Excess commuting and frictions in the labor market

Deschacht, Nick; De Bruyne, Karolien Article Excess commuting and frictions in the labor market Journal of Applied Economics Provided in Cooperation with: University of CEMA, Buenos Aires Suggested Citation: Deschacht, Nick; De Bruyne, Karolien (2020) : Excess commuting and frictions in the labor market, Journal of Applied Economics, ISSN 1667-6726, Taylor & Francis, Abingdon, Vol. 23, Iss. 1, pp. 600-617, https://doi.org/10.1080/15140326.2020.1812476 This Version is available at: https://hdl.handle.net/10419/314109 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ Journal of Applied Economics ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/recs20 Excess commuting and frictions in the labor market Nick Deschacht & Karolien De Bruyne To cite this article: Nick Deschacht & Karolien De Bruyne (2020) Excess commuting and frictions in the labor market, Journal of Applied Economics, 23:1, 600-617, DOI: 10.1080/15140326.2020.1812476 To link to this article: https://doi.org/10.1080/15140326.2020.1812476 © 2020 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. Published online: 30 Sep 2020. Submit your article to this journal Article views: 966 View related articles View Crossmark data Citing articles: 1 View citing articles Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=recs20 ARTICLE Excess commuting and frictions in the labor market Nick Deschacht and Karolien De Bruyne KU Leuven, Department of Economics (ECON) ABSTRACT We propose a model of excess commuting based on search costs in the labor market and show how the equilibrium rate of excess commuting is determined by the degree of geographical job concentration – without neglecting the importance of the size of the labor market and commuting costs. We test – and largely confirm – the main predictions of our model using Belgian population data on commuting flows between all its 589 municipalities. Our approach is to aggregate the data into 640 sectors and skill-specific groups in order to generate heterogeneity in the excess commuting rate. We find that workers in sectors with a high degree of job concentration have lower rates of excess commuting and that workers that operate in larger labor markets, such as higher educated workers and men compared to women, have higher rates of excess commuting. ARTICLE HISTORY Received 10 July 2019 Accepted 15 August 2020 KEYWORDS Excess commuting; job search 1. Introduction Excess commuting refers to the commuting that exceeds the minimum required commuting in an area (Hamilton, 1982; Jun, Chong, Wen, & Kwon, 2018; Kanaroglou, Higgins, & Chowdhury, 2015; Ma & Banister, 2006; Rouwendal, 1998; Small & Song, 1992; van Ommeren & van der Straaten, 2008). This paper investigates how the rate of excess commuting varies across labor markets by proposing a model based on imperfections in the job search process while assuming no residential relocation. Excess commuting is important because commuting time involves opportunity costs in the sense that commuting time cannot be spent working or engaging in leisure activities. Full-time employees in OECD countries on average spend over 30 minutes per day commuting (OECD, 2011), but many workers spend even more time and commuting times are increasing because of congestion problems. For example, the number of hours lost in road traffic jams has almost doubled over the past 10 years in Flanders, the main economic region of Belgium (Vlaams Verkeerscentrum, 2019). In a seminal study, Hamilton (1982) reports very large levels of excess or `wasteful’ commuting and provocatively interprets these as evidence against the monocentric model in which workers choose residential locations by minimizing commuting costs conditional upon housing prices. Hamilton shows that actual commuting is almost what one would expect if commuting were random and concludes that the monocentric model “does an almost unbelievable bad job of predicting commuting CONTACT Nick Deschacht [email protected] Warmoesberg 26, 1000 Brussels, Belgium JOURNAL OF APPLIED ECONOMICS 2020, VOL. 23, NO. 1, 600–617 https://doi.org/10.1080/15140326.2020.1812476 © 2020 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/ by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. behavior.” Hamilton’s findings initiated some debate on the degree of excess commuting and on how to measure it, but the literature appears to agree that excess commuting is substantial (Kanaroglou et al., 2015; Ma & Banister, 2006; Small & Song, 1992). The question is what explains the presence and the degree of excess commuting. First, some scholars suggest that people may not bother about commuting because commuting costs are relatively limited, for example because of low gasoline prices or transport subsidies. Some even suggest that people value commuting itself or value a clear separation between work and residence (see Ma & Banister, 2006), but this is hard to swallow given the importance of commuting times and congestion in the public debate in many countries and the evidence