Journal of Labor Research

, Volume 34, Issue 4, pp 433–454

Discouraging Workers: Estimating the Impacts of Macroeconomic Shocks on the Search Intensity of the Unemployed


    • Department of EconomicsElon University
  • Mark Kurt
    • Department of EconomicsElon University

DOI: 10.1007/s12122-013-9166-0

Cite this article as:
DeLoach, S.B. & Kurt, M. J Labor Res (2013) 34: 433. doi:10.1007/s12122-013-9166-0


Discouraged and marginally attached workers have received increasing attention from policy makers over the past several years. Through slackness in the labor market, periods of high unemployment should reduce the likelihood of receiving a job offer and thus create more discouraged workers. However, the existing literature generally fails to find evidence of such pro-cyclicality in search intensity. Surprisingly, search appears to be acyclical. We hypothesize the observed acyclicality may be the result of coarse measurement of search intensity in previous studies and the failure to account for changes in individuals’ wealth across the business cycle. In this paper we use daily time use dairies from the American Time Use Survey 2003–2011 to examine the cyclicality of search intensity to explain this apparent contradiction between theory and data. Results indicate that workers do reduce their search in response to deteriorating labor market conditions, but these effects appear to be offset by the positive effects on search that are correlated with declines in household wealth.


Search intensityHousing pricesDiscouraged workersBusiness cycles
JEL codes J2J6J1E24E32

In February 2011, over 1 million potential workers in the United States were identified as “discouraged,” a more than 250% increase from 3 years earlier (Bureau of Labor Statistics 2012).1 Given the rapidly deteriorating labor market conditions brought on by the start of the Great Recession, this fact is not terribly surprising. Though these potential workers would accept a full-time job if offered, they have given up searching because the cost of their job search outweighs the expected payoff. As Krause and Lubik (2011) explain, during deep and prolonged recessions, the expected payoff from finding a job match declines, leading to substantial increases in the number of discouraged workers. Moreover, this kind of pro-cyclicality of search intensity associated with the discouraged worker effect is likely to have propagating effects on the business cycle. As search intensity decreases, the probability of a successful job match also decreases, resulting in longer duration of low levels of employment. Given the potential costs of this drop in search intensity both to individuals as well as the economy as a whole, it is surprising labor economists know so little about the determinants of search intensity of the unemployed.2

In general, there is a dearth of empirical evidence regarding the extent to which search intensity varies over the business cycle. The limited evidence that exists suggests search intensity may be acyclical. Shimer (2004) found that, after controlling for former industry, occupation, and duration, search intensity among active searchers did not appear to change significantly during the 2001 recession.3 The finding that search intensity appears to be acyclical is somewhat puzzling in light of the evidence that many other features of the labor market (e.g., total hours worked, unemployment rate, vacancy rate, job destruction rate, job opening rate) are strongly correlated with the business cycle.

One possible reason for the inability to reconcile the apparent acyclicality of search intensity with the existence of discouraged workers may be due to how search intensity is measured. For example, in the Current Population Survey (CPS), workers are asked a series of questions about whether and how they have recently searched (e.g., looking through a newspaper, interviewing for a job, etc.). Using only information from the CPS, Shimer (2004) proxies intensity by counting the number of different search methods reported. The assumption is that the more varied the methods, the more time one spends on search. This assumption, however, is difficult to justify. Fortunately, the American Time Use Survey (Bureau of Labor Statistics 2012a) is designed to address precisely these kinds of problems. Because workers record the actual time they spend on activities related to searching for a job, we have a direct measure for intensity.

To date, Krueger and Mueller (2010) is the only known study to use the American Time Use Survey (ATUS) to look at the search intensity of unemployed workers. In this paper, Krueger and Mueller estimate the impact of unemployment insurance (UI) benefits on search intensity. As predicted by Mortensen (1977), their findings suggest that as UI benefits increase search intensity declines. However, they do not address behavior of search intensity over the business cycle.

A second possible reason for the inability to reconcile the observed acyclicality of search intensity may be due to the existence of non-idiosyncratic shocks to wealth. To our knowledge ours is the first study of search intensity to account for such wealth effects. The availability of unemployment benefits are commonly thought to decrease search intensity. Benefits typically rise during recessions through extensions. This effect would make search intensity pro-cyclical, similar to the discouraged worker effect. However, in times of recessions household wealth typically falls as asset values, such as housing, plummet. This effect would tend to make search intensity counter-cyclical, since a decline in wealth during recessions would tend to increase search intensity. Theoretically, these two wealth effects counteract each other. Ultimately, the extent to which this occurs is an empirical question.

