The Effectiveness of Work-Search Requirements over the Business Cycle: Evidence for Job Rationing∗ Desmond Toohey University of Delaware February 5, 2017

Abstract This paper examines the general equilibrium effects of work search requirements in unemployment insurance systems. The analyses exploit within-state changes in search requirements for unemployment claimants as a source of plausibly-exogenous variation in search effort. Using two proxies for labor market strength, the results show that increasing search requirements reduces unemployment more when the market is strong than when the market is weak. The results are framed in the context of a job-rationing search-and-matching model in which unemployment in downturns is driven not just by matching frictions but by rigidities limiting the employment level. Canonical searchand-matching models do not deliver the same predictions. The paper demonstrates that search requirements should not be expected to support employment rates in downturns. This result is a novel source of reduced form evidence for job-rationing models. JEL Codes: J08, J64, J65

∗

420 Purnell Hall, Newark, DE 19716, (302) 831-3809, Email: [email protected]. I am grateful to Aditya Aladangady, John Bound, Varanya Chaubey, Christina DePasquale, Kathryn Dominguez, Cynthia Doniger, Anne Fitzpatrick, Evan Herrnstadt, Eric Lewis, Jeff Smith, Kevin Stange, and Mel Stephens for helpful comments on earlier drafts and presentations. Excellent research assistance was provided by Sheila Afrakomah. Earlier drafts of this paper were circulated under the title “Job Rationing in Recessions: Evidence from Work Search Requirements.” Thanks also go to Andrew Spisak for help in acquiring the Benefit Accuracy Measurement Data and to seminar participants at the University of Michigan, the Federal Reserve Bank of Cleveland, the University of Delaware, Mathematica Policy Research, the Federal Reserve Board, the Congressional Budget Office, the University of Illinois, and the Federal Trade Commission. All errors and omissions are my own.

1

Introduction

In recent years, policymakers in many states have dramatically increased the requirements for unemployment insurance (UI) claimants to show that they are actively searching for work. Requiring evidence of work search is an attractive option for policymakers facing UI budgets that are weak and unemployment rolls that are long. There is initially good reason to believe that the changes should be effective: randomized experiments examining the effects of UI job search requirements have shown some evidence that they lead to faster reemployment (Johnson and Klepinger 1994; Klepinger, Johnson, and Joesch 2002). Canonical search-and-matching models of the labor market suggest that increasing the number of matches between workers and job vacancies has scope for lowering the unemployment rate. However, a growing literature indicates that negative externalities may limit the benefits of reemployment policies in general equilibrium (Davidson and Woodbury 1993; Lise, Seitz, and Smith 2004; Crépon et al. 2013). Further, recent job-rationing models of the labor market argue that a substantial fraction of unemployment in recessions is driven not by matching frictions but by wage rigidities and the rationing of jobs (Michaillat 2012; Landais, Michaillat, and Saez 2016). Because search requirements are largely aimed at increasing the rate of matches, their efficacy will be limited if other rigidities are the primary drivers of unemployment. In weak labor markets, search requirements may simply increase the rat race effect, leading UI claimants to compete harder for the same number of job openings. It is initially unclear whether search requirements can improve the functioning of the labor market and lead to faster reemployment when implemented at the state level. In this paper, I examine the effects of search requirements in general equilibrium and, in doing so, provide new reduced-form evidence for job-rationing models. The analysis exploits within-state changes in search requirements for UI claimants over time. I show that a number of proxies for search effort are increased when search requirements become more stringent. More importantly, I demonstrate that these policies can reduce

1

unemployment, but the effects are stronger when the labor market is strong. I frame the analysis in the context of a job-rationing search-and-matching model of the labor market. In such a model, unemployment in downturns is driven not just by matching frictions but by rigidities limiting the employment level even in the absence of matching frictions. The implications of this model hold even if search requirements simply function as a leisure tax and do not directly increase search effort. Although job search requirements have returned to near-ubiquity in UI systems and are a major point of contact between claimants and workforce agencies, they are only minimally documented and studied in the existing literature. In general, 75 to 80 percent of all UI claimants in the US are subject to their states’ active search policies. At the height of the Great Recession, more than 3 million new UI claimants each quarter were subject these policies. These policies are also largely missing from the literature on optimal UI design, which broadly assumes that search effort is unobserved by UI agencies. While job-search monitoring is clearly imperfect, this paper further demonstrates that it has impacts on the behavior of claimants. The prototypical job search requirement in the US requires claimants to contact a minimum number of prospective employers each week. The details of the nature and number of contacts required each week vary across states and over time. I catalog the changes in state-level search requirements since the turn of the century using UI handbooks and forms provided to claimants over time. On the whole, the documented policy changes suggest a strong reversal of the slackening in search rules during the 1980s and 1990s, with a number of states implementing more stringent requirements over the last decade. I compare these policy codings to the actual number of contacts reported by claimants when audited by their states’ Benefit Accuracy Measurement (BAM) programs. First, this analysis shows that claimants respond directly to rising search requirements by reporting more contacts to UI authorities. Second, it shows that many of these contacts are genuine: the number of contacts found by auditors to be acceptable also rises. This latter finding suggests that the additional contacts are reflective of actual search and

2

are not bogus contacts invented by claimants. I use within-state changes in search requirements to identify the effects of the policies. Using fixed effects models in state and time, I control for time-invariant state characteristics and national trends over time. As it is possible that the policy changes are correlated with time-varying state and claimant characteristics, I control for a number of variables that are likely to be drivers of individual search effort and labor market outcomes. The results show that likely UI claimants increase some measures of observed search effort when search requirements go up. I show this using measures of search methods as reported in the Current Population Survey (CPS) as in Shimer (2004). Using data on unemployment rates at the level of Core Based Statistical Areas (CBSAs), I test for these business cycle differentials and find that search requirements are relatively less effective in weak labor markets when using both lagged unemployment and an industry shift-share measure to proxy for labor market strength. This paper builds on the experimental results of the search policy literature exemplified by Klepinger, Johnson, and Joesch (2002) and Ashenfelter, Ashmore, and Deschênes (2005) by examining effects in general equilibrium. It also extends search policy analysis in other contexts as in Borland and Tseng (2007), McVicar (2008), and McVicar (2010) to the United States environment by examining a variety of different policies in the US unemployment system. A number of these studies (Ashenfelter, Ashmore, and Deschênes 2005; McVicar 2008, 2010) vary the incidence of job search monitoring, which may have different effects from changing job search requirements under a stable monitoring regime. More generally, this paper directly examines the general equilibrium effects of labor market policies in the spirit of Davidson and Woodbury (1993) and Lise, Seitz, and Smith (2004). Those papers, however, primarily consider search externalities exerted by a treated group on untreated groups, while this paper largely focuses on the mechanisms for crowd-out even if everyone is treated.1 To my knowledge, it is also the first paper to 1

The effects of these policies on nonclaimants or search-exempt workers is an important topic for future work.

3

interpret these kinds of unemployment policy changes in the context of a job-rationing model as in Michaillat (2012). In that sense, it provides new evidence to support the implication that some labor market policies may be less effective during downturns (Landais, Michaillat, and Saez 2016). The paper proceeds as follows. Section 2 describes the relevant institutional features of UI and details the way in which the policies studied in this paper are measured. Section 3 examines the predicted effects of a search requirement in a job-rationing search-and-matching model of the labor market. Section 4.1 describes the nature of the policy changes themselves and section 5 discusses the general empirical strategy employed in much of the paper. Section 6 presents the estimated effects of search policies on observable effort. Section 7 presents evidence on the efficacy of search requirements across labor market conditions. Section 8 discusses and concludes.

2

Institutional Setting

Search requirements are an important feature of UI systems even though they receive limited attention in the existing literature. Many states have implemented dramatic changes in these requirements over the past decade. I document the recent changes in search requirements using the publications available to UI claimants at the time of their claims. In Section 6.1, I compare my coding of the policies to data on the number of job contacts actually reported by claimants who are audited and find a strong relationship between the published rules and the reported contacts.

2.1

Search Requirement History

Unemployment insurance, while governed by a set of federal guidelines under the Federal Unemployment Tax Act (FUTA), differs across states. In general, it provides compensation to full-time, permanent workers who lose their employment through no fault of their own. To be initially eligible, workers must meet thresholds for quarters of employment and earnings in a set period before a job separation. Benefit levels are determined by earnings over the covered period. In general, benefits can be paid for approximately 26 weeks, but this varies across claimants and states. Benefit dura-

4

tions are also increased through both automatic and ad hoc extensions during periods of high unemployment via the Extended Benefits (EB) and Extended Unemployment Compensation (EUC) programs. Unemployment insurance is affected by a well-known moral hazard problem: higher unemployment benefits raise the relative value of remaining unemployed, lowering the incentive to search for and accept new employment. Conversely, UI allows workers to smooth consumption over employment shocks (Gruber 1997). While a second-best result can be reached by setting benefits to balance moral hazard against consumptionsmoothing benefits (Chetty 2006, 2008), policymakers have historically attempted to mitigate the problem by requiring demonstrable search effort and job offer acceptance from claimants. Search requirements of one form or another have been an important part of UI systems since at least the 1980s. While the original UI system established in the Social Security Act of 1935 did not specifically require search from claimants, later additions to federal law called for claimants to demonstrate active search. In particular, federal law first demanded these requirements for claimants receiving benefits under the EB program (Anderson 2001). Over the last few decades of the twentieth century, states variously implemented and eliminated search requirements for claimants of regular UI benefits (Klepinger, Johnson, and Joesch 2002). The 2000s and 2010s have been characterized by a steady increase in the strength of search requirements across many states. While most states already had wording in their statutes requiring that claimants be “able, available, and actively searching” for work, this language was added to the United States Code as a condition for state UI funding as part of the Middle Class Tax Relief and Job Creation Act of 2012. The prototypical search requirement calls for claimants to contact some minimum number of potential employers each week. In some cases, these contacts are required to be in person, while some more general requirements simply disallow phone calls. The most general requirements simply ask claimants to contact employers in the way that is customary for their professions. In general, the same employer cannot be con-

5

tacted again for a minimum number of weeks or unless there is reason to believe that another position has become available. Currently, virtually all states require claimants to at least track their contacts in a diary or work search log. Blank search logs are often provided to claimants with the rest of their UI documentation. However, there is considerable variation in how often these logs are checked by state workforce agencies. A steeper and increasingly common requirement is that claimants must report the details of their employer contacts at the time of making their weekly or biweekly claims. Claimants who are found to have not fulfilled their work search requirements will be deemed ineligible for the week and, potentially, disqualified from receiving future benefits. While search requirements have become common across states, they do not apply to all workers. In particular, most states exempt claimants who find work through a union hiring hall. They are not required to make regular contacts with other employers, though a weekly minimum may be set on the number of times a claimant must contact the hiring hall. Workers who are on layoff and awaiting recall can also be exempt, though they must often have a definite recall date within a set number of weeks. Claimants who are participating in agency-approved training programs may also be exempt.