about the role of congestion in worker’s job acceptance decisions (Flemming, 2019). Second, observed excess commuting might reflect the fact that people are minimizing overall travel rather than just individual commuting. Workers may be willing to make longer commutes if that reduces other travel, such as for recreational trips, and in two-worker households, optimizing one partner’s commute could increase the commute of the other partner. However, many scholars feel this is not the whole explanation (Hamilton, 1982). Third, the more recent literature sees excess commuting as a result of frictions in the labor and housing markets which prevent a more efficient matching of workplaces to places of residence (Crane, 1996; Larsen, Pilegaard, & van Ommeren, 2008; Rouwendal, 1998; van Ommeren & van der Straaten, 2008). This line of reasoning requires imperfections in both the labor market and the housing market because excess commuting would largely be avoided if there were either perfect residential mobility or perfect job mobility. Frictions in the housing market arise from lack of information about alternative housing and from moving costs, which are thought to be higher for homeowners but which also include psychological costs such as the stress of moving house. Frictions in the labor market arise from reduced job mobility through firm-specific human capital and from incomplete information about vacancies and search costs. This paper proposes a simple model of excess commuting as a result of search costs – and hence frictions – in the labor market. We include frictions in the housing market as well by assuming no residential relocation. We focus on how the rate of excess commuting is determined by a limited number of labor market parameters: the size of the labor market, the distribution of jobs and workers across places in the market, the reservation wage of job seekers and the level of commuting costs. We use a dynamic model to derive an expression for the equilibrium rate of excess commuting, which is determined by workers’ endogenous job acceptance decisions who trade-off commuting costs and search costs, where the latter is the time cost of rejecting a job and waiting for an acceptable offer to arrive – unlike in Wasmer and Zenou (2002). The outcome of the model is a spatial matching of jobs to workers which is suboptimal in the sense that some workers could in principle swap jobs in ways that reduce the excess commuting and increase welfare. In other words, one worker may live in place A and work in B while a similar worker lives in B and does the same job in A, because workers accept job offers while they do not know when to expect a similar offer closer to home. While the outcome of the model itself is suboptimal, firms and workers maximize their individual welfare. If, for example, two commuting workers with the same reservation wages swap jobs so that they become local workers, then the elimination of the commuting costs produces an JOURNAL OF APPLIED ECONOMICS 601 additional surplus in the market which could then be shared between the worker and the firm. Our empirical study tests the main predictions of this model using commuting data for the Belgian working population in 2012. We propose a measure for the rate of excess commuting which is bounded between 0 and 1, defined as the non-required commuters as a proportion of the workers who could potentially work locally. We consider commuting flows as trade-in labor between localities in the labor market and illustrate how concepts and indicators that are commonly used in the analysis of international trade patterns can also be applied to commuting. In the main analysis, our data are aggregated into 640 sectors and skill-specific groups in order to generate heterogeneity in the excess commuting rate and study its determinants. This paper contributes to the literature by proposing a model that sets out to explain heterogeneity in the rate of excess commuting across groups. More in particular, our paper predicts (and shows empirically) that a higher job concentration results in less excess commuting. The intuition is that in very concentrated sectors (such as mining), the minimum required commuting is very high so there is less need for excess commuting. The effect of job concentration on excess commuting is mechanical, and therefore not surprising: the novelty of this paper is that our model of excess commuting allows for an analytical derivation of the effect of job concentration, and that we present empirical estimates of the size of this effect. A small number of earlier studies have applied job search models to excess commuting, but these are concerned with obtaining estimates of the degree of excess commuting (Crane, 1996; van Ommeren & van der Straaten, 2008). To the best of our knowledge, there is only one paper that also focuses on the role of search costs in the labor market, assumes no residential mobility and proposes a model that shares with ours the prediction that there is less excess commuting when jobs are spatially concentrated, namely