Understanding the behavior of search intensity over the business cycle is important because of its implications for modern search theory. The literature in this area has been interested in understanding the volatility of labor market tightness, as measured by the vacancy to unemployment (VU) ratio. Recent work has focused on the inability of existing models (e.g., Mortensen and Pissaridies 1994) to account for the variation in the VU ratio over the business cycle while matching many other empirical features (Hall 2005; Shimer 2005; Hagedorn and Manovskii 2008; Pissarides 2009). Specifically, theory predicts that the volatility in the VU ratio should match the volatility in average labor productivity (Shimer 2005). However, these models reply on assumptions about the behavior of search intensity over the business cycle. Specifically, Shimer (2005) assumes that individual search intensity is non-decreasing in unemployment.4 Ultimately, understanding the behavior of search intensity over the business cycle could help explain the observed excess volatility in the VU ratio.

The purpose of this paper is to help fill this void in the literature. Using the data from the ATUS 2003–2011, we estimate the effects of idio- and non-idiosyncratic (macroeconomic) shocks on the search intensity of unemployed workers. We estimate variety of models, all couched in terms of a random search model where intensity is determined by three standard components of the canonical search model: the cost of search, the likelihood an offer is obtained from a given amount of search, and the expected net payoff conditional on an offer (Mortensen 1986). Due to the natural variation in the data over this sample period, we are able to identify a number of correlated, but distinct shocks that have competing effects on the search intensity of individuals. These include monthly changes in regional VU ratios and regional housing prices.

Overall, the paper makes a number of contributions to the empirical literature on the determination of search intensity. Most importantly, the results offer an explanation for the observed acyclicality of search intensity. As standard theory predicts, workers respond negatively to a deteriorating labor market (i.e. the discouraged worker effect), with a one percent increase in the VU ratio resulting in a 0.46% increase in search intensity. However, macroeconomic shocks to household wealth appear to mitigate the discouraged worker effect. We find that a one percent increase in housing wealth results in a 1.64% decrease in search intensity. Ultimately, the combined effects of the large fluctuations in both the VU ratio and housing prices that occurred during our sample period appear to explain the observed acyclical behavior of search intensity across the business cycle. Because of this, the results have important implications for the modern search theorists in their ongoing effort to accurately account for the excess variation in the VU ratio.

Search Theory

A generalized model of search intensity, S, for the ith individual, in the jth labor market, experiencing an unemployment spell at time t can be written as:
$$ {S}_{ijt}=f\left(c,\lambda {\theta}_{ijt},{w}_{ijt}^e,\kern.2em {l}_{it}\left|{S}_i>0\right.\right) $$
$$ {l}_{ijt}=g\left({\phi}_{it},{W}_{ijt}\right) $$

This represents the agents’ decision. First, c represents the disutility, or direct cost of search. Together, λθ is the likelihood of obtaining an offer given a unit of search time, where \( \lambda \) is the arrival rate of offers conditional on search and \( \theta \) is a measure of labor market tightness.5 It is important to note that Eq. 1 is conditioned on an unemployed individual having some positive amount of search during the current unemployment spell, not necessarily observed at time t. Next, we is the expected wage. Finally, l represents the value of leisure. In Eq. 2, we further specify that the value of leisure depends on the opportunity costs, ϕ and W, the wealth of the individual.

Direct Costs (c)

Theoretically, the disutility from search encompasses an individual’s direct costs of search and their indirect costs. We interpret the former as the disutility of effort associated with a unit of time spent searching for employment. For direct search costs to play a significant role in determining search intensity, they must vary either by individual workers or across the business cycle. At the individual level, it is reasonable to imagine that these costs vary across workers. However, such differences will largely depend on factors such as worker education and experience. Since these factors also affect the expected wage, it is not possible to model these idiosyncratic differences as pure direct search costs.

Non-idiosyncratic shocks to direct search costs will ultimately be driven by changes in search technology. While job search technology has undoubtedly changed in the last decade, these are arguably long-term changes unlikely to vary across the business cycle. Because there is little reason to believe that direct search costs vary across the business cycle, they are also unlikely to be correlated with variation in search intensity due to macroeconomic shocks, which is the focus of our study. As a result, the lack of suitable proxies for direct search costs in the econometric model should cause no problems with our fundamental inferences.

Probability of Obtaining a Job Offer (λθ) and the Expected Wage (we)

Labor market tightness has long been a topic of study in labor economics, dating back to the Beveridge curve (Dow and Dicks-Mireaux 1958). Its variation, as measured by the VU ratio, over the business cycle has been well documented and driven much research (Blanchard and Diamond 1989; Eckstein and van den Berg 2007). We focus our attention on how labor market tightness affects search intensity. As the probability of receiving an offer conditional on search decreases, search intensity should fall. Labor market tightness is the key to determining the arrival rate of job offers conditional on search.6,7 Because macroeconomic shocks to the labor market perturb the equilibrium employment level, tightness in the labor market varies widely across the business cycle (Blanchard and Diamond 1989).

The variation in the labor market tightness is driven through two factors: job openings (or vacancies) and unemployment. During recessions there are increasing numbers of job matches dissolve and unfilled vacancies close without replacement. Considering the inverse relationship between vacancies and unemployment, the VU ratio is quite violate over the business cycle.

