe workers investments, the incumbent rm need not fully
compensate the worker.4 It is necessary to understand the importance of asymmetric
information to know whether these ine¢ ciencies should be of concern.
This paper seeks to identify whether asymmetric information is prevalent in the labor
market. Since promotions and non-schooling human capital investment are di¢ cult to observe
in standard data sets, I develop a methodology that uses wages to identify the existence of
asymmetric information. I derive a model that allows the degree of asymmetry to vary and
nests symmetric learning as a special case, as follows. Assume workers have some degree of
unobserved ability. At the point of entry, workers are paid identically, conditional on the
component of ability that the rm observes. Once a worker is hired, the incumbent rm learns
about the worker s initially unobserved ability. Some portion of rm learning is re ected in the
wage. The distribution of wages within a rm, in consequence, becomes more dispersed over
time as employee tenure increases. Wage changes, re ecting how much an employer has
learned between periods, also spread out. Further, the distribution of wage changes track the
distribution of initially unobserved ability, implying that the variance in wage changes is
increasing in the variance in ability.
I employ three identi cation strategies that test di⁄erent implications of the model, using the
National Longitudinal Survey of Youth. First, the asymmetric information model implies that
lower-ability workers are more likely to leave the rm (Greenwald (1986) and Gibbons and Katz
3 See, for example, Chang and Wang (1996), Katz and Ziderman (1990) and Waldman (1990).
4 Alternatively, Acemoglu and Pischke (1998) show that under asymmetric information it might be optimal for
rms to provide general training to workers because, by selecting into a rm that provides general training, workers
can signal that they are high quality.
2
, (1991)) since the incumbent employer but not the outside employer has learned that they are
low ability. I test this prediction by comparing employer learning at the next job of endogenous
versus exogenous movers; under asymmetric information, endogenous movers will have a
lower variance in ability. Following Gibbons and Katz (hereafter GK), I exploit plant closings as a
plausibly exogenous event forcing workers to switch jobs. 5 Consistent with the asymmetric
learning model, I nd that wage changes at a new employer are more spread out for workers
who lost jobs due to plant closings than for laid-o⁄ workers.
The second identi cation strategy tests comparative statics on the variance in ability of
cohorts of workers. Under asymmetric information, outside rms update less on initially
unobserved ability than incumbent rms do. Thus for movers relative to stayers, the wage
change (de ned for movers as the rst wage at the new rm minus the last wage at the old rm) will
be less linked to ability. In particular, the variance of wage changes for movers will be less
related to the variance in initially unobserved worker ability.
For cross-group di⁄erences in the variance in worker ability I use labor market conditions at
time of entry into a rm. I show that workers who enter rms in recessions have lower variance in
ability than workers who enter rms in booms. During a recession, the share of job-seekers who
are unemployed or leaving bad jobs increases relative to boom-times; those workers with better
jobs quit less frequently because they are less likely to nd a better opportunity elsewhere. In
consequence, the variance of worker ability will be lower for workers who enter rms in
recessions these workers are more likely to be of lower quality. 6Consistent with their lower
5 GK hypothesize that if learning is asymmetric, being laid o⁄ should be a negative signal of ability whereas
there should be no stigma associated with losing a job due to a plant closing. Using data from the CPS Displaced
Workers Survey, they show that laid o⁄ workers pay a larger wage penalty at a new job, relative to workers who
lost their jobs due to plant closings.
6 6Workers who enter rms in recessions will also have a lower mean of ability. However, I show below
3
[Job Market Paper]
Abstract
Employer learning about workers abilities plays a key role in determining how workers
sort into jobs and are compensated. This study explores whether learning is symmetric or
asymmetric, i.e., whether potential employers have the same information about worker
ability as the incumbent rm. I develop a model of asymmetric learning that nests the
symmetric learning case and allows the degree of asymmetry to vary. I derive testable
implications for the prevalence of asymmetric learning involving a new dependent variable:
the variance in pay changes. Using the NLSY, I employ three distinct identi cation strategies
to test di⁄erent predictions of the model. I rst test whether laid-o⁄ workers appear
negatively selected compared to workers who lost jobs in plant closings, by comparing the
variances in pay changes at their new jobs. I next exploit the fact that groups of workers
di⁄er in their variances in ability based on economic conditions at time of entry into a rm to
show that incumbent wages track ability more closely than do outside rm wages. Finally, I
provide additional evidence using the fact that learning about ability is more symmetric for
some occupations than for others. All three cases favor the asymmetric learning model and
suggest that the e⁄ect on wage setting is signi cant both statistically and in terms of
economic magnitudes.