van Ommeren and van der Straaten (2008). Our model, however, improves on theirs in several ways: (i) we allow for excess commuting as the direct outcome of the model instead of (indirectly) deriving it through the maximum acceptable commuting distance between two groups (i.e. employed and self-employed); (ii) we theoretically derive the expected impact of the size of the labor market n and the degree of job concentration α on excess commuting directly – instead of focusing on the (indirect) impact of the arrival rate for job offers (λ) and the employment density function F(t); (iii) in our empirical application we are able to analyze the impact of the degree of job concentration at industry level on excess commuting and we focus on studying the variation in excess commuting across labor markets. The remainder of the paper is organized as follows. Section 2 develops a job-search model of equilibrium excess commuting. Section 3 discusses the data and methodology while section 4 presents the empirical results. Section 5 concludes. 2. Model 2.1. A measure of excess commuting In this section, we derive an expression for the rate of excess commuting (ρÞas a function of the percentage of commuters (c) and the rate of required commuting (creq). In order to do so, we also introduce the percentage of local workers (l), the percentage of local 602 N. DESCHACHT AND K. DE BRUYNE workers that could potentially work locally (lp) and the rate of non-required commuting ðcnonÞ. Consider a labor market that consists of n places. Assume that the total number of workers in the labor market equals the total number of jobs in the labor market and that the labor market is a closed system, i.e. workers do not commute to places outside the labor market. The commuting flows within the labor market can then be represented by a commuting matrix A cij  �, which contains as elements cij the percentage of workers, relative to the total number of workers in the labor market, from origin place i who work in destination place j. The sum of all elements of A equals 1 ðP n i¼1P n j¼1 cij ¼1Þbecause the labor market is assumed a closed system. The shares of workers residing in place k are denoted as Wk¼P n j¼1 ckj and correspond to the row sums of A. The shares of jobs in k are denoted as Jk¼P n i¼1 cik and correspond to the column sums of A. Workers either hold a job in their place of residence (local workers) or in another place (commuters). We dichotomize the commuting concept, instead of focusing on a more continuous “commuting distance” measure, in order to simplify our model and because it is much more straightforward in this dichotomous case to define excess commuting – which we define intuitively as someone living in place i and working in place j, while another person lives in place j and works in place i so that they could potentially swap jobs. The percentage of local workers is denoted as l¼P n k¼1 ckk and the percentage of commuters as c¼1l. Since the number of jobs in each place does not necessarily equal the number of workers in that place, a minimal rate of commuting creq could be required in order to balance workers and jobs in each place. The remaining share of workers could potentially work locally lp, with: lpþcreq ¼1 (1) The percentage of workers who could potentially work locally is determined by the distribution of jobs and workers across places. If there are more workers than jobs in k then only Jk workers could potentially work locally. If there are more jobs than workers then all Wk workers can potentially work locally. Thus, lp¼P n k¼1 Min Wk;Jk ð Þ. The rate of commuting c will generally exceed the required rate of commuting creq and the difference will be denoted as the rate of non-required commuting cnon, with: creq þcnon ¼c(2) We now define the rate of excess commuting in the labor market as ρ¼cnon=lp, i.e. the non-required commuters as a proportion of the workers who could potentially work locally. Using (1) and (2) allows us to rewrite the rate of excess commuting as: ρ¼ccreq 1creq (3) JOURNAL OF APPLIED ECONOMICS 603 For a given distribution of jobs and workers across the labor market, creq is constant so that the rate of excess commuting ρ is a linear increasing function of the commuting rate c. The intuition is that, for a given distribution of workers and jobs across space, additional commuting implies a higher rate of excess commuting. A special case arises when in every place the number of workers equals the number of available jobs. In that case, there is no required commuting (creq ¼0), so that all commuting is excess commuting and the rate of excess commuting simplifies to the commuting rate: ρ¼c:From Equation (3) it is also clear that ρ ranges between 0 and 1: there is no excess commuting (ρ¼0) when all commuting is required (c¼creq) and the rate of excess commuting reaches a maximum (ρ¼1) when all workers commute (c¼1). ρ is not defined in the limiting case where all workers are required to commute (creq ¼1Þ. For example, this would be the case in a labor market with two places when