Ann Reg Sci DOI 10.1007/s00168-017-0832-7 ORIGINAL PAPER
Out-migration from the epicenters of the housing bubble burst during and in the aftermath of the Great Recession in the USA Jaewon Lim1
Received: 17 January 2017 / Accepted: 17 May 2017 © Springer-Verlag Berlin Heidelberg 2017
Abstract During the Great Recession in the USA, the four Sand States—Arizona, California, Florida and Nevada—suffered from plunging home prices, massive layoffs and much slower population growth. Some metro areas in these states experienced the unprecedented net loss of population with increasing out-migration. Were the devastating labor market conditions in these four Sand States strong enough to force out a massive number of migrants to other states? Using multinomial logit models, this paper finds that the labor market status of an individual (part-time worker, unemployed and not in labor force) serves as an important factor for out-migration decision during a recession. The pushing effects of labor market status among the at-risk population in devastated labor markets are the largest for movers to intrastate destinations beyond metropolitan borders. The effects of labor market status for interstate out-migration are still significant but largely limited, due to higher moving costs and growing uncertainty. For the at-risk population with lower socioeconomic status, depressing labor market status triggers speculative migration; however, the speculative behavior among the overall at-risk population continuously slowed during the Great Recession, as the rather risk-aversive “wait and see” attitude had increased. JEL Classification J61 · O15 · R23
Electronic supplementary material The online version of this article (doi:10.1007/s00168-017-0832-7) contains supplementary material, which is available to authorized users.
B 1
Jaewon Lim
[email protected] School of Public Policy and Leadership, University of Nevada, Las Vegas, Las Vegas, NV, USA
123
J. Lim
1 Introduction During the housing crisis, the four Sunbelt states, namely Arizona, California, Florida and Nevada, suffered largely from seriously high delinquency rates, rapidly rising foreclosures and devastating labor market conditions. Prior to the Great Recession, these states had benefited from fast population growths, booming housing construction activities, and rapid appreciation in home prices. Some economic analysts have labeled these states “Sand States,” due to similar trends in their housing markets and their proximity to beaches or deserts (Brown 2009). However, during the recession, housing prices in Sand States crashed by nearly twice as fast as the national average. Fast home value depreciation and the resulting underwater mortgages were partially fueled by unsustainable mortgage lending and investor speculation. The rise in unemployment with the subsequent recession further added to rising defaults and the decline in property values. The deteriorating economic conditions in the Sand States resulted in much weaker in-migration and stronger out-migration patterns during the recession. Some metropolitan statistical areas (MSAs) within these states even had net population losses through internal migration, while others experienced much slower population growth. Declining labor demand in these states significantly weakened the attractiveness of these Sunbelt states as labor migration destinations. Moreover, a deteriorating local labor market condition might have been strong enough to push out some local residents to other locations. Comparative analysis on individual socioeconomic characteristics of movers and non-movers among the at-risk population enables an understanding of the indirect effect of the Great Recession on the out-migration decision. The reasons for the internal labor migration within the USA have been a popular research topic in regional science (Plantinga et al. 2013; Lim 2011; Ashby 2007; Francis 2007; Partridge and Rickman 2006; Chi and Voss 2005; Horiba 2000; Saltz 1998; Carrington et al. 1996; Vedder et al. 1986; Gordon and Vickerman 1982; Renas 1978; Silvers 1977; Sjaastad 1962). Most of these studies focused on the determinants for internal migration in terms of the utility-maximizing behavior of potential migrants. However, empirical studies estimating interregional migration with regional differences in various economic conditions intrinsically have ecological fallacy problems. Heterogeneous socioeconomic characteristics among individual migrants either in an origin or a destination cannot be treated effectively; rather, there is a single representative migrant for a pair of an origin and a destination. With annual American Community Survey (ACS) PUMS data, individual socioeconomic characteristics can be controlled. As a result, we can effectively study what individual circumstances, rather than regional conditions, motivate internal migration. The purpose of this paper is to examine the net effects of individuals’ labor market status in the selected MSAs of the four Sand States with devastating labor market conditions on forced out-migration. The identified pushing factors from labor markets enable shedding some light on questions such as who moved facing recession and what triggered them to move. This paper is structured as follows. Section 2 reviews the existing literature on internal labor migration. In Sect. 3, this study develops multinomial logit models using ACS data for the study period (2007–2012). Section 4 synthesizes the analytical
123
Out-migration from the epicenters of the housing bubble…
results, followed by a discussion of the major findings in Sect. 5. The last section concludes with policy implications.