2.2

Existing Evidence on Search Requirements

Characteristics of UI systems are studied extensively in the existing literature, but search requirements themselves receive somewhat less attention. Borland and Tseng (2007) is the only other study of which I am aware that examines broad changes in work-search requirements outside the specific context of an experiment. The authors examine a job-search diary program in Australia shortly after implementation in the late 1990s. Due to a labor dispute involving unemployment caseworkers, some benefit offices did not enforce the requirement that claimants keep a job-search diary satisfying a particular number of employer contacts.2 Those who kept the job-seeker diaries 2

In general, this requirement was for eight contacts every two weeks.

6

experienced shorter unemployment durations. However, part of the identification is generated from variation in the level of implementation between geographic regions and the study does not specifically try to estimate the implied effect of the policy under universal application. Further, workers with relatively poorer labor market options, as inferred from their recent unemployment histories, did not exhibit faster reemployment under the job-search diary regime. This finding in the Australian context is consistent with the idea that work search requirements are not effective in generating reemployment when workers are already constrained in their job market opportunities. Analysis of search requirements also appears in Keeley and Robins (1985) and Decker (1997). The two studies most relevant to the examination of search requirements in the US are Johnson and Klepinger (1994) and Klepinger, Johnson, and Joesch (2002). Johnson and Klepinger (1994) describes the Washington Alternative Work Search Experiment, in which the 9,634 eligible UI claimants who applied for benefits in Tacoma between July 1986 and July 1987 were randomly assigned to different work search treatments. The first group was exempted from search requirements and was not even required to file biweekly continuing claims forms. This resulted in a 3.34 week increase in UI durations over the reference group’s 14.48 average weeks. A second treatment group was assigned individualized work search requirements based on their circumstances. A third treatment group participated in an intensive job-search training workshop early in their employment spells. The individualized requirement group saw no change in benefits drawn while the workshop group drew, on average, half a week fewer benefits. The dramatic increase in UI durations for the first treatment group suggests that, in this context, work search requirements decrease UI claim durations. However, because the sweeping treatment effectively removed all costs of continuing to claim UI, the results likely overstate the effects of the search requirement alone. A follow-up analysis of this experiment in Lachowska, Meral, and Woodbury (2016) shows little overall effect of the experiment on long-term earnings or employment. This overall effect masks measurably faster returns to employment, higher earnings, and longer tenure with first reemployer for permanent job losers. These results indicate that the effects of search requirements

7

on post-reemployment outcomes are an important area for future research. The Maryland Unemployment Insurance Work Demonstration, as detailed in Klepinger, Johnson, and Joesch (2002), provides a sharper test of the direct experimental effects of work search requirements. During 1994, all new claimants at six randomly-selected Maryland UI offices were enrolled in the study. The experiment included treatment groups that faced increased search requirements, decreased search requirements, required participation in a job search workshop, and monitored work search. Both informed and uninformed control groups were included to test for Hawthorne effects. The simple changes in work search requirements are of greatest interest for the purposes of this paper. The decreased search treatment required no contacts as compared to Maryland’s standard of two, while the increased treatment required four weekly contacts. The zero-contact group saw an increase of 0.36 weeks claimed over the control mean of 11.94, and the four-contact group saw weeks claimed fall by 0.72. While the results are less striking than those seen with the sweeping treatment in Washington, they suggest a distinct slope in required contacts. In the case of both experiments, though the effects are well-estimated in partial equilibrium, there is little scope for considering the general equilibrium effects of search policies. Even if the entire labor market were randomized into the experiments, which arguably happened in Tacoma or at each site in Maryland, only some of the claimants saw their work requirements change, and they were always counteracted by a treatment group changing in the opposite direction. Another strand of the literature examines the effects of differential monitoring under constant search requirements. In two papers, McVicar (2008, 2010) uses the refurbishment of unemployment benefit offices in Northern Ireland as a source of variation in the monitoring of claimant search requirements. On the whole, his findings show that eliminating monitoring reduces the unemployment exit hazard and job entry hazard and increases the stock of claimants. If we interpret these findings as being equivalent to moving between a no-search requirement (when claimants are not monitored) and a standard search requirement (under standard monitoring), we would expect changes in search requirements to induce changes in unemployment exit and job entry in other

8

contexts. However, experimental evidence from four US states in Ashenfelter, Ashmore, and Deschênes (2005) suggests that UI claimants complied with work-search requirement even under very limited monitoring regimes. Performing additional verification of reported work-search did not induce meaningful changes in the duration of unemployment claims. As these studies do not address general equilibrium issues, it may be important to consider them as in Davidson and Woodbury (1993), Lise, Seitz, and Smith (2004), and Crépon et al. (2013). Both Davidson and Woodbury (1993) and Lise, Seitz, and Smith (2004) consider the general effects of some manner of reemployment bonus. In both cases, a model that allows for general equilibrium effects is calibrated using the partial equilibrium results of an experiment. Davidson and Woodbury (1993) then discuss the implied equilibrium effects of the policy, while Lise, Seitz, and Smith (2004) compare the model’s predictions to an out-of-sample group before using the results to identify feedback in the policy. In both cases, partial equilibrium experimental results are at least partially reversed when implemented in general. Crépon et al. (2013) endeavors to explicitly measure spillover effects of a French job search assistance program through a two-level randomization process. Municipalities were first randomized into groups that would vary the share of unemployed that would be treated with the program and then the appropriate percentage of individuals within each municipality were randomly treated. They find spillover effects in markets where the treated individuals compete for jobs mostly with other similar individuals and when labor market conditions are poor. This last result relates most strongly to the overall thrust of this paper. Overall, this literature indicates that there is scope for analysis of policies in which general effects should be considered.

3

A Model of Search Requirements

This section details a simple model of vacancy-posting and work-search behavior. Similar to the model appearing in Crépon et al. (2013), it maintains the key features of models appearing in Michaillat (2012) and Landais, Michaillat, and Saez (2016), but

9

focuses on the role of search behavior and its impact on equilibrium under different market conditions. The model employs the simplest kind of rigid wage—one that is fully fixed—and allows for diminishing marginal product of labor in the production technology. With this latter feature, the model demonstrates that, unlike canonical search-and-matching models, search behavior changes have greater effects when the labor market is strong than when the labor market is weak. Further, it formalizes the argument that the implication is the same even if search requirements have no direct effect on search behavior and simply operate as a leisure tax.

3.1

Environment

The model takes place in discrete time. At the beginning of each period t, unmatched workers search for jobs and hires are made. Production then takes place and wages are paid. A share λ of existing employment matches are then exogenously destroyed and the period ends. Both workers and firms have discount factor β. The labor force is measure one, with nt employed workers and ut unemployed workers so that nt + ut = 1. Matches occur between the measure of unemployed workers, who exert average search effort st , and posted vacancies, vt , according to a constant returns to scale (CRS) matching function

mt = m(st ut , vt )

(1)

which is increasing in both of its arguments. Given the CRS assumption, matches per vacancy can be expressed as a function of average search effort, st , and labor market tightness, θt ≡

vt 3 ut : mt vt

= m(st uvtt , 1) = q(st , θt ).

(2)

An individual worker exerting sit efficiency units of search is providing search effort that is a ratio

sit st

of average search effort and receives matches at this ratio times the

3

This setup allows me to examine the equilibria in n-θ space, with search effort as an additional variable that affects the equilibrium. Another option would be to define θ ≡ v/su.

10

average matches per unemployed worker:

sit mt st ut

=

sit vt st m(st , ut )

=

sit st f (st , θt ).

(3)

As in Pissarides (2000), I consider only symmetric Nash equilibria in which all workers exert average search effort (sit = st ). Thus, the transition probability for the representative worker is given by f (st , θt ). The vacancy and representative worker transition probabilities are related through f (st , θt ) = θt q(st , θt ). In steady-state equilibrium, flows into and out of employment are balanced, defining the steady-state relationship between employment and labor market tightness for a given level of search effort:

nt =

3.2

θt q(st , θt ) . λ + θt q(st , θt )

(4)

Worker Behavior

The level of search effort depends on worker behavior. Workers are assumed to face disutility from search effort, given by ψ(sit ), which is increasing and convex in its argument. Workers who are unemployed at the beginning of a period set their search intensity and, if they match with an employer, become immediately productive. An unemployed worker who can freely choose search effort will select their individual optimal effort, s∗it , to satisfy ψ 0 (s∗it ) =

f (st , θt ) [Wt − Ut ], st

(5)

where Wt is the value of being employed and productive in the period and Ut is the value of remaining unemployed. The marginal disutility of search effort is set to equal the marginal effect of search effort on job-finding times the asset gain from reemployment. Restricting to steady states and focusing on symmetric Nash equilibria in which all workers exert the same search effort, this condition can be shown to be

ψ 0 (s∗t ) =

f (s∗t , θt ) w − b + β(1 − λ)ψ(s∗t ) , s∗t 1 − β(1 − λ)(1 − f (s∗t , θt ))

11

(6)

where w is the (rigid) wage and b is unemployment compensation. The derivation of this condition appears in Appendix A.