all workers live in one place while all the jobs are in the other place. Also note that ρ cannot reach 1 in the exceptional case where a place k has more jobs Jk than the total number of workers available in the other places W� k so that some of the workers of k have to work locally (see appendix A1). Table 1 presents numerical examples of commuting matrices that correspond to minimum and maximum excess commuting in a labor market with three places, where workers are uniformly distributed across space but jobs are not. Our aim is to study the heterogeneity of the rate of excess commuting across labor markets and the ways in which the rate of excess commuting depends on labor market characteristics such as its size, the degree of job concentration and the presence of commuting costs. With this purpose in mind, we develop a dynamic model of the labor market in which an equilibrium rate of excess commuting arises endogenously as a steady-state outcome. 2.2. A dynamic model of excess commuting In order to clarify the intuition behind our model and its underlying assumptions, we perform a simulation. Suppose there are 1000 workers who are uniformly distributed across 10 places and who initially work in their place of residence, so that there is no excess commuting in ‘year 1ʹ. Assume that every year 100 randomly drawn workers retire, that their jobs become vacant and remain in the same place and that 100 new workers enter the labor market as job seekers. Vacancies are matched to job seekers by a process in which a wage is drawn from a wage distribution and offered to a random job seeker. If the offered job is located in the place of residence of the job seeker, then the job seeker accepts the offer if the wage exceeds his reservation wage (say 1800). If the job is located in a different place, then the wage is compared to a fixed commuting cost (say Table 1. Numerical examples of minimum and maximum rates of excess commuting. No excess commuting Maximum rate of excess commuting A¼ :17 0 :17 0:33 0 0 0 :33 0 @1 AA¼ 0 0 :33 :17 0 :17 0:33 0 0 @1 A ρ¼ccreq ð Þ=1creq ð Þ ¼ :17 :17ð Þ=1:17ð Þ ¼ 0%ρ¼ccreq ð Þ=1creq ð Þ ¼ 1:17ð Þ=1:17ð Þ ¼ 100% Note: The commuting rate c is the sum of the off-diagonal elements. The rate of required commuting for both matrices is creq ¼1lp¼1P 3 k¼1 Min Wk;Jk ð Þ ¼ 1:17 þ:33 þ:33½ � ¼ :17: 604 N. DESCHACHT AND K. DE BRUYNE 200) in addition to the reservation wage (so 2000 in total). All workers are assumed to have the same reservation wage, which can be thought of as the value of leisure. If the job seeker rejects the wage offer, then the same offer is made to the second job seeker in a random line, and so on until the vacancy is filled. If no job seeker accepts the offer, then a new wage offer is drawn. Figure 1 shows how the rate of excess commuting in this simulation appears to converge towards a steady-state level. This simulation illustrates the payoffs, the maximization problem faced by workers and the timing in our model. First, a group of workers retire and an equal number of job seekers enter the labor market. Wage offers are then randomly drawn and made to each job seeker with three possible outcomes: (i) a local job offer is accepted; (ii) a commuting job offer is accepted or (iii) the job offer is not accepted which leads to a new wage draw (replay). For each job offer the worker uses a decision rule in which the wage offer is compared to the implied commuting cost and her reservation wage, which is set to maximize the expected discounted streams of future income. The next section defines the model in a formal way in order to derive how labor market characteristics determine the rate of excess commuting. 2.3. Equilibrium excess commuting We now derive an expression for the equilibrium rate of excess commuting in the labor market (ρ�), as a function of a limited number of labor market parameters: the level of commuting costs (τ), the reservation wage of job seekers (ϕ), the size of the labor market (n) and – especially – the distribution of jobs and workers across places (by means of a concentration parameter α). Modeling these four determinants is a simplification because factors outside this model, such as preferences for commuting, may affect excess commuting patterns as well. The approach we follow to derive an 0 .5 1 rate of excess commuting 0 10 20 30 40 time Figure 1. Simulation of the rate of excess commuting. JOURNAL OF APPLIED ECONOMICS 605 expression of the equilibrium rate of excess commuting is to look at the steady-state solution in a dynamic model of the labor market in which over a period of time (Δt) a number of workers retire from the labor market (job destruction) and new job seekers enter the labor market and are matched to job offers (job creation). This rate of replacement is indicated by δ:Over this period of time, the stock of excess commuters in the labor market experiences an inflow from non-employment when job seekers accept job offers in places other than their residence and an outflow to nonemployment of commuters who retire from the