2 Literature review The empirical literature on internal migration in the USA is enormous and continues to grow.1 The interregional migration patterns are formulated in a gravity model in line with development in the empirical migration literature (Greenwood and Hunt 2003). Place-to-place migration studies use different modifications of the Harris– Todaro (1970) model to explain either net or gross migration (see, among others, Pissarides and McMaster 1990; Decressin 1994; Oswald 1990; Treyz et al. 1993; Vigdor 2002). It would be more desirable to model gross in- and out-migration rather than net migration due to the volatility of net migration and, more importantly, the fallacy of net migration rates (Plane and Rogerson 1994). Rogers (1990) has indeed demonstrated the violation of the demographer’s principle when using net migration rates, as the atrisk population for net migration used as a denominator is composed of the population in specific sets of origin and destination places included in the study. However, the relevant at-risk population for in-migration is not the population from specific sets of origin but all of those who are not in the destination under consideration (Plane and Rogerson 1994). However, it is relatively easy to identify the at-risk population for out-migration as this includes all of the population within the specific sets of origin. Use of modified gravity models with gross place-to-place migration flows as a dependent variable partly overcomes this problem. However, this approach still has defects mainly due to the lack of socioeconomic characteristic data at the individual level. An individual (either movers or non-movers) is represented by a location-specific characteristic. This can be addressed by employing annual ACS data, which survey socioeconomic characteristics at the individual level. Starting from the mid-1980s, the availability of longitudinal data on migration allowed the adoption of dynamic specifications of the gravity migration model (Molho 1984). ACS data replaced the long-form decennial census for internal migration studies (Franklin and Plane 2006) and are known to be a better source for time-series analysis with detailed information on individuals. Among others, Basile and Lim (2016) summarized the issues associated with the importance of temporal aspects in migration analysis as follows: (1) response lags to market signals, (2) life cycle effects and ‘state dependence’ of individuals, (3) cyclical perspectives of the overall volume of migration, (4) persistence in migration behavior and (5) uncertainty on temporally varying socioeconomic characteristics. Among these potential issues, this paper mainly focuses on the cyclical nature of migration and uncertainty on temporally varying socioeconomic characteristics. A temporal variation of socioeconomic conditions, 1 See, among other, Rogers (1973), Plane (1981), Rogerson and Plane (1984), Rogers and Little (1994), Rogers and Raymer (1998), Rogers et al. (2001, 2002, 2003), Rogers et al. (2003), Rogers and Raymer (2005), LeSage and Pace (2005), Little and Rogers (2007), Taylor et al. (2008), Wilson et al. (2009), Jeanty et al. (2010), Kancs (2011), Parrado and Kandel (2011), Zabel (2012), Partridge et al. (2012), Pendergrass (2013).
123
J. Lim
such as wage differentials and unemployment differentials, has been mainly studied at either national or regional levels. For cross-section gravity models, the implicit constant measures the aggregate volume of migration for a given period, whereas regional push and pull factors measure the deviations in in- and out-migration for each area (Molho 1984). With dynamic panel models, temporally varying the national economic climate makes the implicit constant adjust over time as the aggregate volume of migration varies with the business cycle. Furthermore, the extent of the impact from temporal variations in national business cycles will vary by region based on the regional economic structures. Empirical studies found that the volume of migration is likely to vary counter-cyclically (Hart 1975; Gordon 1985). Because liquidity constraints restrict human capital investment behavior, such as migration, in a recession period, uncertain prospects are discounted more heavily. In addition, employment opportunities are significantly reduced during a recession. However, a local labor market condition greatly lagging behind other regions during a recession may push out the at-risk population. In these cases, not only a decision on ‘where to move’ matters, but also the decisions on ‘whether to move’ and/or ‘when to move’ matter. Silvers (1977) suggested two different migration types in the perspective of the rational behavior of an individual maximizing income: speculative and contracted. The former includes migrants to the “promised land” even without job offers, easily attracted by expected wages in a destination, which is often overestimated by the migrants. Contracted migration includes conservative migrants who move only after confirming the higher wage from new job offers in a destination. For contracted migrants, the regional wage difference and the probability of obtaining a job in a destination do not really matter; rather, this matters more for speculative migrants. Speculative migration with greater uncertainty in destination regions leads migrants to “try their luck in other job markets” due to their belief that “anything can be better than the current situation of the origin regions.” Migration studies for the potential impact of the uncertainty associated with future events on migration decisions are somewhat limited. Burda (1993) shows that prospective migrants become less likely to relocate as the uncertainty about future conditions in the origin and destination regions increases—they “wait and see.” However, this “wait and see” aspect of the migration decision varies with the socioeconomic characteristics of the at-risk population. Molloy et al. (2011) speculated that a cyclical labor market downturn is not a main cause for continuously decreasing internal migration in the USA. Partridge et al. (2012) also found evidence of diminishing net migration responsiveness to regionally varying demand shocks in the post-2000 era. Potentially, increasing tendency toward “wait and see” among the at-risk population during the Great Recession was associated with their findings on the increased risk aversion in the post-2000 period. This study aims to analyze how individual labor market status among the at-risk population in the epicenters of the housing bubble burst affected the out-migration decision with destination choices.