3.3

Firm Behavior

The firm side of the market uses inputs of labor, nt , and technology, at , to produce output via the production function F (nt , at ). A firm makes hires by posting vacancies at cost cat . Each of these vacancies yields q(st , θt ) hires. Thus, the firm posts 1/q(st , θt ) vacancies to make one additional hire, and the cost of that hire is given by

cat q(st ,θt ) .

Labor demand is given by the first order condition on nt in the firm’s profit equation, which in steady state reduces to:

w+

cat cat = Fn (nt , at ) + β(1 − λ) , q(st , θt ) q(st , θt )

(7)

where Fn is the derivative of the production function with respect to n. This condition is derived in Appendix B. Vacancies are opened, increasing θ, until the costs of an additional hire is equal to the benefit of the additional worker. The costs appear on the left side of the equation and include the wage paid to the worker and the recruiting cost. The benefits include the marginal product of the worker and savings on future recruiting costs, appropriately discounted for time and the possibility of an exogenous separation.

3.4

Equilibrium

Steady-state market equilibrium can be shown in n-θ space where the Beveridge curve defined by equation (4), which balances flows into and out of unemployment, intersects the job creation curve defined by equation (7). Both of these depend on worker search effort, s, which is either a function θ as defined by equation (6) or is exogenously set, for example, through governmental rule. The conclusions presented here feature the jobrationing conditions in which search is more effective in tight labor markets. As shown by Michaillat (2012) and Landais, Michaillat, and Saez (2016), the equilibrium in these models is fundamentally affected by the production technology and the wage-setting

12

mechanism. A flexible wage-setting mechanism or linear production technology would produce alternative predictions which, ultimately, are not supported by the empirical results in this paper.

3.5

Illustrative Simulation

The nature of this equilibrium and the theoretical effects of a search requirement can be demonstrated through a simple illustrative simulation. The parameterization for this simulation is intended to approximate flows at the weekly level and draws on a related simulation in Michaillat (2012). In addition to choosing the individual model parameters, the simulation requires functional forms for the matching function, the production function, and the disutility of search. The first two are chosen to be CobbDouglas, as is common in the literature. There is relatively less guidance on the form of search disutility, but I follow the functional form of a simulation in Chetty (2008) s1+κ

t and parameterize it as ψ(st ) = ν 1+κ , with ν and κ both assumed to be positive.4

Additional details of this simulation are not the focus of this paper, so they appear in Appendix C. Four different equilibria of the simulated model are displayed in Figure III. For this initial exposition, search effort is exogenously set and is not determined in the model. As in Michaillat (2012), Crépon et al. (2013), and Landais, Michaillat, and Saez (2016), they are graphed in n-θ space, with full employment at the far right. In this arrangement, the Beveridge curves increase convexly from the lower left to the upper right. The left panel displays equilibria under a high value of productivity, a, and the right panel displays equilibria under a low level. Each panel displays two equilibria, one under a high level of search and one under a low level.5 The key difference between the two panels is the location of the job creation curves, which fall as employment 4

It should be noted that in Chetty (2008), s is scaled to directly correspond to the probability of jobfinding. In the model in this paper, job-finding probability is a function of s and θ. Thus, while the form of disutility matches that of Chetty (2008), the parameters of the disutility function are likely to be quite different. 5 The particular values of search intensity are somewhat arbitrary but are chosen to illustrate differences in equilibria. The increase in search effort from s = 1 to s = 2 has the equivalent effect on number of matches as doubling the number of unemployed workers and keeping search effort constant.

13

increases. The key element leading to this characteristic is the diminishing productivity of labor in production: the value of θ that solves equation (7) falls as n increases. In canonical search-and-matching models, production does not exhibit diminishing returns to labor and the second derivative of the production function with respect to labor is zero (Fnn = 0). A single θ thus solves equation (7) for any value of search effort. If Fnn < 0, firms choose a lower level of θ as n rises. In both cases, the two effects of a change in search effort are clear: the job creation curve rotates to the right and the Beveridge curve shifts down and to the right. The rotation in the job creation curve is owing to the effect search behavior on the firm’s condition given in equation (7). As workers search harder, it is less costly for firms to find a match. The search frictions associated with matching are diminished and more vacancies are opened, leading to an increase in labor market tightness. At the same time, the change in s affects the equilibrium flows condition given by the Beveridge curve in equation (4). Higher search effort increases the match rate from the workers’ perspectives, increasing employment at each level of tightness. When productivity is high, as in the left panel of Figure III, unemployment is low and is entirely frictional. Much of the unemployment can be reduced by lowering the hiring costs for firms. When productivity is low, as in the right panel, total employment is low even in the absence of matching frictions because of the rigidity of the wage and the diminishing productivity of labor (Michaillat 2012).6 There is much less scope for increased search effort to make up ground in this scenario. In the definitions of Michaillat (2012), unemployment in this environment is largely due to rationing and not frictions. In terms of changes in search effort, the effect on employment is larger in the high productivity case. Owing to the convexity of the Beveridge curve, the opposite prediction is attained in a canonical search-and-matching model with linear production and flexible wages (Landais, Michaillat, and Saez 2016). In that canonical case, unemployment is entirely due to matching frictions and is most reduced by extra 6

The wage in this model is fixed, as in Crépon et al. (2013), but the same qualitative result attains with a variable but rigid wage.

14

search effort when unemployment is high.

3.5.1

Optimal Search under Search Requirements

The equilibria displayed in Figure III assume fixed levels of search effort across all displayed values of labor market tightness. If search effort is instead chosen endogenously by workers, it is chosen to satisfy the condition given by equation (6). In this parameterization, equation (6) induces a positive relationship between labor market tightness and search effort. As the labor market tightens, the marginal effect of additional search effort increases, leading to a higher level of search at which the marginal benefit is equal to the marginal cost. This effect is displayed in the left panel of Figure IV. As labor market tightness rises, the optimal choice of search effort increases, shifting the equilibrium from one Beveridge curve with fixed search effort to a higher one. This produces a Beveridge curve with variable search effort that is flatter than any individual fixed-effort Beveridge curve. If a UI agency is able to enforce a floor on search effort, this floor will bind at low levels of labor market tightness. When labor market tightness is high enough, optimally-chosen search intensity will exceed the floor. In terms of the Beveridge curve, the end result would be a composite curve that is the lower envelope of the variableeffort Beveridge curve and one fixed-effort Beveridge curve like those displayed in the left panel of Figure IV.7 This relationship suggests that the effect of a search floor on search effort should be even larger when the labor market is weak because the floor is most likely to bind. If search requirements are binding in a weak labor market, their impacts should be most likely to be visible there. If this is the case, it should bias against the result suggested by the relationship described in Figure III, which is seen in the empirical evidence in this paper. 7

The empirical validity of this prediction is open to some debate. While a positive relationship between search effort and tightness is common to this kind of model, Shimer (2004) suggests an alternative model in which job applications decrease as the probability of job-finding increases. Recent evidence by Gomme and Lkhagvasuren (2015) combining American Time Use Survey (ATUS) data with CPS data suggests that procyclical search effort is supported by the data.

15

3.5.2

Unproductive Search Requirements

One potential concern is that the oprationalization in the prior section is not a realistic form of search requirements. It is possible that work-search tests as they are implemented in most US states produce no benefit in job-finding. If, for example, workers only find jobs through informal channels or through contacts that are not deemed acceptable to workforce agencies, search requirements should not be expected to decrease unemployment durations or increase employment rates. However, if this is true, the search requirements may still operate as a leisure tax for the unemployed. The additional hurdles to UI claiming would lower the flow value of unemployment, a kind of effect that has previously been documented in Black et al. (2003). In the context of the model, this is the same as lowering b and raising the relative value of employment. If this is the case, then workers will optimally increase their actual, productive search effort. The effect of such a decrease on the flow value of unemployment is displayed in the right panel of Figure IV. If b falls, workers optimally increase their search effort and produce a rightward shift in the Beveridge curve. This channel is similar to more conventional analyses of UI systems in which UI policy, mostly through benefit levels or durations, has an indirect impact on search effort through its effects on flow utility in the unemployed state.

4

Policy Measurement

Although state-specific search requirements are quite common, documentation of the policies is not readily available at most points in time. The DOL reported state search requirements in its annual Comparison of State Unemployment Insurance Laws through 1999, but it was determined that there was not enough cross-state variation at the time to justify continuing to include search requirements. Around the same time, Anderson (2001) performed a cross-sectional review of the standing search requirements, notably finding that some workforce agencies’ publications suggested different rules from those listed in the DOL report. Until search requirements were added back into the Comparison of State Unemployment Insurance Laws in 2012, information on policies

16

is largely limited to O’Leary (2004), who provided a cross-section of rules as reported in a survey of state workforce agencies. A handful of other papers mention the search requirements for individual states at various points in time over this period. I overcome the lack of existing information on search policies by constructing new data from state workforce agency publications. All states make some documentation available to claimants, whether through stand-alone UI claimant handbooks or “frequently asked questions” brochures and web pages. Many of these instruct claimants on the exact search requirements they are expected to fulfill. Agency-provided work search logs, which have become more common in recent years, also often detail the rules that claimants are supposed to follow. Although most of these are not directly available online, many are archived through Internet crawler caches and others are available through university and state government libraries. Through these sources, I have collected all relevant and available documents published by workforce agencies between 2001 and 2015. Unfortunately, it is difficult to be certain that I have successfully found every publication over this time period. I am most confident that I have gathered all relevant information starting in 2005, so the analysis in this paper focuses on the period from 2005 to the end of 2014. I examine these documents for information on work search requirements and track within-state changes via document publication dates. The rules implied by the agency publications generally correspond with the other available sources, but there are some discrepancies with DOL’s Comparison of State Unemployment Insurance Laws and O’Leary (2004). For the sake of consistency, I defer to the agency publications throughout. I also argue that the rules as they are described to claimants in this information are the most relevant for measuring the policies. However, I cannot rule out that workforce agency staff provide different information in person. In Sections 4.1 and 6.1, I compare the rules recorded from agency publications to actual numbers of employer contacts reported by claimants in BAM data.