labor market. The steady-state equilibrium in the model is derived by equating these inflows and outflows. The probability that the job seeker receives and accepts a local job offer is P matchl ð Þ while the probability that the job offer is a commuting/local job is, respectively P oc ð Þ and P ol ð Þ. The wage offers wð Þ are assumed to be drawn randomly from a normal distribution with cumulative density function F wð Þ. Finally, the proportion of jobs in a place is given by a probability density function h pð Þ, where p is the percentile rank of the places ranked in an increasing order by their proportion of jobs. Assume that over a period Δt workers leave the labor market at a rate δ to be replaced by an equal number of workers who initially enter non-employment. The distribution of jobs and workers is assumed to remain constant over time, so a new job seeker enters the labor market in the same place where a worker retires and a new job is created in the same place where an old job was destroyed. The latter assumption fixes the distribution of jobs and workers over time so that creq is constant over time and so that ρt, the rate of excess commuting at time t, is a linear function of the rate of commuting at time t: ρt¼ctcreq ð Þ=1creq ð Þ. A steady-state equilibrium in the rate of excess commuting (ρtþ1¼ρtÞrequires the rate of commuting to be in equilibrium (ctþ1¼ct) as well as, which will be more convenient in the derivation, the percentage of local workers (ltþ1¼lt). Over a period Δt, a group of δΔt workers retire and an equal number of job seekers enter the labor market. The outflow to retirement from the stock of local workers is ltδΔt, i.e. the probability that a retiring worker is a local worker (lt) times the number of retiring workers. The inflow of job seekers into the stock of local workers over a period Δt is the number of job seekers (δΔt) times the probability that a job seeker is matched to a local job in his place of residence, which we denote P matchl ð Þ. Equating the inflow to the outflow yields the steady-state equilibrium rate of local workers l�¼P matchl ð Þ, the equilibrium rate of commuting (given that c¼1l): c�¼1P matchl ð Þ (4) and the associated equilibrium rate of excess commuting: ρ�¼c�creq ð Þ=1creq ð Þ (5) In order to bring commuting costs and the size of the labor market into our model, we develop Equation (4) by expanding P matchl ð Þ. Call the probability that a job seeker receives and accepts a local job offer P ol&accl ð Þ ¼ P ol ð ÞPðaccljolÞ, where the first factor is the probability that a job seeker receives a local job offer and the second factor is the probability of accepting conditional on receiving a local offer. The probability that a job seeker is matched to a local job can now be written as the probability that he receives and 606 N. DESCHACHT AND K. DE BRUYNE is .83, which implies that in most sectors the bilateral commuting flows are largely oneway. The degree of one-way trade in labor is particularly high in sectors that are geographically concentrated, such as mining (ONEWAYs¼1) and air transport (.98). The sectors with the largest degrees of two-way commuting are construction (ONEWAYs¼.49), retail (.54), residential care (.56), and education (.57). The third indicator measures the extent to which a municipality is engaged in a pattern of intra-industry trade in labor (exporting labor in one sector in exchange for labor in another sector) or inter-industry trade (two-way commuting in each sector). We define INTER INDUSTRYi¼P iP j XijsMijs j j XijsþMijs :wi ijs as a weighted average of Grubel-Lloyd type indices for each sector and pair of municipalities with weights that represent the share of each bilateral trade flow in the overall trade to and from that municipality: wi ijs ¼ ðXijs þMijsÞ=P iðXijs þMijsÞ. The average of the indicator across all municipalities is .76 which suggests that most municipalities are characterized by an inter-industry pattern of trade in labor. The municipality with the maximum value is Beauvechain (INTER INDUSTRYi¼.89), where 76% of all jobs are in the military (it hosts an air base) and which imports mainly military personnel while its outgoing commuters are in different sectors. The value is lowest in municipalities that are more diversified in terms of sectoral composition, such as Izegem (INTER INDUSTRYi¼ .59). The inter-industry indicator should be interpreted with some caution because the measure is affected by overall imbalances (as measured by ONEWAYi). This is a known issue in international trade but it is more serious here because overall imbalances tend to be limited in international trade. When the sample is restricted to municipalities that have similar levels of overall outgoing and incoming commuting (ONEWAYi<.1), then the mean inter-industry indicator falls – but only to a limited extent (mean = .71, N = 76). Given that a substantial part of the observed commuting is two-way and part of that is of the intra-industry type, the question naturally arises what is the extent of excess commuting. 