3 Methodology The hypothesis of this study concerns whether the labor market status of an at-risk individual affects out-migration behavior. If so, specific questions include the follow-
123
Out-migration from the epicenters of the housing bubble…
ing: 1) How does the labor market status of an at-risk individual affect destination choices? 2) Does the identified effect temporally vary during and in the aftermath of the Great Recession? Among the 100 largest MSAs, 13 MSAs from four Sand States were selected that recorded an at least 1.5 times higher serious delinquency rate2 for March 2010, compared to the average serious delinquency rate among the 100 largest MSAs. In these 13 MSAs, the share of all mortgages either 90 or more days delinquent or in the foreclosure inventory for March 2010 is greater than 15.2%, whereas the average for the 100 largest MSAs is 10.2%. These 13 MSAs are classified as the epicenters of the housing bubble burst. Those who lived in one of the 13 MSAs 1 year prior to the data collection date of the ACS reference year were chosen as at-risk population to test the effects of deteriorating labor market conditions as a pushing factor for out-migrants. Most of the existing empirical studies on labor migration tend to analyze place-toplace migration based on comparative labor market conditions between origin and destination places. However, this study employs multinomial logit models to estimate the net effects of the labor market status in origin places on the out-migration decision by controlling individual socioeconomic characteristics. Unfavorable labor market status of the at-risk population during a recession is expected to serve as pushing factor, particularly for those marginally attached workers in the local labor market. Due to limited opportunities in same MSAs and/or same states, this group is also expected to migrate to other states chasing better employment opportunities. 3.1 Data ACS PUMS3 data are employed for three main reasons. First, personal socioeconomic characteristics are the key control variables for this paper. Second, analysis on the annual migration pattern is critical as this study focuses on temporally varying out-migration patterns over a 6-year period: pre-recession (2007–2008), during the recession (2008–2011) and in recovery (2011–2012). Third, the proposed models do not describe place-to-place migration; rather, they focus on out-migration patterns from the 13 selected MSAs. The selected 13 MSAs are listed in Table 1. The smallest MSA is Modesto in California with 514,453 residents in 2010, while the largest MSA is Miami–Fort Lauderdale–Pompano Beach in Florida with over 5.5 million residents. The minimum serious delinquency rate is 15.7% in Jacksonville, FL, which is 150% higher than the average among the largest 100 MSAs (15.2%). All 13 MSAs have a minimum 6.1% foreclosure rate and a minimum 16.0% subprime foreclosure rate. Because the PUMAs (Public Use Microdata Areas) are used for PUMS data, multiple PUMAs are combined to develop a dataset for each of the 13 MSAs. The at-risk 2 The serious delinquency rate is defined as the percent of all mortgages either 90 or more days delinquent or in the foreclosure inventory in the reference month. (Source: Analysis of LPS Applied Analytics data by the Local Initiatives Support Corporation (LISC); for more detailed information, visit http://www. foreclosure-response.org/assets/maps&data/Methodology_Metro_Delinquency_Rates.pdf). 3 The American Community Survey (ACS) Public Use Microdata Sample (PUMS) data have the full range of population and housing unit responses collected on individual ACS questionnaires. A detailed list can be found at the following: http://www.census.gov/acs/www/data_documentation/pums_documentation/.
123
123
Modesto, CA
1,375,765
Las Vegas–Paradise, NV
Largest 100 Metro Avg.