17

4.1

Changes in Policies

I group search policies by the number required weekly employer contacts. Since the year 2000, various states have specifically required claimants to make between zero and five contacts. States that do not make these specific requirements fall into two groups. First, many states do not indicate an exact number of employer contacts. In general, these are the states listed as “no specific number” in Comparison of State Unemployment Insurance Laws and recorded in O’Leary (2004) as states in which claimants are instructed to follow the customs of hiring in their profession. Throughout this paper, I refer to these policies as “nonspecific” search requirements. A second group of seven states provides search requirements that are in some way individualized for claimants. My research suggests that Arkansas, Colorado, Idaho, Missouri, Ohio, Texas, and West Virginia have all had such “directed” search policies at some point since 2000. I exclude these states from my analysis both because the determinants of the individual requirements are not always clear and because they generate cross-sectional within-state heterogeneity that makes analysis at the state level difficult. Figure I displays the number of states with each policy from zero to five required employer contacts over the ten years from 2005 through 2014. While there are a few changes over the first two-thirds of this sample period, the rate of changes clearly accelerates starting around 2011. The increases are also concentrated at the high end of the requirement distribution. While only four states required at least three weekly employer contacts at the beginning of 2005, 24 required three or more contacts by the end of 2014. Common narratives surrounding the policy changes indicate that search requirements have been raised to get the unemployed back to work by increasing labor force attachment and more or less explicitly raising the burden of drawing UI benefits. Figure II displays the number of employer contacts reported to the BAM program by audited UI claimants. The BAM program is designed to identify the sources of overpayments in the UI system. It accomplishes this task by randomly selecting UI claimants and thoroughly checking their monetary and nonmonetary eligibility for a given week.

18

As part of the audit, claimants are asked to record their requisite employer contacts for the week, even if they are not generally required to explicitly report contacts. While the question of whether the recorded contacts satisfy the actual requirements is of interest, my first goal with this data is to see how many contacts claimants think they should be reporting. The seven different search policies examined in this paper are listed along the horizontal axis of Figure II. The bars above each policy number show the discrete empirical PDF of contacts reported by the 87, 712 claimants who were audited by BAM between 2005 and 2014 and were not union- or attached to an employer with a definite recall date. In general, the patterns suggest a strong relationship between the policies I have recorded and the BAM reports. For each of the specific policies, the required number makes up a plurality of the reports. It is easier to explain apparent overreports than underreports because claimants have an incentive to list extra contacts if they are concerned that some of their contacts will be deemed ineligible. Thus, the most striking deviations are for states with three-contact policies, which have over 30 percent of claimants reporting zero or one contacts.8 On the whole, however, the BAM data are supportive of the policies as they are coded. The relationship between BAM contact reports and state changes is displayed in Figure V. For each of the 20 states with a policy change during the period available in the BAM data, the graphs indicate the policy (as indicated by the thick, dashed line) and the monthly average of reported contacts (the thin, solid line). No dashed line is displayed for periods during which a nonspecific policy prevailed. A number of the changes appear only at the very end of the available BAM data, with Alaska, Georgia, Mississippi, New York, and Oregon having only brief available data after the implementation of search policies. In the available data, Arizona, Florida, Georgia, Hawaii, Nebraska and Tennessee’s all suggest reasonably close adherence to their new 8

Further examination of this pattern indicates that almost half of these observations are from Massachusetts. Bay Staters almost exclusively reported zero contacts over the vast majority of the sample period even though Massachusetts’ UI claimant handbooks very clearly require three contacts throughout. As several other sources also suggest a three-contact coding in Massachusetts, it appears that there is some idiosyncrasy with the state’s BAM codings. Efforts to learn more about this from Massachusetts workforce authorities have been unsuccessful.

19

policies after these states switched away from nonspecific rules. Likewise, North Dakota, Pennsylvania, Utah, and Wisconsin all generally follow changes in their explicit policies. While the data do seem to reflect the more recent change in Louisiana’s requirements, they follow the earlier increased period somewhat less closely. Overall, there is some variability in these averages over time, but the measured search requirement policies appear in line with the number of contacts actually reported to BAM.

5

Primary Empirical Strategy

The primary analysis in this paper involves the estimation of regressions on aggregate and individual outcomes with state fixed effects and time fixed effects. In particular, I consider individual-level models of the form

yist = α0 + α1 Dst + α2 xist + α3 Xst + γs + δt + ist

(8)

where yist is an outcome for individual i in state s and month t, and Dst is a vector measuring the prevailing search policy. Controls for individual characteristics are included in the vector xist and aggregate state-level controls are included in the vector Xst . State and month fixed effects are given by γs and δt , respectively, and ist is an idiosyncratic error term. The coefficient vectors of interest are given by α1 . These coefficients estimate the difference in outcomes between state-months with differing search requirements. The estimates are identified by within-state variation in policies over time. I parameterize the policies in two different ways. First, I estimate the effect of moving through the zero- to five-contact policies with a continuous linear measure of the number of required contacts. While this specification is simple in that it provides a single estimate of the effects of an additional contact, it constrains that effect to be constant as the number of contacts increases. In these regressions, I include nonspecific policies as a separate indicator and assign them a zero in the continuous measure. The coefficient estimate on the indicator indicates the estimated difference between zero-contact and nonspecific

20

policies. Second, I estimate the effects of each policy individually by including a full set of indicators, one for each policy with zero-contacts omitted.9 This specification is clearly more flexible, but suffers from relatively few observations for some of the policies. The precision and reliability of the estimates are thus reduced. To ameliorate the problem slightly, I group policies requiring greater than four contacts into a single measure. No matter the exact policy coding, consistent estimation relies on the assumption that there is no correlation between search policies and outcomes for the sampled respondents, conditional on the fixed effects and covariates. One source of concern is the possibility of policy endogeneity: states with weak labor markets may implement policy increases in response. To control for this possibility, I include controls for the unemployment rate as measured by the Bureau of Labor Statistics’ (BLS) Local Area Unemployment Statistics (LAUS) program. Depending on the model, these controls are either at the state or CBSA level. Another concern is correlation between search policies and other aspects of state UI systems. A long literature demonstrates the importance of Ui benefit levels and durations on outcomes for claimants, so it is important to control for changes in these policy characteristics. In general, the data I examine do include the necessary measures to calculate benefits or durations at the individual level. Therefore, I include controls for average benefits and average potential benefit durations at the state-quarter level as reported to the DOL. These measures should control for major changes in UI policy within states over time.

6

Evidence on Requirements Affecting Search Effort

In this section, I examine the estimated effects of search requirements on measures of search effort using data from the BAM program and the CPS. The econometric models are broadly of the form described in Section 5. All regressions include demographic, occupation, and industry codes. Table I displays the estimates from regressions of the form of equation (8). The first two columns regress the number of contacts reported in 9

I choose the zero-contact policies as the omitted group for these specifications because it seems most straightforward to compare more stringent policies to a simple zero-contact requirement.

21

the BAM data on measures of search policies, quantifying the effects shown in Figure V. Columns (3) through (6) use different measures of genuine, acceptable employer contacts as the dependent variables. Columns (7) through (10) extend the analysis to more general measures of search methods and effort in the CPS.10

6.1

Employer Contacts Reported to Auditors

The number of employer contacts reported in the BAM data is regressed initially on a linear measure of required contacts with an indicator for nonspecific policies. Claimants living under nonspecific policies are assigned a zero in the linear term. Thus, the estimate in the first row indicates the number of additional reported contacts associated with a one-contact increase in the requirement. The estimate in the last row indicates the conditional expectation of the difference between reports under a nonspecific policy and a zero-search policy. A single additional required contact is estimated to increase reported contacts by 0.571 and claimants under nonspecific policies are estimated to report 0.828 more contacts than claimants under zero-search policies. The estimates reflect a number of patterns apparent in Figures II and V. First, nonspecific policies are associated with some variation in number of reported contacts, but the average level is greater than that of the lowest specified search policies. Second, the estimated marginal effect of increasing a search policy is statistically significantly less than one, though much greater than zero. This likely reflects both measurement differences between the BAM data and the policies11 and less-than-full adjustment to policies by claimants. Note that this latter effect does not imply anything about noncompliance: because the share of claimants exceeding the requirement decreases as the policy increases, the estimated effect of the policy should be attenuated below one. The second column of Table I further demonstrates this pattern. When estimated with individual indicators, the number of reported contacts increases as search requirements get more stringent, but the marginal 10

Other research has also used data from the ATUS to examine search effort. Unfortunately, owing to the relatively small sample size in that data and the limited variation of interest in this paper, estimates using the ATUS are quite sensitive and imprecise. 11 See discussion of discrepancies in Section 4.1.