4.2. Excess commuting We now present the main results of our empirical analysis regarding the determinants of the rate of excess commuting, where we focus on the impact of both the size of the labor market and the degree of geographical job concentration. How do we define the size of the labor market when we take our model to the data? The literature on geographical labor markets is well aware of the fact that the frequently used regional administrative boundaries are a poor approximation for actual labor markets, especially since this does not capture trends over time in the geographical mobility of workers (Nimczik, 2018). In our model, the actual (or effective) size of the labor market is the number of places with accepted job offers – where the largest possible labor market is the total number of places n in the area under study. The actual size of a labor market differs across groups of JOURNAL OF APPLIED ECONOMICS 613 workers because some groups are more likely to search, receive, or accept job offers at larger distances. For example, low-skilled workers may have less information about vacancies at larger distances and women may be less likely to accept offers for such jobs because of family obligations. This understanding of the actual size of a labor market can also be found in the work of Manning and Petrongolo (2017), who find that labor markets are quite local since the matching rate sharply decays with distance. As explained before, the commuting data are aggregated into 640 groups defined by sector, age categories, and gender-to-generate heterogeneity in the excess commuting rate. The regression analyses estimate linear equations of the type ρijk ¼θαijk þβxijk þsiþajþgkþεijk, where ρijk is the rate of excess commuting in a particular labor market for workers in sector i, age j and gender k, αijk is the degree of geographical job concentration in the labor market defined by i, j and k, xijk the educational level in the labor market and si, aj and gk are vectors of fixed effects indicating the group’s sector, age and gender, and εijk is a random error term. Figure 3 describes the distribution of the rate of excess commuting and its relation with the core explanatory variables. Panel (a) demonstrates that there is a substantial degree of heterogeneity in the rate of excess commuting across groups, with values ranging from close to zero to .86. Panel (b) shows that there is less excess commuting in female labor markets and that excess commuting is lower in age groups above 40 years old (for both men and women), which is suggestive of on-the-job search 2 or residential relocation. Panel (c) shows a negative bivariate correlation between excess commuting and the degree of geographical job concentration across sectors (r = -.24, p < .05, N = 81). The residuals in panel (c), i.e. the vertical distances between the data points and the estimated regression line, are indicative for the role of education: for example, domestic workers (sector 97), who are mostly low educated, have a relatively low rate of excess commuting, whereas software developers (62) have a relatively high rate. Panel (d) further demonstrates the positive correlation between excess commuting and the proportion of highly educated workers across all 640 groups (r = .29, p < .001). Table 3. Regression of the rate of excess commuting. (1) (2) (3) (4) Base Clustered Unweighted Industry FE Gender: female (ref = male) −.062*** −.062*** −.059*** −.066*** Age: 30–40 (ref = .<30) .008 .008 .006 .029*** Age: 40–50 −.010 −.010 −.008 −.004 Age: 50+ −.047** −.047*** −.044** −.051*** Highly educated (%) .286*** .286*** .330*** .017 Concentration: ln(α) −.048*** −.048** −.055*** No Industry (81 categories) No No No Yes R-squared .22 .22 .23 .89 Number of groups (N) 640 640 640 640 * p < .05; ** p < .01; *** p < .001. All models are linear and include a constant. All models, except (3), use weights that are proportional to the log of the number of workers in each group. All tests use robust SE, except in (2) where SE are clustered at the industry level. 2 The finding about the above-40-year-olds is indeed suggestive of on-the-job search if one looks at panel (d) in Figure 2. Combining it with the estimation results in Table 3, we might however obtain a more nuanced result. It appears to be mainly the above 50-year olds that have a lower rate of excess commuting. Although this might indeed be due to onthe job-search, in this age category it could also be caused by the fact that workers are less “prepared” to commute as they grow older. 