The third component of the search model in Eq. 1 is the expected wage offer conditional on search. The logic here is straightforward. The higher the predicted wage, the more incentive there is for each unemployed worker to increase her/his search intensity. Like labor market tightness, the expected wage also varies across the business cycle. Indeed, a number of recent studies find new wage offers exhibit pro-cyclical behavior (Pissarides 2009; Kudlyak 2010).

Opportunity Costs (ϕ)

The value of leisure plays an instrumental role in determining search. Not only does it vary substantially across workers, but there also is good reason to believe that it varies over the business cycle as well. The value of leisure is a function of two factors: opportunity costs and wealth.

First we address those idiosyncratic factors that affect an unemployed individual’s opportunity costs. One way of thinking about this is in terms of each worker’s value of household production. For example, the presence of children in the household will increase the value of household production and increase the disutility of searching. Thus, the presence of children is expected to decrease search intensity. Similarly, because they may more easily (i.e., with less opportunity costs) substitute their time from paid work to housework, unemployed workers who are married may search less intensively than their unmarried counterparts. In addition, the reason for one’s unemployment (e.g., on temporary layoff, voluntary job leaver, re- or new entrants into the labor market, or job loser) will serve as a proxy for unobserved personal preferences. For example, if someone has voluntarily left a job it could be the result of family preferences, health issues, etc. In such a case, this would reveal a preference for non-work activity over work. Finally, age could proxy disutility from search as younger workers may value non-work activity relatively more than middle aged workers. For example, younger workers are more likely to view additional education as an option during times of high unemployment than older workers. At the other end of the spectrum, older workers may find searching for a job just a few years before planned retirement to be relatively costly as the value of their leisure time is likely to increase as they near retirement.

Wealth Effects (W)

In addition to the opportunity costs varying across workers, the value of leisure depends upon the wealth of workers. Wealth directly affects the marginal utility of leisure. Because wealth increases the value leisure, wealthier unemployed workers should search less intensively, all else equal.

Unlike the opportunity costs discussed above, the value of leisure that results from changes in wealth are likely to vary substantially over the business cycle. The Great Depression has also been referred to as the “Great Compression” due to the massive loss of wealth (Goldin and Margo 1992). The decline in wealth resulted from sharp decreases values across many classes of assets, including housing. From 2001–2007, there was wide-spread appreciation in housing prices. Large decreases in the price of housing starting in late 2007 caused record defaults and record low housing starts. This variation of wealth across individuals and time generates substantial variation in the value of leisure across workers.

In addition to housing wealth, the availability of unemployment insurance (UI) benefits represents another important source of wealth for the unemployed. Mortensen (1977) develops a model that endogenizes the decision to search as well as the amount of time spent searching. In his model, unemployment benefits directly affect search intensity. In general, he finds the effect of an increase in UI benefits on search intensity to be ambiguous depending on state unemployment benefit structure. Krueger and Muller (2010) empirically test this result and find that unemployed workers who are eligible for UI benefits actually respond to an increase in UI benefits by searching less.


To estimate the model given in Eq. 1, data are pooled from several sources. All data on individual workers come from the American Time Use Survey (ATUS) 2003–11 (Bureau of Labor Statistics 2012a), a multi-year dataset is a pooled cross-section of its annual surveys. Data on current labor market conditions come from the Bureau of Labor Statistics (2012c). Data measuring changes in household wealth come from Case and Shiller’s index of housing market prices (Standard and Poor’s 2012). Descriptive statistics for all the variables used in the regressions are summarized in the Appendix in Table 1.

The ATUS is a sub-sample of the Current Population Survey (CPS) (Bureau of Labor Statistics 2012b). Individuals selected for the CPS are interviewed monthly for 8 months in total. Following that, a sub-sample is selected for the ATUS. The ATUS interviews are conducted 2–5 months following the final CPS monthly survey. Each respondent is randomly assigned 1 day (diary day) on which to record their activities. Respondents record activities starting at 4 a.m. of the interview day and end at 4 a.m. on the following day. In addition to the nature of the activity, respondents are asked where the activity took place, who else was present during the activity, and duration of the activity.

Following Krueger and Mueller (2010), unemployed individuals are defined using the ATUS labor force status recode (see Table 2 in the Appendix). Like them, we also restrict the sample to unemployed workers between the ages of 20 and 65 who are not enrolled full-time in school. While constructed like Krueger and Mueller (2010), our sample includes an additional 4 years of data. Most importantly, these additional years of data include the Great Recession. This sample provides a great deal of variation in employment status and macroeconomic shocks that are likely to affect the search intensity of workers.