22

differences are less than one. The estimates in columns (1) and (2) indicate that higher requirements are generally associated with more reported contacts. However, it is initially unclear whether these reports reflect actual contact with employers or fake contacts by claimants trying to appear as though they have satisfied the requirement. To separate these two possibilities, I use data on the results of BAM audits themselves. As part of determining claimant eligibility, BAM auditors attempt to verify that the employer contacts were made and were legitimate under state rules. Ultimately, employer contacts are identified as acceptable, unacceptable, or not verified. A large portion of the contacts (42%) fall into this last category, indicating that there was not enough information to rule the contact as acceptable or unacceptable.12 Of those that can be verified, approximately 88% of the contacts in the sample are found to be acceptable. Columns (3) and (4) of Table I repeat the same specifications as the first two columns but use the number of contacts that are not found to be unacceptable as the dependent variable. This measure is an upper bound on the number of contacts that would be found to be acceptable if all contacts could be verified one way or the other. Columns (5) and (6) use a lower bound: the dependent variable is the number of reported contacts that are found to be acceptable. In either case, the coefficient on the linear measure remains positive and significant. While only half of contacts are unverified, the linear estimate for the verified, acceptable contacts in column (5) is attenuated rather more than half as compared to column (1). This suggests that, on average, the additional contacts provided under higher search requirements are found to be unacceptable at a higher rate, but the total measured search effort is increasing regardless.

6.2

Search Effort Proxies in the Current Population Survey

A limitation of the analysis based on reported contacts is that, even if all contacts are genuine, the differences across policies may simply represent differences in reporting. The above results could be generated by all claimants actually contacting well more 12 For the purposes of determining eligibility for UI benefits, the audits generally treat these contacts as acceptable. Payments are not ruled as improper because of unverified contacts.

23

than the requirement and only reporting the required number. Therefore, I turn to the CPS for additional measures of search effort that are not driven directly by reporting requirements. Columns (7) and (8) of Table I demonstrate that search requirements broadly increase the number of job-search methods used by basic monthly CPS respondents between 2005 and 2014. Unemployed CPS respondents are asked to indicate which of twelve possible job-search methods they have used in the past four weeks. These regressions estimate the policy effects on the total number of indicated methods among those respondents who indicate they were laid off from or otherwise lost their most recent job. This is intended to capture the population of the likely-UI-eligible unemployed. It is important to note that the population differs from that sampled by the BAM data, which restricts to UI claimants. Measuring the usage of the 12 search methods described in the CPS provides a reasonable proxy for search effort. Because job-seekers are likely to diversify the ways in which they look for jobs as they put more effort into search, these measures provide readily-available proxies for effort. On a sample average of approximately 2.5 methods, the linear estimate in column (7) suggests that an additional required contact increases the number of search methods used by 0.02. In terms of the overall point of this paper, the magnitude of this estimate is less important than the fact that it is positive and measurable. The result indicates that there is a detectable increase in search activity, on average, as search requirements increase. The individual policy indicators in column (8), however, show a less clean pattern than in the results using BAM data. The indicator for three-contact policies has the largest coefficient, while a smaller but positive coefficient is associated with one-contact policies. Columns (9) and (10) of Table I display the effects on whether respondents made use of one particular search method: directly contacting or interviewing with employers. This method was chosen because it appears to correspond most directly to the definitions of employer contacts in search requirements. The linear estimate is again positive and significant, indicating that an additional contact requirement raises the probabil-

24

ity of contacting employers directly by 2.5 percentage points. The specification with indicators shows that the effect is largely driven by a difference between zero-search policies and the other policies. All of the estimates, including the one for nonspecific policies, put the effect between 0.083 and 0.143. In general, this result should not be surprising. If respondents in states with one-contact policies are following the requirements and making at least one employer contact, there should not be much effect on this probability from adding additional contacts. Table II presents the estimated linear effects on each of the search methods in the CPS individually. The first row repeats the linear estimate from Table I column (9). The remaining rows repeat the same exercise for the rest of the possible search methods. In addition to eh employer contacts measure, significant positive coefficients are also estimated for contacting both public and private employment agencies. The largest negative effects are associated with contacting friends or relatives and sending out resumes or filling out applications, though only the former is significant at conventional levels. These results may reflect the emphasis workforce agencies place on in-person employer contacts over networking and sending out resumes. If this emphasis is illadvised, or particularly ill-advised in certain labor market conditions, that would be a concern for the conclusion that search requirements have any positive impact on search.

7

Differential Effects by Market Conditions

The theory in Section 3 suggests that the efficacy of search requirement policies may vary considerably across labor market conditions. If unemployment is always driven by matching frictions, then search requirements may be very effective in reducing unemployment during recessions due the slack in the labor market. If, on the other hand, jobs are rationed in recessions because of wage rigidities, then search requirements will have little ability to reduce unemployment. Therefore, at the level of the labor market, I test whether search requirements are effective at reducing unemployment, allowing the effects to vary according to labor market strength. One strategy for examining differential effects across market conditions would be to

25

simply interact the policy measures with the unemployment rate, a commonly-chosen measure of labor market strength. However, in the current analysis, it is also the outcome of interest. Therefore, I use two plausibly-exogenous measures of market strength. First, I separate labor markets by lagged unemployment, testing to see whether the implementation of a search requirement differentially impacts the equilibrium in labor markets that were weaker before the policy change. Second, I use a shift-share measure of employment-by-industry to proxy local labor market demand with national employment trends.

7.1

Equilibrium Effects by Lagged Unemployment

Within states that implement changes to their search requirements, I explicitly examine differences across labor markets by their conditions before implementation. If search requirements are more effective when the labor market is weak, then unemployment should fall relatively more for markets with initially high unemployment. If search requirements are not as effective in weak markets, then unemployment rates should fall relatively more in the initially low unemployment locations. I implement these tests via baseline specifications of the form umt = αLm 1(t > t∗ ) + γm + δt + εmt ,

(9)

where umt is the unemployment rate in market m at time t, Lm is an indicator for being a low-unemployment market prior to implementation, which is multiplied by an indicator function for time t being post-implementation. Market and time fixed effects are given by γm and δt . Estimates of the coefficient α indicate the average change in unemployment differential between high- and low-unemployment markets following implementation of a search requirement increase. Negative estimates of α show that lowunemployment markets had relatively even lower unemployment after implementation. Positive estimates of α show that low-unemployment markets had relatively higher unemployment after implementation. The former suggests that search requirements are relatively more effective in low unemployment markets, while the latter suggest

26

that they are relatively more effective in high unemployment markets. I estimate regressions of the form of (9) using LAUS data for CBSAs that do not cross state lines. These data are useful in that they provide monthly estimates of unemployment and labor force participation at relatively disaggregated levels. A disadvantage of the data are that they are partially constructed by BLS models that combine data from a number of different sources. Unemployment estimates at the level of the labor market are constructed from data on current and past UI claims and estimates of new entrants and reentrants into the labor force.13 Entrants and reentrants are estimated using state-level estimates of these groups modeled from current and past CPS data, which are then divided into market areas based on the relative age distributions of the markets. Each market is assigned new entrants in proportion to its share of the state’s age 16–19 population. Reentrants are assigned based on the market’s share of the population ages 20 and over. In essence, the process combines high-quality data on local unemployment claims, which are not always otherwise available, with an averaged apportioning of entrants. The estimates may be biased if state policy changes are correlated with changes in the number of state entrants and reentrants and these groups are differently assigned to markets defined as high- or low-unemployment before implementation. It is difficult to explicitly control for these concerns because the BLS does not reveal the exact models used in creating LAUS data. I therefore simply proceed using the LAUS data as they are published. Table III reports the regression results using data from 2005 to 2014 in Louisiana, North Dakota, Pennsylvania, Utah and Wisconsin. These are five states explicitly increased from one specific number of required contacts to a larger one. In the case of multiple policy changes in these states, the most recent change is examined. Alaska similarly made an increase from no requirement to a specific number of contacts but is excluded because of its small number of CBSAs combined with the lateness of its 13

This description draws heavily www.bls.gov/lau/laumthd.htm.

on

the

LAUS

27

estimation

methodology

details

found

at

policy change in the sample period. Owing to the small number of CBSAs in some of the regressions, I also report wild bootstrap p-values for the nulls that the estimates are zero (Cameron, Gelbach, and Miller 2008). Within each state, I divide the CBSAs by their unemployment rates pre-implementation using two measures. The first row of results in Table III are generated defining lowunemployment areas as those that had average unemployment rates below the state median between 2005 and the implementation of the policy change. The second set of results defines low-unemployment areas as those that had average unemployment below the state median in the year before the policy change. The former identifies areas that are consistently low in unemployment, while the latter identifies those that may have been transitively so. The estimates are almost universally negative, the only exception being the first estimate for Wisconsin, suggesting that low-unemployment markets do relatively better after a search requirement increase under either definition. However, both the standard errors and wild bootstrap p-values indicate that many of the estimates are not statistically distinguishable from zero. The exceptions, at conventional significance levels, are North Dakota, Pennsylvania, and one of the Utah estimates. On the whole, these results indicate at least some evidence that the search requirements were relatively more effective in the stronger labor markets.