614 N. DESCHACHT AND K. DE BRUYNE Table 3 presents the results of the regression analysis of the rate of excess commuting. Our main specification (the base model in column 1) indicates that the predicted rate of excess commuting among women is 6.2 percentage points below that of men and that workers over 50 years old have a predicted rate that is 4.7 points below that of workers younger than 30 years old. The estimated coefficient for education implies that the predicted rate of excess commuting in a sector with only highly educated workers (100%) is 28.6 percentage points above that of a sector with no (0%) highly educated workers. In order to appreciate the size of this effect, it is worth noting that the interquartile range for the educational variable is around 25 percentage points, which implies that a sector at the third quartile in the education distribution has a predicted rate of excess commuting that is about 7 percentage points (.25 times .286) above that of a sector at the first quartile in the education distribution. Thus, the effect of education appears to be of the same order of magnitude as the effects of gender and age. The estimated effect of the degree of geographical job concentration implies that doubling the concentration parameter α reduces the predicted rate of excess commuting by 4.8 percentage points. The interquartile range on α is about 1.4 log points, so a sector at the first quartile of the concentration distribution has a predicted rate of excess commuting of 6.7 percentage points above that of a more concentrated sector at the third quartile. Models (2) and (3) demonstrate that the results are robust to the use of industry-level clustering of standard errors and the application of weights in the base specification. Model (4) adds industry-fixed effects which leave no variability in the job concentration variable and little variation in the educational variable. The gender and age effects largely remain, except for the 30–40 year olds who have higher predicted rates of excess commuting than younger workers in the same sector. 5. Discussion and conclusion We develop a simple model of excess commuting as a result of search costs in the labor market and focus on how the resulting excess commuting in equilibrium is determined by a limited number of labor market parameters. We show that a larger labor market implies more excess commuting, while higher commuting costs and a higher degree of geographical job concentration decrease the rate of excess commuting. We test the main predictions of our model using data on commuting flows between all 589 Belgian municipalities and find confirmation of our theoretical predictions. We aggregate the data into 640 sectors and skill-specific groups in order to generate heterogeneity in the excess commuting rate. We find that women, workers aged above 50 and workers in sectors with a higher job concentration have a lower rate of excess commuting. Higher educated workers on the other hand have a higher rate of excess commuting. We can only speculate about the reasons for these group differences in excess commuting rates, so further research is needed to investigate to what extent these differences are due to preferences versus more structural constraints and determinants. An important implication of our findings is that there are benefits to what is generally considered to be “excess” commuting. It is useful to consider excess commuting as excess trade in labor, which relates it to the literature on excess trade in goods and opens avenues for both methodological and theoretical renewal. In the international trade literature, intra-industry goods trade was initially considered wasteful. The development JOURNAL OF APPLIED ECONOMICS 615 of New Trade Theory altered the common conception of intra-industry trade by showing how, in the presence of internal economies of scale, it may produce gains from trade and increase the variety of products available to consumers. The debate on excess trade in labor is very similar. If excess commuting arises from search costs in the labor market, then there are benefits to (“excess”) commuting. Just like reducing intra-industry trade in goods by producing locally would increase fixed costs per input, reducing excess commuting (for example, by a tax on commuting costs to discourage excess commuting) might result in longer job search. As far as our model is concerned, we see some possibilities for future extensions. First, we model and measure excess commuting using essentially a binary distance measure since commuting is defined as not working locally. Future work could include other distance measures, for example, by constructing a distance-weighted measure of excess commuting. Second, the model could be extended to allow for residential relocation. In a world of perfect residential mobility in which every worker moves to her job location, there would be no excess commuting (in fact, there would be no commuting whatsoever). In a more realistic world of imperfect residential relocation towards jobs, the equilibrium rate of excess commuting can be expected to be smaller than in the case of no residential mobility which our model assumes, because the residential relocation partly offsets the continuous inflow of excess commuters resulting from the job search process. Finally, our empirical analysis could be strengthened by using more detailed data on the skills offered by workers and required by jobs. Within each labor market, workers are assumed to be interchangeable so that errors in defining and delineating labor markets