,1 Introduction
The sorting of workers into rms is an important aspect of the labor market. A signi cant part of
this sorting involves both worker and rm learning about the quality of workers over time. For
example, many patterns observed in the job mobility literature that young workers change jobs
often (Topel and Ward 1992) yet long-term employment relationships are common and the
probability of remaining at a rm rises with tenure (Farber 1999) are consistent with employer
learning over time about overall or match-speci c worker quality. However, the extent to which
information about workers spreads across the market is less clear. Asymmetric information
exists when incumbent employers have more information about worker quality than potential
employers do. Consider a team of research assistants working for a professor. The professor can
see exactly which worker performed each task and who contributed important ideas, yet the
outside world observes only what appears on the bottom of an academic paper: the professor is
grateful for the excellent research assistance
of....
Under asymmetric information, ine¢ ciencies can arise in both allocating workers to jobs and
investing in worker human. Waldman s 1984 model assumes that outside rms can observe
promotions but not wages and implies that a worker s outside option changes sharply upon
promotion.1 The incumbent rm must o⁄er a large wage increase to keep the worker so there are
fewer promotions than would be optimal. 2 Turning to human capital, several theoretical papers
1 Other papers focusing on the promotion-as-signal hypothesis include Bernhardt (1995) which contains an
extended version of the Waldman (1984) model with more empirical predictions and Zabojnik and Bernhardt
(2001) which brings tournament theory and general human capital investment into the model.
2 Similarly, Milgrom and Oster (1987) hypothesize that skills of minority workers are di¢ cult for outside rms to
discern, but promotions allay this problem by increasing observability. Firms thus underpromote minority workers
to retain their informational advantage.
1
,point out that under asymmetric information, workers could underinvest in general skills. 3
When outside rms cannot observe workers investments, the incumbent rm need not fully
compensate the worker.4 It is necessary to understand the importance of asymmetric
information to know whether these ine¢ ciencies should be of concern.
This paper seeks to identify whether asymmetric information is prevalent in the labor
market. Since promotions and non-schooling human capital investment are di¢ cult to observe
in standard data sets, I develop a methodology that uses wages to identify the existence of
asymmetric information. I derive a model that allows the degree of asymmetry to vary and
nests symmetric learning as a special case, as follows. Assume workers have some degree of
unobserved ability. At the point of entry, workers are paid identically, conditional on the
component of ability that the rm observes. Once a worker is hired, the incumbent rm learns
about the worker s initially unobserved ability. Some portion of rm learning is re ected in the
wage. The distribution of wages within a rm, in consequence, becomes more dispersed over
time as employee tenure increases. Wage changes, re ecting how much an employer has
learned between periods, also spread out. Further, the distribution of wage changes track the
distribution of initially unobserved ability, implying that the variance in wage changes is
increasing in the variance in ability.
I employ three identi cation strategies that test di⁄erent implications of the model, using the
National Longitudinal Survey of Youth. First, the asymmetric information model implies that
lower-ability workers are more likely to leave the rm (Greenwald (1986) and Gibbons and Katz
3 See, for example, Chang and Wang (1996), Katz and Ziderman (1990) and Waldman (1990).
4 Alternatively, Acemoglu and Pischke (1998) show that under asymmetric information it might be optimal for
rms to provide general training to workers because, by selecting into a rm that provides general training, workers
can signal that they are high quality.
2
, (1991)) since the incumbent employer but not the outside employer has learned that they are
low ability. I test this prediction by comparing employer learning at the next job of endogenous
versus exogenous movers; under asymmetric information, endogenous movers will have a
lower variance in ability. Following Gibbons and Katz (hereafter GK), I exploit plant closings as a
plausibly exogenous event forcing workers to switch jobs. 5 Consistent with the asymmetric
learning model, I nd that wage changes at a new employer are more spread out for workers
who lost jobs due to plant closings than for laid-o⁄ workers.
The second identi cation strategy tests comparative statics on the variance in ability of
cohorts of workers. Under asymmetric information, outside rms update less on initially
unobserved ability than incumbent rms do. Thus for movers relative to stayers, the wage
change (de ned for movers as the rst wage at the new rm minus the last wage at the old rm) will
be less linked to ability. In particular, the variance of wage changes for movers will be less
related to the variance in initially unobserved worker ability.
For cross-group di⁄erences in the variance in worker ability I use labor market conditions at
time of entry into a rm. I show that workers who enter rms in recessions have lower variance in
ability than workers who enter rms in booms. During a recession, the share of job-seekers who
are unemployed or leaving bad jobs increases relative to boom-times; those workers with better
jobs quit less frequently because they are less likely to nd a better opportunity elsewhere. In
consequence, the variance of worker ability will be lower for workers who enter rms in
recessions these workers are more likely to be of lower quality. 6Consistent with their lower
5 GK hypothesize that if learning is asymmetric, being laid o⁄ should be a negative signal of ability whereas
there should be no stigma associated with losing a job due to a plant closing. Using data from the CPS Displaced
Workers Survey, they show that laid o⁄ workers pay a larger wage penalty at a new job, relative to workers who
lost their jobs due to plant closings.
6 6Workers who enter rms in recessions will also have a lower mean of ability. However, I show below
3