Search Intensity

The data used to construct search intensity come from the ATUS. Because the ATUS provides detailed data on time use, we are able to measure the time each individual spends searching during their diary day. In the 2003–11 multi-year dataset from the ATUS, activities related to job search include job search activities, job interviewing, waiting associated with job interviews, security procedures associated with search or interviews, and other job search activities not otherwise specified.8

Estimated search intensity for unemployed workers over the sample period is given in Fig. 1. Here we report the average search time for those participating in search activities during their diary day (i.e., mean search | search > 0).9 The most interesting fact in this graph is the dramatic rise in search intensity at the beginning of the recession in 2008, followed by an equally dramatic decline. In addition, there is no obvious relationship between search intensity and the unemployment rate. This is consistent with Shimer’s (2004) argument that search intensity appears acyclical.
Fig. 1

Search Intensity (average minutes per day, weighted) (All descriptive statistics are weighted using the probability weights provided by the ATUS 2003–11 (tufnwgtp).). Source: American Time Use Survey 200311 (Bureau of Labor Statistics 2012a) and CPS (Bureau of Labor Statistics 2012b)

Labor Market Tightness (θ)

The standard measure for tightness in the labor market condition is the VU ratio.10 Job vacancies are not available at the state level, but they are available at the regional level through the Job Openings and Labor Turnover Survey (Bureau of Labor Statistics 2012c). Dividing this by the regional unemployment rate (Bureau of Labor Statistics 2012b) results in the regional VU ratio, our measure of regional labor market tightness. This is shown in Fig. 2 over the sample period. It is easy to see that the VU ratio varies significantly across both regions and time.
Fig. 2

Vacancy to Unemployment Ratio across Census Regions. Source: Job Openings and Labor Turnover Survey (Bureau of Labor Statistics 2012c) and CPS (Bureau of Labor Statistics 2012b)

As will be discussed later, identifying the effect of changes in the VU ratio on search intensity will be confounded if there are other macroeconomic shocks that affect search intensity in ways other than through changing the probability of finding a job. The obvious example of this are changes in wealth. Wealth falls during the business cycle. At the same time, the VU ratio falls. Failing to account for changes in wealth over the business cycle, then, introduces simultaneity since it is correlated both with the VU ratio as well as search intensity.

Expected Wage (we)

For each unemployed worker we estimate a predicted real wage to proxy their expected offer conditional on search. This is a predicted wage for each worker conditional on their human capital and geographic location. To estimate the predicted wage, we run regressions for all men and women who are employed full time using data from the CPS outgoing interview file for the period 2003–2011. This resulted in 187,714 men and women between the ages of 20 and 65 who reported earnings and were not full-time students. Regressors included sex, age, age squared, dummies for the level of education, and dummies for the state of residence. In addition, we also run separate regressions for each sample year to generate the predicted wage for each worker. This allows for potential changes in the marginal returns to experience and education that may vary over time. It also controls for time-varying cost-of-living changes that affect market wages.

Idiosyncratic Characteristics (ϕ)

Unemployed workers are classified into four groups: job losers, workers on temporary layoff (expecting to be recalled by employer), job leavers, and re-entrants and new entrants. These classifications are constructed from information in the CPS and ATUS. Table 2 in the Appendix lists the precise CPS and ATUS variable codes used to construct these categories. Since the ATUS interview occurs 2–5 months following the CPS interview, it is not possible to rely solely on the classifications in the CPS. The CPS variable which contains this information asks all unemployed workers the reason for their unemployment. The four categories are job losers, job leavers, those on temporary layoff, and re-and new entrants. Job leavers are those who have voluntarily left their jobs. The remaining categories, job losers, those on temporary layoff, and re- and new entrants require additional information from the ATUS interview. Re- and new entrants include those who have not recently been in the labor force but are currently seeking employment in the CPS. This group also includes workers who were not in the labor force in their CPS interview but consider themselves unemployed in the ATUS interview. To determine if an unemployed worker is on temporary layoff, we use information from the ATUS on whether they have been given any indication that they will be recalled to work within the next 6 months and whether they have been given a date for their return. Job losers include those who do not expect to be recalled or whose temporary job ended. We also have a number of workers who have lost their jobs between the time they did the last CPS interview and the time they do the ATUS interview. These workers are those who were working in the CPS but who are unemployed in the ATUS.

As outlined previously, individuals vary in their opportunity costs of search time. Because we have no way to proxy personality differences, we are forced to rely on basic demographic characteristics likely to be related to these preferences and proxy the individual’s returns to household production. To account for these characteristics, we use basic information from the ATUS regarding factors such as sex, marital status, age, and the presence of children in the household. Because we expect large differences across sex, we use a number of interactions between sex and marital status and the presence of children. For age, we include age and age2 because we expect the youngest and the oldest unemployed workers to search less intensively than middle-aged workers.

Wealth Effects (W)

In addition to idiosyncratic differences across workers, the opportunity cost of search is also affected by macroeconomic conditions and policies that generate wealth effects. Because of data constraints, the only macroeconomic shocks to wealth we are able to identify are regional housing market fluctuations.11 We also consider government policies that generate wealth effects, namely, unemployment insurance benefits that vary in level and availability across states. Changes in regional housing prices over the sample period are shown in Fig. 3.