7.2

Equilibrium Effects by Industry Shift-Share

I next proxy for labor market strength using a shift-share measure as in Bartik (1991). This method uses changes in the national distribution of employment-by-industry to proxy for local labor demand in individual markets with different industry mixes. Intuitively, if employment in manufacturing declines nationally, one would expect markets that have larger shares of workers in manufacturing to see larger employment declines. If the national movements in industry employment are exogenous to an individual market’s local labor conditions, then the measure is a plausibly exogenous measure of local labor demand. In practice, I calculate the proxy using predicted employment in market m at time

28

t as ˆmt = E

X Nkt Emkb , Nkb

(10)

k

where b is a chosen baseline date and k indexes industries. Nkt is national employment in industry k at time t, Nkb is national employment in the industry at baseline, and ˆmt is predicted Emkb is employment in market m in industry k at baseline. Thus, E using baseline employment in each industry (Emkb ) multiplied by national growth in Nkt that industry since the baseline ( N ) and summed across industries. For the purposes kb

of the regressions that follow, I use predicted employment growth, ˆ ˆ mt = Emt − Emb , G Emb

(11)

which has the advantage of being the same scale for all markets. While this measure can be used as an instrument, I interpret it directly and estimate reduced-form regressions given by

ˆ mt + α3 Dst + α4 Dst G ˆ mt + γm + δt + mt . umt = α0 + α1 G

(12)

The estimates of α4 indicate the differential effect of a search policy when predicted employment growth is 100 percent higher. The inclusion of CBSA fixed effects and time fixed effects make these coefficient estimates conditional on baseline differences ˆ mt itself is a predicted and average changes across CBSAs. In other words, although G growth rate, the coefficient estimates are identified off of relative differences in predicted growth rates across areas. Negative estimates for these coefficients suggest that search policies do more to lower unemployment when labor demand is relatively stronger. I calculate the shift-share proxy using the Quarterly Workforce Indicators (QWI) for all available CBSAs. The QWI provide quarterly estimates of various employment stocks and flows using Longitudinal Employer-Household Dynamics (LEHD) microdata. The QWI are sourced with high-quality administrative data from a number of sources, but have some noise infused to protect individual confidentiality. For the purposes

29

of constructing the shift-share measure, I use employment counts in two-digit North American Industry Classification System (NAICS) sectors at the level of the statistical areas. I use a baseline of the second quarter of 2005 because it is before most of the policy changes of interest, but is at the point at which almost all states appear in the QWI.14 The first row of Table IV indicates the negative relationship between the shift-share measure and the local unemployment rate, again taken from LAUS data. In these ˆ mt are scaled as percentage points (e.g., the results, both the unemployment rate and G ˆ mt is 2). While there is some variability in the magnitude unemployment rate is 5 and G of the interaction estimates, they are universally negative even with the inclusion of a number of different included trends. Looking across specifications, a one percentagepoint increase in predicted growth strengthens the decrease in the unemployment rate from each additional required contact. Increasing from zero contacts to five should lower unemployment by 0.125 to 0.360 percentage points for each additional percentage point ˆ mt in the estimation sample is 1.86, outside of expected growth. The average value of G of specification a relative predicted growth rate of just over three is enough to swamp the positive main effect of the number-of-contacts parameter and induce a negative effect of contacts on unemployment. The nonspecific policies have both a larger main effect and a larger interaction estimate, though they require a relatively higher predicted growth rate in order to make the effect negative in a number of the specifications.

7.3

Micro Evidence on Reemployment Hazard

I next show that search requirements have positive effects on reemployment hazard as measured in micro data. I further test for whether these effects are attenuated when unemployment is high. Unfortunately, the precision of the estimates is extremely limited. Although the results point in the same direction as the results in the prior 14

Not all geographies are available in the QWI data. Massachusetts is does not appear until 2010 and is excluded from this analysis. Similarly, no estimates are available for New England City and Town Areas (NECTAs) in other states. Alternative data sources, like the Quarterly Census of Employment and Wages (QCEW) would not have these particular limitations, but face other limitations, like lacking information on individual micropolitan statistical areas.

30

sections, the estimates are effectively zeros. I perform these tests using Cox proportional hazard models with the same general form of controls as the linear models of the previous section. I estimate the models on a subset of monthly CPS respondents who can be linked across months. For the unemployed each month, failure is defined as reemployment the following month. In addition to the policy variables, I allow the hazard to vary proportionally in month, year, state, education, two-digit occupation occupation codes, two-digit industry codes, race, ethnicity, sex, the state unemployment rate, and average UI benefits and potential durations at the state-quarter level. Coefficient estimates from these models are reported in Table V. The first column reports the coefficients for a simple model in which the policy measures and the unemployment rate enter linearly, as they have in the previous models. The linear number-of-contacts and nonspecific measures are both positive, but with estimated standard errors of similar magnitude to the estimates. While these coefficients are not directly interpretable in terms of durations, they indicate that increased employer contacts in search requirements lead to faster reemployment. Perhaps the most interesting feature of these estimates is the relative size of the nonspecific indicator coefficient, as it implies an effect on the hazard as large as a five-contact requirement. The negative coefficient on the unemployment rate variable is unsurprising, as workers are slower to find reemployment in high-unemployment time periods. Of course, there is an issue with using the unemployment rate as an explanatory variable in a model were reemployment is the outcome, as the two are simultaneously determined in the actual labor market. This simultanaeity problem concerning labor market conditions is more full accounted for in the previous section. The second column of Table V reports estimates from a proportional hazard model in which the policy measures are also interacted with the unemployment rate. The coefficients on the main effects of the policies now report the effects in the (clearly out-of-sample) environment in which the unemployment rate is zero. Both estimates are larger than their counterparts in the first column, which is closely related to the

31

fact that the point estimates on the interaction terms are negative. The interaction point estimates suggest that as the unemployment rate increases, the reemployment effects of the search policies are diminished. The linear interaction term, in particular, is negative but effectively zero. Although these estimates are imprecise, they point in the same direction as the estimates in the previous section. To demonstrate the nature of this relationship, I graph the predicted survival curves for different unemployment rates and search policies using the interacted model. These curves are graphed in Figure VI. The low unemployment curves are predicted using an unemployment rate of 4.0 and the high unemployment curves are predicted using an unemployment rate of 12.0. These rates are chosen to approximate states near the low and high ends of state unemployment distributions. Survival curves for zero, two, and four contact search policies are predicted for each of these unemployment rates. As expected, given the signs of the estimates, the low unemployment curves have higher rates of reemployment. Within each unemployment, there is a much smaller difference between the different search policies, with faster reemployment predicted by the higher requirements. The interesting feature of these predictions, in line with the rest of the results in the paper, is that the spread between the search policy lines is wider when unemployment is low. This is consistent with the idea that search policies induce a greater effect in stronger labor markets. However, this is driven by the small and imprecise estimate on the interaction term in Table V.

8

Conclusion

This paper provides new reduced-form evidence that labor market policies aimed at increasing search effort are less effective at reducing unemployment in downturns. The results are consistent with job-rationing models in which frictional unemployment is relatively less important in weak labor markets. The results are complementary to findings presented in Michaillat (2012), Crépon et al. (2013), and Landais, Michaillat, and Saez (2016), but uses very different methods and a more direct analysis of search effort itself. The variation examined in this paper affects a very large number of workers

32

and exists in full general equilibrium. This paper documents the numerous changes in search requirements for UI claimants across states in the US. Increases in these requirements are shown to be associated with higher measures of search effort across multiple proxies. Using two different proxies for labor market strength, I show that policies have less of a reductive effect on unemployment when the market is weak. Micro-level evidence from duration models point toward the same conclusion, but are highly imprecise. Although UI search requirements may be unambiguously effective for small groups of workers in partial equilibrium, their effectiveness appears muted in general equilibrium in weak labor markets. Thus, this paper also contributes to a growing literature considering the roles of spillovers and externalities in interpreting the findings of randomized controlled trials. While policymakers may wish to continue raising search requirements as a way of increasing the burden of UI claiming, these changes are unlikely to improve outcomes for UI systems through faster reemployment in recessions. Other outcomes, like match quality, may be affected, but future work is needed to fully examine them.

Desmond Toohey, Department of Economics, University of Delaware

33

Appendix A

Worker Behavior

The value of being employed at the time of production is given by

Wt = w + β[(1 − λ)Wt+1 + λSt+1 ],

(13)

where w is the wage and St is the value of being an unemployed job-seeker at the beginning of period t. The value of being unemployed during the time of production is given by Ut = b + βSt+1 ,

(14)

where b is the flow utility received by the unemployed. The value of searching for a job at the beginning of a period is given by

St = −ψ(sit ) +

sit st f (st , θt )Wt

+ (1 −

sit st f (st , θt ))Ut ,

(15)

where ψ(sit ) is the disutility of exerting search effort sit , which is assumed to be increasing and convex in its argument. Workers take the variables st and θt as given. If workers are free to choose search intensity, then their choice will satisfy the first order condition given by ψ 0 (s∗it ) =

f (st , θt ) [Wt − Ut ]. st

(16)

Combining equations (13) and (14), we have

Wt − Ut = w − b + β[(1 − λ)(Wt+1 − St+1 )].

(17)

Using the value of St as indicated by equation (15) evaluated at symmetric equilibria where sit = st , this worker surplus becomes

Wt − Ut = w − b + β(1 − λ) [ψ(st+1 ) + (1 − f (st+1 , θt+1 ))[Wt+1 − Ut+1 ]] .

34

(18)

Workers’ chosen search intensity can be determined using equations (16) and (18). While next period’s search intensity and job finding rate can used to create a sufficient statistic for next period’s worker surplus (Wt+1 −Ut+1 ), it is more convenient to consider steady state equilibria. That is, in steady state, Wt − Ut = Wt+1 − Ut+1 and st = st+1 , so equation (18) can be solved for Wt − Ut and plugged into equation (16) to give the condition for optimal search intensity:

ψ 0 (s∗t ) =

B

f (s∗t , θt ) w − b + β(1 − λ)ψ(s∗t ) . s∗t 1 − β(1 − λ)(1 − f (s∗t , θt ))

(19)

Firm Behavior

The firm’s value of entering period t when it had nt−1 employees during the productive portion of the previous period, is given by

Π(nt−1 ) = max F (nt , at ) − wnt − nt

cat [nt − (1 − λ)nt−1 ] + βΠ(nt ), q(st , θt )

(20)

where nt − (1 − λ)nt−1 is hires made during the matching period. Vacancies are posted at flow cost cat . Each of these vacancies yields q(st , θt ) hires. Thus, the firm posts 1/q(st , θt ) vacancies to make one additional hire, and the cost of that hire is given by cat q(st ,θt ) .