476,230
2,395,997
Tampa-St. Petersburg–Clearwater, FL
1,644,561
Palm Bay–Melbourne–Titusville, FL
Orlando–Kissimmee, FL
1,951,269
2,783,243
543,376
2,134,411
702,281
5,564,635
589,959
Miami–Fort Lauderdale–Pompano Beach, FL 5,007,564
North Port–Bradenton–Sarasota, FL
602,095
1,345,596
685,306
1,122,750
563,598
514,453 4,224,851
446,997
3,254,821
483,924
Lakeland–Winter Haven, FL
Jacksonville, FL
Stockton, CA
Riverside–San Bernardino–Ontario, CA
839,631
4,192,887
Population (2010)
661,645
3,252,822
Phoenix–Mesa–Scottsdale, AZ
Bakersfield, CA
Population (2000)
MSA
41.8
16.2
14.1
29.8
19.0
11.1
24.4
19.8
21.6
29.8
15.1
26.9
28.9
Population Change (2000–2010)
10.2
25.1
19.6
15.9
21.8
18.7
26.0
20.0
15.7
17.7
18.7
16.8
16.5
15.8
4.9
13.1
13.1
10.3
13.3
13.1
17.8
11.7
9.1
7.0
7.4
6.9
6.1
6.8
18.8
30.2
35.2
32.9
31.8
39.0
40.5
27.0
25.6
18.8
19.7
20.2
16.0
19.0
(%) Serious (%) Foreclosure Rate (%) Subprime (%) Delinquency Rate (March 2010) Foreclosure Rate (March 2010) (March 2010)
Table 1 13 MSAs in Sand States with highest serious delinquent rate (2010). Source: US Census Bureau, 2010 Census and Census 2000 for population data, and Analysis of LPS Applied Analytics data by the Local Initiatives Support Corporation (LISC) for delinquency and foreclosure data
J. Lim
Out-migration from the epicenters of the housing bubble…
population from the 13 MSAs is defined as the non-institutionalized civilian householder who is 16 years or older. The total sample size for the aggregated 6-year data is 582,042 cases with the annual sample ranging from 95,300 to 99,600. 3.2 Multinomial logistic regression model Proposed multinomial logit models estimate the effects of various socioeconomic variables on the out-migration decision. Among the socioeconomic variables under consideration (Table 2), an individual’s labor market status may vary greatly during a recession. As identified by White and Mueser (1994), temporally varying factors such as falling barriers to migrate can have a large impact on migration probabilities by affecting the structural relationship in migration estimation models. The key determinant of the migration decision in this study is the labor market status and its impacts on migration probabilities. However, the labor market condition is not measured as an aggregated average for an origin; rather, an individual’s status in a labor market is captured using personal data from PUMS ACS for 6 consecutive years during the study period. The 16 total variables fall into the following 3 categories: (1) demographic characteristics, (2) educational attainment and (3) economic & labor market status. For the demographic category, there are 5 variables: Age, Gender, Living with Own Child, Hispanic and Black. For the educational attainment category, the variables measure the 4 levels of education attainment with the reference category of the highest attainment level (graduate degree or higher). For the economic category, 7 variables are included: Annual Personal Income, Below Poverty Level, Part-Time Worker, Unemployed, NILF (currently or in last 5 years), Construction Employee (anytime within last 5 years) and Professional and Business Service Employee (anytime within last 5 years). Two variables—Age and Annual Personal Income—are continuous, while all of the others are discrete choice variables. Out-migration is modeled as one of the four mutually exclusive discrete choices ( j = 1, 2, 3, 4) for the current year’s location compared to the location in the previous year: (1) staying in the same house (stayers or non-movers), (2) moving to another house within the same MSA (intra-MSA out-migration), (3) moving to another region outside an MSA but within the same state (intrastate out-migration) or (4) moving to another state (interstate out-migration). The multinomial logit model is defined as shown in Eq. (1). exp xit β tj t (1) Pr yi = j = 4 t βt exp x j=1 i j where Pr(yit = j) is the probability of choosing migration type j of an individual i at year t i = 1, . . ., N(N = sample size) t = 2007, 2008, . . ., 2012 j = 1(Non-mover/stayer)
123
123
A respondent lives with own child
A respondent is Hispanic
A respondent is an African-American
Living with Own Child
Hispanic
African-American
Some College, but No Degree
College Degree
College Degree or Higher
HS Grad
Some College, No Degree
College Degree
Graduate Degree or Higher
Experience in Professional & Business Service in last 5 years
PBS Worker (5 years)
DD
DD
DD
DD
DD
DD
C
DM
DM
DM
DM
DM
DD
DD
DD
DD
C
Type*
* Variables belong to one the following three types: continuous (C), discrete dummy (DD), discrete multiple choice (DM)
Not in labor force currently or in last 5 years
Unemployed
Experience in Construction in last 5 years
A respondent is unemployed.
Part-time Worker
Construction Worker (5 years)
A respondent is working less than 35 hours per week.
Below Poverty Level
Not in Labor Force (5 years)
Annual personal income (in $1000)
A respondent is below poverty level.
