جستجوی کار در بازار ضخیم

I analyze empirically the effects of urban and industrial agglomeration on both search behavior and the efficiency of matching. The analysis is based on a unique panel data set from the Italian Labor Force Survey micro-data, covering 520 randomly drawn Local Labor Market Areas (66% of the total) over the four quarters of 2002. I compute transition probabilities from non-employment to employment by jointly estimating the probability of searching and the probability of finding a job conditional on having searched. I then test whether these are affected by market size, industrial variety and/or industry specialization. The main results indicate that market size and industry-specialization raise job seekers’ chances of finding employment (conditional on having searched), while industrial variety is not significantly different from zero. Finally, the effect of agglomeration on non-employed individuals’ search behavior cannot be significantly distinguished from zero.

In spite of the recent developments of the theoretical literature on urban search and matching models, the amount of empirical work examining the relationship between agglomeration and search is surprisingly scarce. The applied economic geography literature has generally focused on the impact of agglomeration on productivity, wages or employment growth (see Rosenthal and Strange (2004) for a review), rather than on the process of job search and matching.1 The first empirical papers in the search and matching framework with a local dimension, on the other hand, were mainly directed at correcting the regional aggregation bias, arising when estimating the degree of returns to scale in the matching function without taking into account the interaction of local labor markets (see, for instance, Coles and Smith, 1996 and Petrongolo, 2001). More recently, Petrongolo and Pissarides (2006) decompose individual job finding rates into the product of the probability of receiving a job offer and the probability of accepting it. They find that market size has a positive impact only on the latter (proxied by the mean of the wage offer distribution), although reservation wages rise to fully offset the increase in acceptance rates. Thus, market size does not influence job finding rates on the whole. In this paper I test whether agglomeration affects more the effort individuals devote to job seeking or their employment chances per unit of search (conditional hazard rates), by decomposing unconditional transitions to employment into the product of the probability of searching and the probability of obtaining employment conditional on having searched.2 Knowing whether the shifts in the matching function are due more to technological advances in matching or to individuals’ search choices is important, because local hazard rates and job seekers’ propensity to search are likely to be differently affected by agglomeration externalities (see Section 2). For instance, if agglomeration raised conditional hazard rates to the same extent that it lowered individuals’ search propensity (or vice versa), we would not find any effect on unconditional hazard rates. While the impact of agglomeration is usually studied either at the city or at the industry level, I am able to compare the magnitude of the effect of both market size and industrial agglomeration on job seekers’ probability of finding employment. I measure the effects of agglomeration with three variables at the local labor market (LLM) level: market size, industry specialization and industry diversity. I proxy the former with LLM population level, the latter with the inverse of an Herfindal index of LLM sectoral employment concentration, and industry specialization with either an industrial district (ID) or a super-district (SID) dummy. IDs are LLMs characterized by a high presence of spatially concentrated small- and medium-sized manufacturing firms (see Section 3 and de Blasio and Di Addario (2005) for further details). SIDs are a subset of IDs with a higher incidence of small- and medium-sized manufacturing employment (see Cannari and Signorini, 2000). Most ID and SID enterprises are specialized in one or few stages of a main manufacturing production; one or more firms of the cluster assembles the parts produced by each subcontractor. This system enables the district to achieve economies of scale (external to the single firm but internal to the cluster) that would not be possible to reach otherwise. The advantage of measuring industrial specialization with ID dummies is that: (i) they are officially devised by the Italian Institute of Statistics (ISTAT) and, thus, are reliable; (ii) they partition the entire Italian territory; (iii) they are based on LLMs, and thus are potentially reproducible in the countries where LLMs have already been singled out (e.g. the UK, France, etc.); (iv) they are comparable to the US Cluster Mapping Project (see Porter, 1990 and Porter, 1998); (v) they are an entity recognized by the central and local governments, and have received specific subsidies over time. 3 The results obtained in this paper might then be useful to assess and to inform policy making. To my knowledge, the impact of agglomeration on individuals’ search propensity and conditional hazards to employment has not been analyzed before. Including agglomeration among regressors is important: the probability of the average agent finding a job conditional on having searched falls from 53.6% points (when not including agglomeration variables) to 52.5% points. Thus, computing conditional hazard rates without controlling for agglomeration would overestimate the true effect. Italy is a good country to study agglomeration-induced effects because it has a very limited mobility of labor across LLMs. LLMs can therefore be considered as separated markets to a large extent, minimizing the potential problem of spatial sorting into the most agglomerated markets (for more details see Sections 3.2, 6.2 and 6.3). Indeed, in Italy even the unemployed job seekers, who are generally the most likely to migrate (Dohmen, 2005), are unwilling to move out of their town of residence to find a job. Table 1 provides evidence on the limited mobility of labor in Italy: up to 80% of the non-employed individuals who look for a job are ready to accept an offer only in their LLM of residence, and almost half of them do not intend to seek employment out of their own municipality.4 The table indicates that just 1.1% of the non-employed persons of working age have been absent from their household of residence for more than 1 year and that just 0.2% of the interviewed individuals were also looking for a job. Moreover, none of the interviewed people changed municipality of residence between two consecutive quarters in 2002.5 Overall, my findings indicate that agglomeration affects job seekers’ chances of employment: a 100,000-inhabitant increase in LLM population raises job seekers’ probability of employment by 3.7–3.9% points (up to a 1,900,000-inhabitant threshold), and living in an ID or in a SID increases the probability of finding a job by, respectively, 3.8% and 5.6% points. Conversely, the effect of industrial diversity on employment chances and the impact of any of the agglomeration variables on individuals’ search behavior cannot be significantly distinguished from zero. These results are robust to the use of alternative econometric models (bivariate probit with sample selection, probit and linear probability models). Since selected migration is not the only sort of bias potentially plaguing agglomeration studies, because even in the absence of labor mobility there might be omitted variables correlated to both the agglomeration proxies and the error term of the hazard rate equation, I also run instrumental variable estimations. I instrument agglomeration with three sets of variables at the LLM level: population size in 1861, seismic hazard, and 13 soil-type characteristics. Results confirm the existence of agglomeration externalities improving the efficiency of matches: adding 100,000 inhabitants to the average LLM increases the chances of employment by 1.4–1.6% points, and residing in a SID raises the probability of finding a job by 4.4% points. However, in the IV regressions the ID coefficient is not statistically distinguishable from zero, possibly because the large increase in standard errors with respect to OLS does not enable its precise estimation.6 The paper is structured as follows. The next section presents the theoretical framework, Section 3 the dataset, the variables and the descriptive statistics. Section 4 reports the empirical model, Section 5 discusses the estimation results and Section 6 the robustness checks. Finally, the last section concludes.