The first order condition for an interior solution on nt is

w+

cat+1 cat = Fn (nt , at ) + β(1 − λ) . q(st , θt ) q(st+1 , θt+1 )

(21)

That is, firms hire until the marginal costs equal the marginal benefits. The costs include the marginal worker’s wage, the change in wages for all other workers, and the cost of posting the marginal vacancies. The benefits include the marginal production of the worker and savings on the following period’s hiring costs.

35

C

Simulation Details

The matching function is Cobb-Douglas, given by

m(su, v) = µ(su)φ v 1−φ ,

(22)

the disutility of search is given by

ψ(s) = ν

s1+κ , 1+κ

(23)

and F (n) = anα

(24)

is the production function. For parameters that are not well-established in the literature, these are primarily sourced from Michaillat (2012). The parameters of the search disutility function are chosen by the author to generate variation in the optimal search effort over the ranges of θ displayed in this paper. Similarly, the level of low productivity is chosen to be 0.96 and the low unemployment benefit is chosen as 0.1665 in order to show variation in the outcomes. A full listing of the parameters appears in Table C.I.

36

References Anderson, Patricia M. 2001. “Monitoring and Assisting Active Job Search.” In Labour Market Policies and the Public Employment Service. 217–239. Ashenfelter, Orley, David Ashmore, and Olivier Deschênes. 2005. “Do unemployment insurance recipients actively seek work? Evidence from randomized trials in four U.S. States.” Journal of Econometrics 125 (1–2):53–75. Bartik, Timothy J. 1991. “Who Benefits from State and Local Economic Development Policies.” W.E. Upjohn Institute for Employment Research: Kalamazoo, MI. Black, Dan A., Jeffrey A. Smith, Smith C. Berger, and Brett J. Noel. 2003. “Is the Threat of Reemployment Services More Effective Than the Services Themselves? Evidence from Random Assignment in the UI System.” American Economic Review 93 (4):1313–1327. Borland, Jeff and Yi-Ping Tseng. 2007. “Does a Minimum Job Search Rquirement Reduce Time on Unemployment Payments? Evidence from the Jobseeker Diary in Australia.” Industrial and Labor Relations Review 60 (3):357–378. Cameron, A. Colin, Jonah B. Gelbach, and Douglas L. Miller. 2008. “Bootstrap-Based Improvements for Interference with Clustered Errors.” The Review of Economics and Statistics 90 (3):414–427. Chetty, Raj. 2006. “A general formula for the optimal level of social insurance.” Journal of Public Economics 90 (10-11):1879–1901. ———. 2008. “Moral Hazard versus Liquidity and Optimal Unemployment Insurance.” Journal of Political Economy 116 (2):173–234. Crépon, Bruno, Esther Duflo, Marc Gurgand, Roland Rathelot, and Philippe Zamora. 2013. “Do Labor Market Policies Have Displacement Effects? Evidence from a Clustered Randomized Experiment.” Quarterly Journal of Economics 128 (2):531–580. Davidson, Carl and Stephen A. Woodbury. 1993. “The Displacement Effect of Reemployment Bonus Programs.” Journal of Labor Economics 11 (4):575–605. Decker, Paul T. 1997. “Work Incentives and Disincentives.” In Unemployment Insurance in the United States: Analysis of Policy Issues, edited by Christopher J. O’Leary and Stephen A. Wandner. Kalamazoo, MI: W.E. Upjohn Institute for Employment Research, 285–320. Gomme, Paul and Damba Lkhagvasuren. 2015. “Worker search effort as an amplification mechanism.” Journal of Monetary Economics 75 (106–122). Gruber, Jonathan. 1997. “The Consumption Smoothing Benefits of Unemployment Insurance.” American Economic Review 87 (1):192–205. Johnson, Terry R. and Daniel H. Klepinger. 1994. “Experimental Evidence on Unemployment Insurance Work-Search Policies.” The Journal of Human Resources 29 (3):695–717.

37

Keeley, Michael C. and Philip K. Robins. 1985. “Government Programs, Job Search Requirements, and the Duration of Unemployment.” Journal of Labor Economics 3 (3):337–362. Klepinger, Daniel H., Terry R. Johnson, and Jutta M. Joesch. 2002. “Effects of Unemployment Insurance Work-Search Requirements: The Maryland Experiment.” Industrial and Labor Relations Review 56 (1):3–22. Lachowska, Marta, Merve Meral, and Stephen A. Woodbury. 2016. “Effects of the unemployment insurance work test on long-term employment outcomes.” Labour Economics 41:246–265. Landais, Camille, Pascal Michaillat, and Emmanuel Saez. 2016. “A Macroeconomic Approach to Optimal Unemployment Insurance: Theory.” American Economic Journal: Economic Policy Forthcoming. Lise, Jeremy, Shannon Seitz, and Jeffrey Smith. 2004. “Equilibrium Policy Experiments and the Evaluation of Social Programs.” NBER Working Paper 10283. McVicar, Duncan. 2008. “Job search monitoring intensity, unemployment exit and job entry: Quasi-experimental evidence from the UK.” Labour Economics 15 (6):1451– 1468. ———. 2010. “Does Job Search Monitoring Intensity Affect Unemployment? Evidence from Northern Ireland.” Economica 77 (306):296–313. Michaillat, Pascal. 2012. “Do Matching Frictions Explain Unemployment? Not in Bad Times.” American Economic Review 102 (4):1721–1750. O’Leary, Christopher J. 2004. “UI Work Search Rules and Their Effects on Employment.” Tech. rep., W.E. Upjohn Institute. URL http://research.upjohn.org/ reports/83. Report prepared for Center for Employment Security Education and Research, National Association of State Workforce Agencies. Petrongolo, Barbara and Christopher A. Pissarides. 2005. “Looking into the Black Box: A Survey of the Matching Function.” Journal of Economic Literature 39 (2):390–431. Pissarides, Christopher A. 2000. Equilibrium Unemployment Theory. Cambridge, MA: The MIT Press, second ed. Shimer, Robert. 2004. “Search Intensity.” University of Chicago.

38

39

80,585 2.110 26.0 0.000

N: Sample Mean: F-stat, all=0: p-value:

80,585 2.110 36.1 0.000

0.647 (0.072) 1.106 (0.256) 1.410 (0.115) 2.712 (0.338) 0.778 (0.161) 80,576 1.957 23.4 0.000

0.773 (0.280)

0.576 (0.088)

(3)

80,576 1.957 30.7 0.000

0.488 (0.074) 1.039 (0.262) 1.284 (0.119) 2.697 (0.351) 0.639 (0.154)

(4)

69,052 1.003 5.3 0.008

0.555 (0.362)

0.251 (0.103)

(5)

69,052 1.003 21.9 0.000

1.726 (0.237) 1.607 (0.344) 1.897 (0.237) 2.066 (0.355) 1.596 (0.307)

(6)

Acceptable Contacts (lower bound)

BAM

111,548 2.470 2.3 0.112

0.066 (0.043)

0.020 (0.009)

(7)

111,548 2.470 4.0 0.004

0.066 (0.020) -0.006 (0.061) 0.096 (0.026) 0.021 (0.075) 0.035 (0.040)

(8)

Number of Search Methods

CPS

111,548 0.566 8.8 0.001

0.092 (0.022)

0.025 (0.006)

(9)

111,548 0.566 179.5 0.000

0.115 (0.004) 0.083 (0.016) 0.108 (0.009) 0.143 (0.018) 0.120 (0.013)

(10)

Binary: Contacted Employers?

CPS

Notes: Huber/White/sandwich standard errors clustered at the state level appear below estimates in parentheses. Regressions are of the form described in equation (8). Columns (1) and (2) display the estimated effects of search policies on the number of contacts reported by UI claimants audited by the BAM program in the 44 sample states described in Section 4. The dependent variable in columns (3) and (4) is the number of contacts that are not explicitly found to be unacceptable. The dependent variable in columns (5) and (6) is the number of contacts that are explicitly found to be acceptable. Regressions include state fixed effects, year-by-month fixed effects, and controls for industry, occupation, education, UI claim duration, race, ethnicity, sex, and state-by-quarter measures of average UI benefits and average potential UI duration. Columns (7) and (8) regress the number of search methods reportedly used in the prior 4 weeks by unemployed job-losers in the monthly CPS, again for the 44 states in question. Columns (9) and (10) report estimates from linear probability models for contacting employers directly using the same CPS sample. Both sets of CPS models include state fixed effects, year-by-month fixed effects, and controls for occupation, industry, race, education, unemployment duration, unemployment reason, sex, and state-by-quarter measures of average UI benefits and average potential UI duration.

0.828 (0.254)

0.571 (0.082)

(2)

Acceptable Contacts (upper bound)

Number of Reported Contacts

(1)

BAM

BAM

Nonspecific

4+ contacts

3 contacts

2 contacts

1 contact

# contacts

Outcome:

Data:

Table I Evidence on Efficacy of Search Requirements

Table II Share of CPS Respondents Reporting Each Search Method

Contacted employer directly/interview Contacted public employment agency Contacted private employment agency Contacted friends or relatives Contacted school/university employment center Sent out resumes/ filled out applications Checked union/professional registers Placed or answered ads Other active Looked at ads Attended job training programs/courses Other passive

Estimated Linear Effect 0.025 (0.006) 0.012 (0.004) 0.009 (0.002) −0.011 (0.006) −0.003 (0.002) −0.009 (0.006) 0.002 (0.002) −0.004 (0.004) 0.002 (0.003) −0.004 (0.004) −0.000 (0.001) 0.001 (0.001)

Mean Dependent Variable 0.566 0.238 0.103 0.297 0.031 0.532 0.052 0.194 0.084 0.351 0.016 0.006

Notes: Huber/White/sandwich standard errors clustered at the state level appear below estimates in parentheses. Regressions are of the form described in equation (8). The table reports the estimated coefficient on the linear measure of search requirements. Each estimate is from a separate regression where the outcome is an indicator for CPS respondents having used that particular search method over the prior four weeks. The sample is the same 111, 548 observations appearing in columns (7) through (10) of Table I. All models include state fixed effects, year-by-month fixed effects, and controls for occupation, industry, race, education, unemployment duration, unemployment reason, sex, unemployment reason, and state-by-quarter measures of average UI benefits and average potential UI duration.