Econometrically, it is important to be able to identify wealth effects that can be attributed to changes in housing values across individuals. This is accomplished in two ways. First, we are able to construct regional housing values using the Case-Shiller housing market data. As we see in Fig. 3, there are substantial differences both in level and variation across the four census regions. For example, the West region had the largest changing in prices over the series. It went from some of the least expensive housing to the most expensive region before falling to third by December 2009. In addition, housing prices started falling in most regions in late 2006, while labor market conditions did not deteriorate until late 2007. Second, we know from the ATUS whether or not workers own a home. Given that we have variation across time and region that do not coincide directly with labor market conditions, coupled with the fact that we know whether our workers are homeowners, there is reason to believe that changes in wealth due to housing values can be identified apart from changes in labor market conditions.
Fig. 3

Case-Shiller regional housing price indices 2003 to 2011. Source: Standard and Poor’s (2012)

Wealth effects are also generated from the availability of UI benefits. These vary by state of residence and by individuals, since UI benefits depend on one’s previous earnings. Unfortunately, because we do not know from the ATUS data what the individual’s earnings were previous to becoming unemployed, there is no way to accurately estimate individual UI benefits. As a result, we follow Krueger and Mueller (2010) and use the maximum weekly benefits allowed by each worker’s state of residence to proxy unemployment benefits for those eligible (Department of Labor 2012).

To proxy the wealth effects related to the availability of UI benefits, it is necessary to be able to determine which workers are eligible. Consistent with the literature, we define UI eligible individuals as those who have lost their job as well as those on layoffs. For the period 2003–07, workers unemployed for more than 26 weeks are considered not eligible. After 2008 the federal government passed two extension laws, the Emergency Unemployment Compensation act of 2008 (UEC08) and the Extended Benefits (EB) act of 2009. The 2008 legislation, passed in multiple supplemental bills, provided for four tiers of extended benefits. Tier 1 provided for 20 additional weeks, Tier 2 additional 14 weeks, Tier 3 13 additional weeks and Tier 4 6 additional weeks, for a total of 53 additional weeks. The EB act of 2009 provided for up to 20 additional weeks on top of that. To complicate matters, extended benefits in each state (and each tier) have different effective dates due to the criteria that “trigger” the additional weeks of UI. In the later years of our sample, extended benefits began to “switch off” in some states, depending on their level of unemployment. Further complicating matters, some states’ unemployment rates and average duration fluctuated above the threshold to trigger additional weeks in 2011. Using the information provided by the Department of Labor (2012), we are able to determine how many weeks of unemployment benefits for which each unemployed worker is eligible. This information is used to measure with some degree of certainty whether an individual is eligible for benefits at any given time.

Econometric Model

Estimation of the model is complicated due to the fact that less than 20% of unemployed workers in our sample record time spent on job search activities. Because time spent searching is censored at zero, there is some question about how to best estimate the model. The classical censoring model would be one in which the latent variable, s*, may take on negative values, but that the observed, s, is censored at zero. In time use surveys, this is not strictly the case. Because the observations are bound by zero, many researchers have chosen to use Tobit to deal with this econometric issue. However, the implementation of Tobit assumes that the factors determining the likelihood that a respondent reports zero minutes of a given activity on a given day are the same as those determining total time spent, conditional on the respondent engaging in that activity on that day. Recently, economists have questioned the validity of that assumption in the context of time use data. Stewart (2009) shows that when this assumption is violated, the bias in Tobit estimates is large and that the size of this bias increases with the proportion of censored observations. Stewart (2009) goes on to show that OLS estimates are robust over various specifications of the data-generating process. As a result, he strongly recommends the use of OLS over Tobit in cases with data similar to ours.12

In addition to the issue of censoring, there is a question of simultaneity. In particular, there is concern that the VU ratio is simultaneously determined along with search intensity. There are several ways of dealing with this. First, dummies at the local level (state) can be included to account for institutional or structural characteristics that may be correlated with the regional VU ratio as well as search intensity. More likely, however, it is omitted macroeconomic shocks that vary across the business cycle that are the cause of the problem. The main way we account for this potential issue is to introduce housing prices into the model to account for changes in wealth.

As a final check, we also estimate an instrumental variables (2SLS) version of the model to account for any endogeneity in the VU ratio that may remain after the introduction of housing prices. Even though the VU ratio is measured at the regional level while search in at the individual level, it is possible for omitted variables to simultaneously affect both search and the VU ratio. For example, suppose workers expect an improvement in the labor market for some exogenous reason, such as changes in fiscal policy. This would result in an increase in search intensity and possibly a short-term increase in the unemployment rate as workers re-enter the labor market. The result would be a simultaneous increase in search and decrease in the VU ratio. We estimate two versions of the 2SLS using different instruments: (1) mass layoffs at the regional and national levels; and (2) the S&P 500.