Annual Personal Income
Economic & Labor Status
Less than High School Graduate
High School Graduate
Less than HS Grad
Education attainment
Age of a respondent (in years)
Sex of a respondent
Gender
Definition
Age
Demographic
Variable
Table 2 Socioeconomic variables for out-migration decision
No experience in PBS
No experience in construction
Currently in labor force
Not Unemployed
Not Part-time Worker
Not below poverty level
Graduate Degree of Higher
Graduate Degree of Higher
Graduate Degree of Higher
Graduate Degree of Higher
Graduate Degree of Higher
Non-African-American
Non-Hispanic
Not living w/ own child
Female
Reference Category
J. Lim
Out-migration from the epicenters of the housing bubble…
2(Intra-MSA Out-migration) 3(Intrastate Out-migration) 4(Interstate Out-migration) xit is a vector of the socioeconomic variables of an individual i at year t. β tj is a vector of the coefficients for migration type j at year t, using maximum likelihood estimation. Odds ratios (ORs) are calculated as the exponential of the estimated coefficients for different migration types j at year t and used for the comparative probabilities of different types of out-migration against the reference category (no-move), given a set of socioeconomic variables. The odds ratio can be used to determine whether a particular individual socioeconomic characteristic encourages or discourages a specific type of out-migration against no-move and to compare the magnitude of various socioeconomic factors for specific out-migration decisions. • OR = 1 (or β = 0): A variable under consideration does not affect out-migration probabilities. • OR > 1 (or β > 0): A variable under consideration increases out-migration probabilities. • OR < 1 (or β < 0): A variable under consideration decreases out-migration probabilities. For the 6-year study period (2007–2012), the odds ratios for labor market status variables, ‘Part-time Worker’, ‘Unemployed’ and ‘NILF’ (not in labor force), are compared to analyze how a labor market status affects the out-migration decision among the atrisk population in the 13 MSAs. Two scenarios are developed to test how the same set of labor market statuses of two individuals affects their interstate out-migration decision differently when these two individuals have dissimilar demographic and education attainment characteristics. To analyze the responsiveness of an individual regarding the out-migration decision, it is important to compare how the comparative probabilities vary during a recession.
4 Analysis results This section summarizes the descriptive statistics of individuals under consideration, the model estimation results, temporal variation in probabilities and scenario analysis results. 4.1 Regional economic condition during the Great Recession Devastating local housing markets during the Great Recession brought a negative shock to the labor markets of these 13 MSAs. Between 2007 and 2012, the labor market condition captured by the total of 582,042 individual samples (householders 16 years and older) from the annual ACS clearly shows how local labor markets had deteriorated (Fig. 1). The number of part-time workers steadily increased, while the unemployed and NILF population increased more rapidly. In addition, that of those
123
J. Lim
Fig. 1 Economic conditions in 13 MSAs (2007–2012)
working in the construction sector dropped significantly. Consequently, the share of the population below the poverty line increased significantly, particularly between 2010 and 2011. However, even during the Great Recession, the share of the Hispanic population steadily increased. These trends in the labor market may have acted as pushing factors for those vulnerable populations out of these MSAs. Proposed multinomial logistic models estimate both annual and aggregated effects of various socioeconomic variables on out-migration probabilities of the residents within the 13 MSAs who are householders aged 16 years and older. The mean values for all socioeconomic variables are in Table 3. 4.2 Estimation results The goodness of fit with likelihood ratio test results of the multinomial logit model is summarized. The proposed model with the likelihood ratio Chi-square of 90,087.765 indicates that this model as a whole fits significantly better than a model with no predictors (Table 4). The likelihood ratio test results for all 13 variables show that the effects of all of the variables are statistically significant (Table 5). The parameter estimation result for the aggregated sample over the 6-year study period (2007–2012) is given in Table 6. The odds ratio represents the effect of a variable on the comparative probabilities of out-migration against staying, holding other variables constant. Among the demographic variables, AGE discouraged out-migration for all types of moves, with the largest effect on intrastate out-migration. With one more year of age, the probabilities of out-migration against no-move for ‘Intra-MSA’ and ‘Interstate’
123
45.9%
87.9%
12.1%
Female
Living with Own Child
Not Living with Own Child
18.6%
Some College, No Degree
College Degree
84.6%
11.9%
88.1%
9.4%
90.6%
25.4%
Not Below Poverty Level
Part-time Worker
Not Part-time Worker
Unemployed
Not Unemployed
Not in Labor Force (5 years)
11.1%
33.0%
HS Grad
15.4%
24.3%
Less than HS Grad
Below Poverty Level
