1. UNEMPLOYMENT, JOB SEARCH BEHAVIOR AND
MOBILITY COSTS:
An Empirical Analysis of the Spatial Mismatch Hypothesis
Boccardo Serena – 11 February 2015
!
2. SUMMARY
Introduction
Theoretical framework for spatial mismatches
State of the art
Research Question (1)
Methodological Specification
Research Question (2)
Reasons for the Italian case
Further RQ & Related Fields
Comments & Suggestions
PhD Student: Boccardo Serena –
University of Trento
3. INTRODUCTION:
RELEVANCE OF THE UNEMPLOYMENT – MOBILITY RELATIONSHIP IN EU
ü Mounting signs of entrenched mismatch across euro area labour markets
ü Diversity in Beveridge curve movements at the country level, with strong evidence of notable outward shifts
ü Typical cyclical movements along a pre-existing Beveridge curve (and thus the transitory effects of low demand)
or the first signs of an outward shift,marking the start of a structural change?
An obvious factor that may help to explain the Beveridge curve movements observed over the course of the crisis would be an
increase in skill mismatch (that is, the discrepancy between the skills of labour force participants and the skill needs of employers)
across the euro area.
(...)The gap appears higher at the regional level than at the intra-country level, suggesting that at
least part of the strong skill mismatch evident at euro area level could be significantly alleviated
through higher inter-regional labour mobility. Source: ECB Monthly Bullettin, October 2014
4. Cross-country heterogeneity in LT
unemployment"(...) The significant downward trend in job-
finding rates among those unemployed for 12
months or more warrants particular attention
from policy-makers as it points to an elevated risk
of a marked increase in structural
unemployment."
"Stressed economies have, on the whole, suffered much
steeper increases (...)"
Some scholars claimed that declining geographic mobility
has been a source of mismatch (Frey, 2009; Katz, 2010)
Relevance of LT for job finding
RELEVANCE OF THE LONG-TERM UNEMPLOYMENT FOR JOB FINDING
RATES
Source: ECB Monthly Bullettin, October 2014
5. Theoretical Framework for Structural Unemployment
Herz and van Rens (2011)
Local sub-markets with inside search frictions - Cross-markets workers movements entail adjust. costs
Homogeneity in the above factors lead market conditions to be identical in all submarkets =
unemployment entirely due to search frictions
Worker and job mobility costs, wage bargaining costs and heterogeneity in matching efficiency generate
dispersion in labor market conditions and therefore structural unemployment.
6. State of the art (1)
• Theoretical framework for the SMH
- Residential segregation explains (part of) unemployment rates gaps for minorities (Kain, 1968; Holzer, 1986)
- Physical distance limits job search efficiency (Seater, 1979; Holzer, 1986; Barron&Gilley, 1981; Chrinko,1982)
- Spatial distribution of residential allocation is a result of equilibrium on LM (Smith&Zenou, 2003; Bar-El, 2006)
• Main arguments of the SMH (Selod et al.,2007):
1) residential segregation deteriorates "employability" (Houston, 2005; Mc Quaid&Lindsay, 2002 )
2) r.s. affects the quality of the social networks used in job-search activities - info barriers (Selod et al.,2 007)
3) r.s. spurs labor-market discrimination (Kain, et al. 1968; Shen, 1998, 2001)
• Motivation for the SMH
- Less explored approach wrt skills/edu mismatch (Ilhanfeldt et al., 1998)
- Labor mobility explained traditionally with push&pull determinants (migration studies)
- Job decentralization as a causative factor in the growth of the underclass (Wilson, 1987; Kasarda, 1989)
- High emphasis on free mobility in EU to overcome LM mismatch
7. SMH in empirical works – State of the Art (2)
ü Employment rates by ethnicity (Kain, 1968)
ü Weeks worked in central/suburban areas by ethnicity (Harrison, 1974)
ü Wage gradient changes moving from central-city to suburban areas (Madden, 1985)
ü Net wages/Annual earnings = net of higher cost/time of travel (Holzer, 1987; Ihlanfeldt,
1988)
ü Drop-out rates as wages fall below the res.wage (Holzer, 1986)
ü Unempl_duration as geogr. distance increases (Rogers, 1997)
SAMPLE? Minorities (Kain, 1968; Holzer, 1987) – Youth (Ellwood, 1986;
Ihlanfeldt&Sjoquist, 1989,1990) – Living with family of origin (Selod et. al, 2007)
SPATIAL LEVEL? Intra/inter Metrop.Areas in US but in EU only Paris and Brussels
INDEX? "Job accessibility" instead of individual mobility preferences
a) "house locking" hypothesis - Oswald's thesis – (Farber, 2012; many others)
b) residential distribution with respect to distance from central districts (Kain, 1968), cross-
Metrop. Areas distances (Masters, 1974,1975)
c) commuting times, average travel times (Hutchinson, 1974; Ihlanfeldt&Sjoquist, 1989, 1990)
d) jobs availability within a certain radius (Hutchinson, 1974); import ratio (n.jobs/people over
area or travel times (Ellwood, 1986); commuting times weighted by vacancy durat & turnover
rates for transport/auto-ownership (Shen, 1998, 2001)
8. Research Question (1)
(1) Net of individual mobility preferences, do physical distance from job
vacancies affect the probability of employment?.....
a) Individual-level Probabilities
b) 2 Local-level job accessibility indexes
8
Job Density
JDn= N.Jobsn+m/LFn+m
Distance-to-jobs
n, m = local UA/LADs/NUTS-3 areas
depending on data availability
mi,j=commuting timei,j (car times in Rogers, 2007)
Em= Empl.growth in m since....last year?