In general, the model outlined in the previous section can be written as
$$ {S}_{ijt}=\alpha +\beta V{U}_{jt}+\delta {w}_{ijt}^e+{W}_{ijt}\varGamma +{X}_i\varPi +{\xi}_j+{e}_{ijt} $$

Where sijt is the search intensity of the ith worker in jth regional labor market at time t. VUjt is the regional vacancy-to-unemployment ratio, wijte is the expected wage, Wijt is a matrix including wealth effects and Xi includes all controls for idiosyncratic differences across workers. These include sex, marital status, age, age squared, and the presence of children in the household. As previously discussed, Wijt consists of three components: (1) whether the individual owns a house, Hi; (2) housing market values, Vjt; and (3) unemployment compensation, UIijt. State fixed effects, ξj, are also included, as are dummies for the day of the week, month, and whether diary day was a holiday.13

In addition to the two primary estimation techniques (OLS and 2SLS), we estimate a number of specifications. Estimates are weighted using the probability weights provided by the multi-year ATUS dataset and standard errors are clustered by year and month of interview. Though not reported in the results in Table 4, all specifications do include the controls for idiosyncratic differences across workers (Xi). Full results are, however, reported in Table 5 in the Appendix.
Table 1

Descriptive statistics




Weighted mean

Std. Dev.










Ln(VU ratio)







ln(real Case-Shiller index)







Predicted ln(real wage)







ln(max weekly UI benefits)














Job leaver







New entrant







Owns house











































Recall that the puzzle in the recent search literature is the apparent acyclicality of search intensity observed by Shimer (2004). This finding is inconsistent with standard search theory that predicts that a rise in the VU ratio would induce workers to increase search intensity. The results from our baseline model in Table 4 (model 1) also reveal this apparent acyclicality as changes in the VU ratio have a negative, but insignificant effect on search. However, once wealth effects are added to the model in the form of regional housing prices and home ownership (model 2), the coefficient on the VU ratio becomes positive and significant. This is consistent with results from the Tobit estimation given in Table 3 in the Appendix.14 Not only is this consistent with standard search theory, but it is consistent with the central argument put forth in this paper about the importance of controlling for changes in wealth.

Wealth is important due to its effects on the value of leisure for individuals.15 Consistent with theory, home ownership is negative and significant (model 2). All else equal, those who own a house search 8 min less per day, or about 25% less than renters.16 Moreover, increases in regional housing prices have a negative and significant effect on search intensity. Theoretically, a significant effect of housing prices on search intensity could be caused by either a change in wealth or labor market conditions resulting from broader macroeconomic instability. As a wealth effect, a drop in housing prices would decrease an individual’s level of wealth and increase search intensity, a negative relationship. On the contrary, if a drop in the housing prices is just a sign of worsening macroeconomic conditions, then search intensity should decrease due to the expectation of a decreased likelihood of finding a job, a positive relationship. The fact that housing values negatively affect search intensity appears to support the existence of a wealth effect. In other words, whereas changes in the VU ratio is proxying the effect of labor market conditions on search, changes in housing values is proxying the effect of wealth on search.17

As a wealth effect, we also expect homeowners to be most responsive to declining housing conditions. This is because homeowners have greater wealth, on average, and a higher portion of their wealth is tied to their home. The results from model 3 are at least partially consistent with this argument. Looking at the coefficient on the dummy interaction between homeownership and housing prices, we see that homeowners are more responsive to changes in housing values. However, the relative magnitude of this effect is economically small.

These results offer one plausible explanation for the acyclicality of search observed in Fig. 1. The large increases in wealth leading up to 2008 led to a decrease in search intensity, even given the relatively strong job market. Between 2003 and 2007 the VU ratio doubled in most regions and housing prices increased roughly 50% in the South and West. All else equal, the doubling of the VU ratio results in about a 45% increase in search intensity, an additional 100 min a week. At the same time, the 50% increase in housing prices results in an 82% decrease in search. The net effect in this period is an overall decrease in search intensity, consistent with Fig. 1.

After the start of the financial crisis in 2008, there were unprecedented drops in household wealth (housing, stock, and otherwise) leading to an increase in search intensity. At the same time, however, the dramatic decrease in the VU ratio acted to decrease in search intensity. For example, the estimates in Table 4 suggest that a 30% drop in housing values like those experienced in some regions between 2006 and 2008 would result in roughly a 50% increase in time spent searching for a job. That translates into nearly an additional 2 h of searching per week.18 After 2007, the VU ratio decreased by as much as 75% in some areas. This effect alone would decrease the search time about 35%. However, combining the effects of the drop in the VU ratio with the drop in housing prices still results in a net increase in search time between 2007 and 2009. Again, this is precisely what we observed in Fig. 1.

As mentioned previously, we also estimate the model with 2SLS as a check to see if there is any remaining unobserved heterogeneity that might simultaneously cause the VU ratio and search intensity. The primary concern is whether there could be another macroeconomic shock correlated with the business cycle that affects search in ways other than through the probability of finding a job (i.e., through the VU ratio). However, the results across the various OLS and 2SLS specification summarized in Table 4 suggests there is no significant simultaneity between the VU ratio and search intensity. Even though the instruments used (mass layoffs at the regional and national level in models 4–6, and the S&P 500 in model 7) strongly identify the VU ratio in the first stage, the Hausman test fails to reject the null of exogeneity. As is well-known, in the absence of endogeneity, OLS is more efficient than 2SLS. This provides additional evidenced of the validity of the OLS results discussed above.