13.0%
Not African-American
Graduate Degree or Higher
10.9%
89.1%
African-American
18.9%
54.1%
Male
81.1%
47,200
Avg. Annual Personal Income
Not Hispanic
52.7
Avg. Age
Hispanic
Aggregated
Variable
23.4%
92.1%
7.9%
89.5%
10.5%
87.6%
12.4%
11.0%
18.9%
31.7%
25.0%
13.4%
90.1%
9.9%
81.9%
18.1%
12.4%
87.6%
44.9%
55.1%
52,000
51.7
2007
Table 3 Mean values & share of householder for socioeconomic variables
23.9%
92.0%
8.0%
88.7%
11.3%
86.9%
13.1%
11.1%
19.0%
33.3%
23.9%
12.8%
89.9%
10.1%
81.6%
18.4%
12.5%
87.5%
45.4%
54.6%
50,800
52.4
2008
24.8%
91.3%
8.7%
87.7%
12.3%
85.9%
14.1%
11.3%
18.8%
33.3%
23.7%
12.8%
89.4%
10.6%
81.5%
18.5%
12.2%
87.8%
45.5%
54.5%
48,000
52.8
2009
25.5%
89.9%
10.1%
87.6%
12.4%
84.7%
15.3%
11.1%
18.9%
33.4%
24.0%
12.6%
89.1%
10.9%
80.9%
19.1%
12.3%
87.7%
46.4%
53.6%
46,000
53.0
2010
27.2%
89.3%
10.7%
87.4%
12.6%
80.8%
19.2%
10.8%
17.6%
33.1%
24.8%
13.6%
87.9%
12.1%
80.6%
19.4%
11.7%
88.3%
46.4%
53.6%
43,000
53.1
2011
27.5%
89.2%
10.8%
87.7%
12.3%
81.7%
18.3%
11.4%
18.5%
33.2%
24.4%
12.6%
88.3%
11.7%
80.4%
19.6%
11.6%
88.4%
46.7%
53.3%
43,800
53.5
2012
Out-migration from the epicenters of the housing bubble…
123
123
9.5%
582,042
N
94.3%
Not CONST Employee (5 years)
90.5%
5.7%
Construction Employee (5 years)
Not PBS Employee (5 years)
74.6%
In Labor Force (5 years)
PBS Employee (5 years)
Aggregated
Variable
Table 3 continued
96,390
90.6%
9.4%
93.2%
6.8%
76.6%
2007
95,360
90.6%
9.4%
93.7%
6.3%
76.1%
2008
95,328
90.2%
9.8%
94.1%
5.9%
75.2%
2009
95,841
90.4%
9.6%
94.5%
5.5%
74.5%
2010
99,427
90.7%
9.3%
95.1%
4.9%
72.8%
2011
99,696
90.5%
9.5%
95.3%
4.7%
72.5%
2012
J. Lim
Out-migration from the epicenters of the housing bubble… Table 4 Model fitting information for aggregate sample (2007–2012) Model
Model fitting criteria
Likelihood ratio tests
−2 Log likelihood
Chi-square
df
Sig.
90,087.765
48
0.000
Intercept Only
594,726.669
Final
504,638.904
Table 5 Likelihood ratio test results for aggregate sample (2007–2012) Effect
Model fitting criteria
Likelihood ratio tests
−2 Log likelihood of reduced model
Chi-squarea
Intercept
504,638.904b
Age
543,046.443
Male Own child
df
Sig.
0.000
0
38, 407.539
3
0.000
504,986.913
348.009
3
0.000
507,496.762
2, 857.857
3
0.000
Hispanic
505,366.035
727.131
3
0.000
Black
505,126.165
487.261
3
0.000
Education attainment
505,345.003
706.099
12
0.000
P. income (in $1000)
505,216.985
578.081
3
0.000
BLPV
515,193.516
10, 554.612
3
0.000
PT worker
504,822.773
183.869
3
0.000
Unemployed
505,390.826
751.921
3
0.000
NIFL
505,525.303
886.399
3
0.000
Construction employee
504,715.115
76.211
3
0.000
PBS employee
504,690.022
51.118
3
0.000
a The Chi-square statistic is the difference in −2 log likelihoods between the final model and a reduced
model. The reduced model is formed by omitting an effect from the final model. The null hypothesis is that all parameters of that effect are 0 b This reduced model is equivalent to the final model because omitting the effect does not increase the degree of freedom
moves decreased by 5.1 and 5.7%, respectively. The matching odds ratio (discouraging effect of age) for ‘Intrastate’ out-migration was highest at 7.3%. The Gender variable had mixed results. While male householders were 11.7% less likely to experience residential relocation (move within a same MSA) compared to female householders, males were 39.0% more likely to move to another region within the same state (intrastate outmigration). A householder living with Own Child was less likely to have all types of out-migration. The discouraging effect was the largest for intrastate migration (82.0% less likely than those without own child), followed by interstate out-migration (61.4% less likely to move) and finally by residential relocation (36.3% less likely to move). For all destination choices, Hispanic was less likely to out-migrate, contrary to the expectation that Hispanics are more vulnerable to labor market downturns during a
123
123 0.580 0.000
0.170*
0.010*
−0.003*
Some College vs. Grad. Degree
College Grad. vs. Grad. Degree
0.140
0.197*
−0.026*
NILF (currently or in last 5 years)
PBS Employee (last 5 years)
0.046*
0.000
0.020*
Unemployed
Construction employee (last 5 years)
0.001
−0.044*
PT worker
0.001
0.230
0.000
0.000
0.878*
BLPV
P. Income (in $1000)
0.206*
0.000
0.000
0.295*
LTHS vs. Grad. Degree
HS Grad vs. Grad. Degree
0.000
−0.071*
Hispanic 0.000
0.000
−0.451*
Own child
0.214*
0.000
−0.124*
Male
Black
0.000 0.000
1.097*
−0.052*
Intercept
1.047
0.974
1.218
1.020
0.957
2.406
0.997
1.010
1.186
1.229
1.343
1.239
0.932
0.637
0.883
0.949
1.487*
0.000
0.057*
0.148*
1.052*
1.032*
0.473*
2.016*
0.170
0.001
0.000
0.000
0.000
0.000
0.000
0.000
−0.373* −0.003*
0.222
0.411
0.071*
0.229
0.075*
0.006
0.000
0.000
0.000
0.000
−0.049*
0.087*
−0.285*
−1.713*
0.329*
−0.076*
Sig
1.059
1.160
2.862
2.807
1.605
7.508
0.997
0.689
1.073
0.952
1.078
1.090
0.752
0.180
1.390
0.927
Odds ratio
B
Odds ratio
B
Sig
To same state, but other parts (intrastate)
To Same MSA (Intra-MSA)
Age
Types of migrationa (j)
Table 6 Multinomial logit model estimate results for aggregate sample (2007–2012)
0.000 0.000
−0.313*
0.000
0.000
0.052
0.000
0.000
−0.185*
0.300*
0.222*
0.056*
1.414*
−0.001*
0.000
0.000
−0.241* −0.258*
0.000
0.000
−0.536* −0.299*
0.000 0.000
−0.655*
0.000
0.348
−0.331*
−0.951*
0.020*
0.000
0.000
−1.061* −0.059*
Sig.