are likely to affect the resulting estimates of excess commuting. Richer data could allow future research to distinguish more clearly between and identify the effects of the determinants of excess commuting. Our model of excess commuting has implications for the development of excess commuting over time. The tendency towards market concentration in many industries, with an increasing average firm size and decreasing number of firms (De Loecker & Eeckhout, 2017; The Economist, 2016) can be expected to mechanically reduce the rate of excess commuting. However, this decrease in the rate of excess commuting can be attenuated if concentration in the presence of economies of scale leads to an increase in productivity. In that case, the overall effect of increased spatial concentration on excess commuting is ambiguous, which is clear from Equation (3). On the other hand, and more importantly, increasing levels of education and improvements in work/life balance through policies such as childcare provision or teleworking, allow workers to find and accept commuting job offers, so that excess commuting is not likely to end being a feature of labor markets any time soon. Disclosure statement No potential conflict of interest was reported by the authors. Funding The research leading to these results has received support under the European Commission’s 7th Framework Programme (FP7/2013-2017) under grant agreement n°312691, InGRID – Inclusive Growth Research Infrastructure Diffusion. 616 N. DESCHACHT AND K. DE BRUYNE Notes on contributors Nick Deschacht is Associate Professor at the Faculty of Economics and Business of KU Leuven and guest professor in labor economics at Antwerp University. His research is focused on applied micro-econometrics and the political economy of labor markets, with an emphasis on issues relating to discrimination in the labor market and wage formation. Karolien De Bruyne is an Assistant Professor of Economics at KU Leuven, Belgium. Her primary research interests are international trade, development economics, labour economics and location decisions of firms. References Borjas, G. J. (2003). The labor demand curve is downward sloping: Reexamining the impact of immigration on the labor market. Quarterly Journal of Economics, 118(4), 1335–1374. Crane, R. (1996). The influence of uncertain job location on urban form and the journey to work. Journal of Urban Economics, 39(3), 342–356. De Loecker, J., & Eeckhout, J. (2017). The rise of market power and the macroeconomic implications. NBER WP. Dickerson, A. P., Hole, A. R., & Munford, L. A. (2014). The relationship between well-being and commuting revisited: Does the choice of methodology matter? Regional Science and Urban Economics, 49, 321–329. The Economist. (2016, March 26). Too much of a good thing. Flemming, J. (2019). Costly commuting and the job ladder. Unpublished working paper. Grubel, H. G., & Lloyd, P. J. (1971). The empirical measurement of intra-industry trade. Economic Record, 47(4), 494–517. Hamilton, B. W. (1982). Wasteful commuting. Journal of Political Economy, 90(5), 1035–1053. Jun, M., Chong, S., Wen, F., & Kwon, K. (2018). Effects of urban spatial structure on level of excess commutes: A comparison between Seoul and Los Angeles. Urban Studies, 55(1), 195–211. Kanaroglou, P. S., Higgins, C. D., & Chowdhury, T. A. (2015). Excess commuting: A critical review and comparative analysis of concepts, indices, and policy implications. Journal of Transport Geography, 44, 13–23. Larsen, M. M., Pilegaard, N., & van Ommeren, J. (2008). Congestion and residential moving behaviour. Regional Science and Urban Economics, 38(4), 378–387. Ma, K.-M., & Banister, D. (2006). Excess commuting: A critical review. Transport Reviews, 26(6), 749–767. Manning, A., & Petrongolo, B. (2017). How local are labor markets? Evidence from a spatial job search model. American Economic Review, 107(10), 2877–2907. Nimczik, J. S. (2018). Job mobility networks and endogenous labor markets. Unpublished Working Paper. OECD. (2011). “Graph 6.3 - Commuting time: Minutes per day, persons in full-time employment. In How’s Life?: Measuring Well-being.Paris: Author. doi:10.1787/9789264121164-graph57-en. Roberts, J., Hodgson, R., & Dolan, P. (2011). “It’s driving her mad”: Gender differences in the effects of commuting on psychological health. Journal of Health Economics, 30(5), 1064–1076. Rouwendal, J. (1998). Search theory, spatial labor markets and commuting. Journal of Urban Economics, 43(1), 1–22. Small, K. A., & Song, S. (1992). “Wasteful” commuting: A resolution. Journal of Political Economy, 11(4), 888–898. van Ommeren, J. N., & van der Straaten, J. W. (2008). The effect of search imperfections on commuting behaviour: Evidence from employed and self-Employed workers. Regional Science and Urban Economics, 38(2), 127–147. Vlaams Verkeerscentrum. (2019). Verkeersindicatoren. indicatoren.verkeerscentrum.be Wasmer, E., & Zenou, Y. (2002). Does city structure affect job search and welfare? Journal of Urban Economics, 51(3), 515–541. JOURNAL OF APPLIED ECONOMICS 617