While the estimated wealth effects appear large, there is reason to be cautious in interpreting this effect as a pure wealth effect. If the estimated effect was completely due to changes in housing wealth, then only homeowners would respond to changes in housing prices. However, this is not the case. In fact, the evidence suggests that both homeowners and non-homeowners increase their search intensity when housing prices are falling.

One possible explanation for why non-homeowners are also affected could be due to spillover effects caused by falling housing values on the financial sector. For example, it is well-documented that the housing market crash resulted in a credit crunch starting in late 2008. The effect of this credit squeeze obviously had an impact on both homeowners and renters. Likewise, easy credit, such as that experienced in the mid-2000s, can simultaneously stimulate housing prices and increase welfare for renters. A second reason for this surprisingly large effect could be that changes in housing prices are highly correlated with changes in other forms of wealth, such as stock prices. As mentioned previously, unfortunately, we have no information on other forms of wealth in this sample. Because changes in aggregate stock indices are so closely tied to the macroeconomy (i.e., VU ratio, housing prices, etc.), it is impossible for us to identify a separate “stock wealth effect” in this data sample.

Even though changes in housing prices in our data are likely to be picking up additional wealth effects beyond merely those tied to home equity, we can still reasonably infer that housing wealth does affect search intensity. The reason is that through the CPS data, we know which workers own homes and which do not. Arguably the most conservative estimate of a true “housing wealth effect” is provided by the estimated elasticity of search intensity with respect homeownership. The results from model 2 suggest that all else equal, an unemployed homeowner will search about 30% less time than a comparable unemployed renter. This translates into about 1 h less search time per week.


In this paper we investigate the search intensity of unemployed workers. Using data from recent time use surveys, we are the first to document the variation in search intensity in response to a number of recent, large-scale macroeconomic events including the recent housing market collapse and the deteriorating availability of jobs. In the process, the paper makes a number of contributions to the literature on job search intensity of the unemployed.

First, our research establishes how search intensity responds to non-idiosyncratic shocks using time use data. The evidence confirms that unemployed workers do decrease their search intensity in response to deteriorating labor market conditions just as standard search theory predicts. We also find suggestive evidence of wealth effects, proxied by homeownership and changes in housing prices. As wealth declines, unemployed workers respond by increasing the intensity of their job search. Due to some limitations in the data on financial holdings, it is unlikely that all of the estimated wealth effect can be attributed completely to changes in regional housing values. However, we find evidence of a wealth effect on searchers that is independent of other macroeconomic shocks that would directly affect one’s probability of finding a job. This wealth effect interpretation is supported by unemployed homeowners searching significantly less and appearing to be more responsive to changes in local housing market conditions than renters.

Second, these results offer insight into the acyclicality of search intensity cited previously by Shimer (2004). The wealth and substitution effects that come from our results are not part of standard aggregate model typically used in the literature, and may have important implications for policy makers. Unemployed workers searching for a job appear to respond as expected to decreases in the demand for labor. As layoffs increase, they decrease their intensity due to the diminished expected probability of obtaining a job offer. This effect in isolation is indicative of the pro-cyclicality of job search. However, wealth also varies across the business cycle and workers respond to those shocks too. When housing prices fall, for example, search intensity increases. This offers a plausible reason for why search intensity increased in 2008 at the beginning of the Great Recession. Thus, during these months observed search intensity appeared to be behave counter-cyclically. Ultimately, the extent to which the counter-cyclical wealth effect found here is due to declines in housing wealth or other consumer wealth remains a question for further research.

Third, this work points out the importance of finer measurement of economic variables. By constructing search intensity from time use data we are better able to identify the effects of changes in local labor market conditions on search intensity than previous studies. The inability to link data pertaining to complete unemployment histories, as well as individual and household wealth portfolios to the time use data obscures the true effect of wealth on search intensity. Nevertheless, our results are suggestive of deeper wealth effects, and point to the need for data with more complete measures of wealth to better understand how individuals’ search intensity varies across the business cycle.

Finally, this paper can be seen as offering evidence to inform economists’ thinking about future general equilibrium models of search. Most obviously, search intensity is neither exogenous nor time invariant as is assumed in standard labor search models. The fact that search intensity responds to changes in asset prices across the business cycle has implications for the modeling of search over the business cycles.


Calculated using Table A-1 and A-15 from Monthly Employment Situation Report.


There have been a number of papers that model the effects of search intensity over the business cycle on certain labor market outcomes such as the matching function (see Merz 1995; Andolfatto 1996; Gautier et al. 2007, and Shimer 2008).


In his paper, he uses data from the Current Population Survey (CPS) from 1994 to 2004. However, there are limitations to using the CPS in this regard. The CPS records worker responses regarding the types of search they used. Search intensity is defined somewhat narrowly as the number of search tasks undertaken in efforts to secure employment, such as searching for employment in the newspaper, or interviewing for a job. Moreover, the CPS does not measure the actual time spent on these search activities.