B
0.831
0.731
1.350
1.248
1.057
4.114
0.999
0.773
0.786
0.742
0.585
0.718
0.519
0.386
1.020
0.943
Odds ratio
To other states (interstate)
J. Lim
Out-migration from the epicenters of the housing bubble…
recession, consequently being more likely to out-migrate in a speculative manner. This can be partly explained by the lack of resources to out-migrate during the Great Recession as well as by the concentration of the undocumented Hispanic population in these states. The discouraging effect for Hispanics was the highest at 48.1% for out-migration to other states relative to the non-Hispanic population, followed by intrastate at 24.8% and by residential relocation at 6.8%. For African-Americans, the effect on out-migration was mixed. African-Americans were 23.9% more likely to experience residential relocation within the same MSAs and 9.0% more likely to experience intrastate migration. However, the odds ratio on interstate out-migration was 0.718, indicating that they were 28.2% less likely to move to another state. Again, a lack of necessary resources for relatively longer-distance moves can partly explain the odds ratio being lower than 1 for African-Americans, whereas the residential relocation caused by foreclosure may have increased out-migration within same MSAs and/or same states. Regarding education attainment variables, those with lower attainment levels compared to a group with the highest education attainment (graduate degree or higher) had distinctive patterns for out-migration probabilities. The lower the education attainment, the higher was the residential relocation probability, while the lower the education attainment, the lower was the interstate out-migration probability. This is in line with the previous finding on residential displacement and the lack of resources for longerdistance moves among African-American and Hispanic householders. Finally, our attention is on a set of economic conditions of a householder for the out-migration decision. A $1000 higher annual income slightly decreased the probability to out-migrate for all types of out-migration, but the effect was very small. A householder below the poverty line was more likely to out-migrate in the following order: intrastate out-migration, interstate out-migration and residential relocation (intra-MSA). For instance, the odds ratio for an intrastate move was 7.508, followed by 4.114 for an interstate move and 2.406 for an intra-MSA move. With the aggregated data, part-time workers (working less than 35 hours per week) were 4.3% less likely to experience residential relocation, while they were 60.5% more likely to out-migrate to other regions within the same states. Part-time workers preferred shorter-distance moves within the same state to interstate moves. The unemployed tended to have a higher probability of moving (intrastate and interstate) in contrast to ‘not unemployed’ including employed and NILF. For intrastate moves, the unemployed were 2.807 times more likely to move, whereas the unemployed were 1.248 times more likely to move to other states. Employment status was not a significant factor for residential relocation. Those NILF are more likely to out-migrate to all destinations. The highest odds ratio was also found with intrastate, followed by interstate and residential relocation. All of the odds ratios for the NILF variable were statistically significant. Work experience in the construction sector within the past 5 years stimulated intrastate out-migration by 16.0%, whereas it curbed interstate out-migration by 26.9%. However, work experience in the professional and business service sector within the past 5 years also discouraged interstate moves by 16.9%, but it had no effect on intrastate moves. The estimation results for six individual years can be found in the Appendix of Electronic Supplementary Material.