40

Table III Relative Policy Effects by Pre-period Labor Market Strength

Lm defined 2005 to t

Pretreat. avg:

∗

High u: Low u:

Lm defined t∗ − 1 to t∗

High u: Low u: Number of CBSAs Policy change Pretreat. avg:

All -0.083 (0.127) [0.570] 6.796 5.598

-0.157 (0.121) [0.210] 8.484 6.780 85

Outcome=Unemployment Rate LA ND PA UT -0.023 -0.342 -0.381 -0.474 (0.314) (0.104) (0.175) (0.173) [0.932] [0.062] [0.054] [0.032] 7.551 3.328 7.146 5.309 5.914 2.253 5.807 4.541

WI 0.084 (0.155) [0.626] 7.130 5.937

-0.464 (0.272) [0.128] 9.722 7.381 17 1 to 3

-0.060 (0.156) [0.722] 7.856 6.405 22 2 to 4

-0.342 (0.104) [0.068] 3.426 1.642 5 2 to 4

-0.367 (0.176) [0.042] 8.926 7.160 32 0 to 3

-0.277 (0.216) [0.288] 9.257 7.656 9 2 to 4

Notes: Table reports regression estimates from a specification as described by equation (9). Huber/White/sandwich standard errors clustered at the CBSA level are presented in parentheses. Wild cluster bootstrap p-values are displayed in brackets. Estimates indicate the differential change in unemployment rates for initially-low-unemployment statistical areas when a search requirement increase is implemented at time t∗ . Low unemployment statistical areas are defined as those that have average unemployment below the state median in the indicated period (2005 to t∗ or t∗ − 1 to t∗ ). Monthly data on statistical area unemployment rates come from the LAUS program.

41

Table IV Policy and Shift-Share Interaction Effects Outcome= Unemployment Rate ˆ mt G ˆ mt × # contacts G ˆ mt × Nonspecific G # contacts Nonspecific CBSA FE Time FE State linear time trends CBSA linear time trends State quadratic time trends

(1)

(2)

(3)

(4)

-0.090 (0.176) -0.025 (0.028) -0.126 (0.067) 0.274 (0.144) 1.469 (0.506) X X

-0.431 (0.247) -0.072 (0.032) -0.282 (0.093) 0.234 (0.112) 1.699 (0.318) X X X

-0.431 (0.250) -0.072 (0.032) -0.282 (0.094) 0.234 (0.113) 1.699 (0.322) X X

-0.239 (0.213) -0.038 (0.030) -0.153 (0.074) 0.115 (0.135) 0.777 (0.616) X X

X X

Notes: Huber/White/sandwich standard errors clustered at the state level appear below estimates in ˆ mt is parentheses. Reported coefficients are estimated analogs of α1 and α4 from equation (12). G calculated at quarterly frequency using QWI data as described in Section 7.2. In addition to the reported coefficients, main effects of the policies are included along with statistical area fixed effects and month fixed effects. The outcome measure is the unemployment rate from the LAUS program. Estimation is performed on 25, 272 area-month observations from 648 individual statistical areas in 37 states.

42

Table V Cox Proportional Hazard Models of Policy Effects on Reemployment Failure= Reemployment # contacts Nonspecific Unemployment rate

(1)

(2)

0.022 (0.023) 0.117 (0.105) -0.068 (0.016)

0.030 (0.060) 0.279 (0.179) -0.047 (0.035) -0.001 (0.009) -0.022 (0.026)

94,310 1.2 0.541

94,310 4.4 0.353

Unemployment rate × # contacts Unemployment rate × Nonspecific Observations: χ2 -stat, all=0: p-value:

Notes: Table reports estimated proportional hazard model coefficients for measures of search requirement policies using longitudinally-linked CPS data. Included controls allow the hazard to vary proportionally in month, year, state, education, occupation, industry, race, ethnicity, sex, unemployment reason, state-by-quarter average benefit levels, and state-by-quarter average benefit duration.

43

Figure I Number of States with Each Required Number of Employer Contacts 40

5 Number of States

30

4 20

3

10

2 1

0 2005

2007

2009

2011

2013

2015

Notes: Figure indicates the number of states with each specific search requirement from one employer contact per week to five employer contacts per week. Policies are identified using state workforce agency publications as indicated by the author’s research.

44

Figure II BAM Distribution of Reported Contacts by Policy 1.0

0

5

Share Reporting

0.8

2

Contacts Reported

4

0 1 2 3 4 5+

1

0.6

3

0.4 1

5

0.2 0

0.0

1

3 4

23 45

0

3

0

2

1

0 5

3 1

1 4

2

5

2

2 4

0

2 5

3

0123

01

4

3

5

4

4

5

Nonspec

Number of Required Contacts

Notes: Figure indicates share of BAM-audited UI claimants reporting each number of employer conducts by search requirement policy. Data is from 2005 to 2014 and includes 87,712 observations of individuals who are not specifically exempted from active search.

45

Figure III Equilibria for Fixed Levels of Search Effort High Productivity

Low Productivity

Notes: Figure displays simulated Beveridge curves and job-creation curves from the model described in Section 3 and the parameters defined in the appendix. The left panel sets the productivity parameter a equal to 1.00 and the right panel sets it equal to 0.96. In both panels, the “Low s” curves have s fixed at 1 and the “High s” curves have it fixed at 2.

46

Figure IV Beveridge Curve Variation Variation in Unemployment Flow Utility

Variation in Search Effort

Notes: Figure displays simulated Beveridge curves from the model described in Section 3 and the parameters defined in the appendix. In the left panel, the “Low s” curve has s fixed at 1 and the “High s” curves has it fixed at 2. The “Variable s” curve allows s to be optimally chosen according to the condition in equation (6). The right panel similarly allows s to be optimally chosen according to the condition in equation (6). The “High b” curve sets b equal to 0.333, a replacement rate of 50 percent, and the “Low b” curve sets b at half of that.

47

48

15

15

05

05

05

05

07

07

07

07

11

11

11

09

11

South Carolina

09

Nebraska

09

Louisiana

09

Arizona

13

13

13

13

15

15

15

15

0

1

2

3

4

5

6

0

1

2

3

4

5

6

0

1

2

3

4

5

6

0

1

2

3

4

5

6

05

05

05

05

07

07

07

07

11

11

11

09

11

Tennessee

09

New York

09

Massachusetts

09

Connecticut

13

13

13

13

15

15

15

15

0

1

2

3

4

5

6

0

1

2

3

4

5

6

0

1

2

3

4

5

6

0

1

2

3

4

5

6

05

05

05

05

07

07

07

07

11

11

09

09

Utah

11

11

Oregon

09

Michigan

09

Florida

13

13

13

13

15

15

15

15

0

1

2

3

4

5

6

0

1

2

3

4

5

6

0

1

2

3

4

5

6

0

1

2

3

4

5

6

05

05

05

05

07

07

07

07

11

11

11

09

11

Wisconsin

09

Pennsylvania

09

Mississippi

09

Georgia

13

13

13

13

15

15

15

15

Notes: Figure displays search requirement policies and average contacts reported in the BAM data for 20 states that changed their policies between 2005 and the end of 2014. Thick dashed lines indicate search requirements as found in workforce agency publications. Any time without a dashed line indicates a nonspecific search policy. Thin solid lines show the monthly average number of employer contacts reported by audited claimants in the BAM data who are not specifically exempted from active search requirements.

0

0 13

1

1

11

2

2

09

3

3

07

4

05

5

6

4

Rhode Island

5

6

0

15

0 13

1

1

11

2

2

09

3

3

07

4

05

5

6

4

North Dakota

5

6

0

15

0 13

1

1

11

2

2

09

3

3

07

4

05

5

6

4

Hawaii

5

6

0

0 13

1

1

11

2

2

09

3

3

07

4

05

5

6

4

Alaska

5

6

Figure V Search Policy and Average Contacts Reported by State, states with policy changes

Figure VI Proportional Hazard Model Predicted Survival Curves 1.0

Share not reemployed

0.8 High unemployment 0.6

Low unemployment

0.4 Required Contacts 0 2 4

0.2

0.0 0

5

10

Weeks

15

20

25

Notes: Figure displays predicted survival curves from the proportional hazard model displayed in the rightmost column of Table V. The “Low unemployment” curves have unemployment set to 4 percent and the “High unemployment” curves have it set to 12 percent. All variables other than unemployment and the contact requirement are set to their estimation sample averages.

49

Table C.I Simulation Parameters Name Discount factor Separation rate Matching efficiency su-elasticity of matching Production function curvature Vacancy cost Wage Search disutility scale Search disutility shape Unemployment benefit Productivity (high)

Symbol

Value

Source

β λ µ φ α c w ν κ b a

0.999 0.0095 0.233 0.5 0.666 0.215 0.666 2.64 0.026 0.333 1.00

Weekly value, 0.95 annual Michaillat (2012) Michaillat (2012) Petrongolo and Pissarides (2005) Approximate labor share Michaillat (2012) Approximate labor share Author’s choice Author’s choice 50 percent replacement rate Normalization

Notes: See Appendix C and Section 3 for additional details.

50