If individual search intensity is non-decreasing in unemployment, u, then aggregate search intensity is increasing in unemployment. The implication is that the matching function, m(u, v) is increasing in u, ceteris paribas. This ignores how changes in wealth, opportunity costs, or other factors many influence the search intensity decision.


An alternative would be to allow the arrival rate, λ, dependent on the duration of individual i’s current unemployment spell at time t. See Eckstein and van den Berg (2007) for a survey of models which incorporate duration. Because of limitations in the ATUS data, however, we eschew this approach.


Job search technology also affects the probability of obtaining an offer conditional on search. The internet has reduced the cost to post and apply for a particular vacancy. The total amount of help-wanted ad space has decreased during the last 10 years as the unemployed and firms substitute away from the relatively more expensive newspaper ads to electronic ones. The economic recovery from the 2001 recession has virtually no increase in the column square inches of ad space. This is likely due to the increase in internet penetration rates from 1992 recession to the 2001 recession. Internet penetration rates increased from 1.7% of Americans with Internet access to 59.8% (World Bank 2012). Both of these features support an increasing efficiency of job search over time but not cyclicality over the business cycle. Before 2001, newspapers ads were negatively correlated with business cycles.


In general, the probability of receiving a job offer is also dependent on idiosyncratic differences across workers. Most specifically, the length of one’s unemployment spell is likely to affect their search intensity, even holding conditions in the labor market constant. In other words, there are non-idiosyncratic as well as idiosyncratic shocks that affect an individual’s search intensity. The natural way to model this would be to include the observed duration of unemployment for each worker. As is discussed later in the paper, however, the data is limited on this account. But even when added to the model, it has no effect on search intensity in our sample. See footnote 11.


ATUS codes for these activities are t050481, t050403, t050404, t050405 and t050499.


Though not shown in Figure 1, average search intensity for all workers follows the same general trends over time.


Unfortunately, using the respondent’s unemployment duration to proxy for idiosyncratic shocks of an individual's arrival rate is not easily done with the data. While the CPS interview records the duration of unemployment, the ATUS survey, which is conducted 2–5 months following the final CPS interview, does not. While we can accurately infer those workers who lost their jobs between the CPS and ATUS interview dates, we have no way of knowing when they lost their job. Given that our data encompasses the first years of the recent recession, a large number of the unemployed in our sample turn out to be exactly these “recently unemployed” workers. This study has the same limitations in this regard as Krueger and Mueller (2010). Nevertheless, when duration is added to the model it is insignificant and does not affect the sizes or significance of any of the other variables. The inference we take from this is that all the important effects proxying the probability of receiving a job offer are captured in the VU ratio.


In a previous version of the paper we also included stock market fluctuations. However, the ATUS data does not include information on stock ownership. In addition, aggregate stock market fluctuations are highly correlated with labor market conditions. For this reason, we estimate a model where we use stock prices as an instrument for the VU ratio.


While our results indicate Steward’s critique is valid in our case, we do report Tobit estimates from various model specifications for completeness (see Appendix, Table 3).


It is important to control for these because each individual is interviewed only on one day. As a result of the survey design, there is tremendous variation in activities depending on the day of the week and whether or not the diary day was a holiday. Likewise, there are also seasonal differences, which is why controlling for month of the year is important.


In fact, the Tobit estimates are slightly more significant in the statistical sense. However, the elasticities are roughly half of the size of those found in OLS (see Table 6 in the Appendix). This is precisely what previous research suggests should happen, as Stewart (2009) shows that when the proportion of censored observations exceeds .80 (as ours does) the bias in Tobit estimates grows to over 50%.


In our model, wealth effects are proxied by maximum weekly UI benefits and regional housing prices. Across all the specifications, UI benefits appear to have a small, but positive effect on search intensity. In fact, the elasticity of search with respect to UI benefits is less than 0.10. However, if we run the model only for those workers who are eligible for UI benefits following Krueger and Mueller (2010), we do get a negative coefficient on UI benefits, though it is not statistically significant.


Note that this holds even though we have controlled for other characteristics such as predicted wage, marital status, the presence of children, and other demographics that would be positively correlated with homeownership.


The fact that we are able to identify a significant wealth effect through housing prices is especially noteworthy given the relatively crude proxy we have for changes in housing wealth. All we know is whether a worker owns a home or not. Neither the CPS nor the ATUS provides information regarding the amount of equity individuals have in their home or other the existence of other assets. Presumably, better information about each individual’s wealth would result in more precision in these estimates. Nevertheless, the results appear to support the hypothesis that wealth affects search intensity. Further interpretation of these wealth effects are discussed in the following sections.


For example, a 30% drop in housing values results in a 0.30*32/53.04 = 49.26% increase in search time. Using the weighted average of 32 min a day, this is 16 additional minutes per day, or 111 min per week of search time.


Copyright information

© Springer Science+Business Media New York 2013