Euclidian distance:
center-to-center
9. PhD Student: Boccardo Serena – University of Trento
Methodological Specification(1)
Variables of interest
Unempl.Status i,t -> LONGITUDINAL (Selod, 2007)
DISTANCEn and DENSITYn : as described in previous slide (Selod et al., 2007)
INDIVIDUAL MOBILITY PREFERENCESi : house/car/family/UBs
RESID_NEIGH: very deprived/depr/etc. -> avoiding collinearity
(n=local area; i=individual; t=time period)
Longitudinal vs cross-section
IF i –level & n-level variables are time invariant & Job access measure is time-variant:
we can pick up the effect of local_empl_growth on unempl_prob (Hausman test)
Methodological problems
• ENDOGENOUS JOB/RESID_LOCATION: solved by Selod et al., 2007 through SAMPLE
RESTRICTION e.g. to Youth living with parents
• SELF-SELECTION of local labour demand : out of "bad" neighborhoods
• OMITTED VARIABLES : EDU/PARENTAL -> Sensitivity analysis (Selod et al,2007)
10. Methodological Specification (2)
Logit model
• Model I : unemployment rates + demographics (edu, citizenship, age, gender)
• Model II: includes UA/LADs "type" as controls (e.g. deprivation levels)
• Model III: includes job accessibility measures (density & distance) as controls
Sensitivity analysis
• Mobility preference measures (affecting both resid_location and empl_status)
• Further unobserv_variables check : Rosenbaum&Rubin(1983)
Innovation
• Spatial dimension of analysis
• Longitudinal dimension -> Effect of job opportunities distance and density in a local area
through time
10
11. Brussels area analysis (Selod et al., 2007)
PhD Student: Boccardo Serena – University of Trento 11
12. Research Question (2)
Does physical distance from job vacancies, all else equal, affect
unemployment spell duration? ....
12
ü Standard job-search model à la Mortensen + w*(t) includes search costs &
commuting costs which both increase with distance (Rogers, 1997).
ü The total effect of distance on the escape probability is given by:
Distance affect both the arrival rate and commuting costs : if this effect > search costs -> individuals prefer to remain unemployed
(Escape probability falls and unemployment duration increases)
HP (Rogers, 1997): (1) Both commuting and search costs increase with distance since an individual can sample fewer jobs in a
given period and is likely to have less information about job openings that are farther away; (2) All jobs are centrally located;
(3) Individuals search in locations with the highest associated reservation wage first and continue to search across locations in a
decreasing order according to location specific reservation wages.
ü For each individual there is a reservation wage associated with each location,
reflecting location-specific arrival rates, wage offer distributions, and search costs:
Unempl_Duration : Hazard ratio (Logit form– covariates as above)
13. Little innovation
ü Distance index based on job density rather than employment growth levels
ü NB: different functional forms to discount distance (Rogers, 1997; Hutchinson,
1974)
ü Time-invariant dnm (Euclidian distance)
ü Multiple unempl_spells: longer or average duration – left censoring avoid
ü Recall vs new jobs : look at the distinction
ü Endogeneity of early retirement : sample selection
ü Descriptive statistics: Kaplan-Meier hazards spells
PhD Student: Boccardo Serena –
University of Trento
13
E m JobDensm(t)= (Em(t)- Em(t-1)) /LabForcem(t)
14. PhD Student: Boccardo Serena – University of Trento
A) Long-term internal migration trends relatively under-investigated - new indicators,
new perspective of analysis could explain:
• why inactivity in the 2000s has not been replaced by migration, given the increasing
North-South gap in PIL and income levels?.... (V-L, 2011)
• the circularity of internal migration with long unemployment spells in between (V-L,
2011)
B) Availability of new longitudinal dataset (ADSILC) with job_location & residential
location –> identifying "movers" but not causal relationship
C) Peculiarity of the Italian sample (living with the family of origin)
D) Strong attitude to homeownership -> Evidence in favour of the house-locking HP
(both for owners and for renters)
E) Descriptive evidences showing that discontinuing careers discourage permanent
transfer -> "residential transfer only occurs when job is stable" (Colucci, 2014)
Reasons for the Italian case
14
15. Further RQs & Related Fields
ü Internet access (=proxy for info barriers) lowers the effect of physical
distance?...
ü Comparative perspective : job accessibility in Europe – limit: data on
residential mobility abroad are missing
ü Local level shocks (e.g. lowering local fixed costs of firm entry) effects on
unempl_spells
PhD Student: Boccardo Serena –
University of Trento
15