In this paper I analyze agglomeration effects on both individual’s search intensity and hazard rates from non-employment to employment. More specifically, I empirically examine whether population size, industry specialization and industrial variety generate overall net positive or negative externalities. I find that only the matching process is affected by agglomeration. In particular, the probability of non-working job seekers finding employment increases with market size and with working in a SDI, while industry variety does not have any statistically significant effect. As to search intensity, on average it is not possible to distinguish an effect statistically different from zero by any of the agglomeration variables. A possible explanation of why the intensity of search does not increase in spite of greater chances of employment is that job seekers are discouraged from bearing the greater commuting costs produced by the presence of a large population mass (i.e. traveling on congested public transportation, spending time in traffic, etc.). These findings suggest that the magnitude of the externalities generated by agglomeration on employment probabilities varies according to both the type and the degree of agglomeration considered. This has potentially important policy implications. It would thus be useful, in terms of policy recommendations, to know whether the absence of market size effects on job seekers’ hazard rates beyond the 1,900,000-inhabitant threshold is due to an excessive congestion in Rome, Milan and Naples, or to the choosiness of job seekers in these cities. If these markets were too congested, reducing their dimension (for a given industrial composition) would generate productivity gains.44 In contrast, if the mechanisms leading to the empirical results involved mainly the productivity of the match (i.e. a lower job seekers’ acceptance probability), the three largest LLMs would not be over-sized (they would, on the contrary, be endowed with higher-quality matches, firms offering more attractive jobs, etc.). Future analysis can investigate in detail the channels through which the empirical results have been established in this paper.