123
J. Lim
Odds Ra o of Inter- & Intra-state Out-Migra on (vs. No-Move) 3.50
PT Worker (Intra-State)
Unemployed (Intra-State)
NILF (Intra-State)
PT Worker (Inter-State)
Unemployed (Inter-State)
NILF (Inter-State)
3.00
2.50
2.00
1.50
1.00
0.50
0.00
2007
2008
2009
2010
2011
2012
Fig. 2 Effects of employment status on inter- & intrastate out-migration
4.3 Temporal variation in the effects of labor market status (2007–2012) Since the beginning of the Great Recession in 2008, the deteriorating local labor market condition in the 13 MSAs definitely weakened their attractiveness as migration destinations. More importantly, it potentially forced out some of the at-risk population that was more vulnerable to negative economic shock in these MSAs. Figure 2 compares the temporal variation in the odds ratios between two types of migration: interstate and intrastate. In all cases, the reference category is ‘No-Move’. The probability for intrastate out-migration was larger than interstate moves. In particular, the intrastate out-migration probability for the unemployed had been highest before to during the start of the recession (2007–2009), while the matching probability for NILF had been highest during the recovery (2011–2012). An unemployed person was at least 2.4 times more likely to move within the same state, with its peak being 3.3 times higher in 2008. Although this probability had continuously declined from 2008 to 2012, it was still much higher than the matching probabilities for interstate out-migration of the unemployed. The odds ratios of the unemployed for intra-MSA residential relocation were insignificant. The matching odds ratios for interstate outmigration also decreased over time but at a much slower pace, from 1.458 (in 2008) to 1.041 (in 2012). The unemployed had much higher mobility compared to ‘not unemployed’ including employed and NILF, and this was much more evident for relatively shorter-distance moves within the same states. In the midst of the Great Recession, the effect of unemployed status on out-migration had slightly decreased. In the meantime, the probability of intrastate out-migration continued to increase for the NILF group between 2007 and 2011. The matching probability of NIFL for
123
Out-migration from the epicenters of the housing bubble…
interstate out-migrants also increased between 2007 and 2010 but started to decrease from 2010. The labor market conditions in the 13 MSAs started to improve from 2010, and higher moving costs over longer-distance interstate migration might have made the NILF population postpone out-migration and possibly put more weight on staying in their current locations. The odds ratio of NILF for intrastate out-migration started to decline in 2011. Part-time workers’ probability of moving out of the 13 MSAs was also higher for intrastate moves than interstate moves. In fact, the part-time worker effect was not significant for most of the study years for interstate out-migration. The odds ratio for part-time workers on intrastate moves had steadily increased to its peak at 1.712 in 2011, indicating that part-time workers were 71.2% more likely to move to other parts of the same state than ‘non-part-time workers’. For part-time workers, intrastate out-migration might have been viewed as an affordable option with relatively lower cost in seeking full-time employment status.
4.4 Scenario analysis for interstate out-migration probability Two individuals with contrasting socioeconomic conditions are assumed for simulation scenarios (Table 7). For each case, out-migration probabilities are estimated with varying labor market status variables: NIFL, unemployed, part-time worker and full-time worker. An individual, named B (case 2) has relatively higher socioeconomic status compared to individual A (case 1). Individual B is expected to move as a contracted migrant, while individual A is prone to be a speculative migrant. Individual A has a higher probability of moving to another state compared to individual B in most cases (Figs. 3, 4). For both cases, the employed (either part-time or full-time) tend to have higher probabilities for interstate out-migration in the midst of a recession. This is in line with the contracted migration discussion by Silvers (1977). Prior to the recession, either NILF or the full-time employed had the highest outmigration probability, as those NILF had no strings attached and the full-time employed might have job offers and moved as contracted out-migrants chasing better opportunities. The probabilities have generally decreased since 2008, but there have been unexpected hikes. For individual A with work experience in the construction sector, the hike
Table 7 Two cases with varying socioeconomic characteristics Socioeconomic variable
Case 1
Case 2
Name
A
B
Age
28 years old
45 years old
Income
$ 30,000
$ 85,000
Gender
Male
Male
Living with Own Child
No
Yes
Race
Hispanic
Non-Hispanic
Education Attainment
Less than High School
College Degree
Below Poverty Level
No
No
123
J. Lim
Construcon Workers' Probability to Outmigrate to Different State 6.00% 5.50% 5.00% 4.50% 4.00% 3.50% 3.00% 2.50% 2.00% 1.50% 1.00%
2007
2008
2009
2010
2011
2012
NILF (case 1)
4.57%
4.92%
2.79%
2.89%
2.88%
2.59%
Unemployed (case 1)
3.74%
4.48%
3.07%
3.34%
2.66%
2.69%
Part-Time Worker (case 1)
4.25%
4.77%
3.50%
3.58%
3.25%
2.55%
Full-Time Worker (case 1)
4.44%
5.40%
3.11%
3.82%
3.06%
2.48%
Fig. 3 Construction worker’s probability to out-migrate to other states (Case 1: A)
Construcon Workers' Probability to Outmigrate to Different State 6.00% 5.50% 5.00% 4.50% 4.00% 3.50% 3.00% 2.50% 2.00% 1.50% 1.00%
2007
2008
2009
2010
2011
2012
NILF (case 2)
3.74%
4.01%
2.41%
2.41%
3.10%
2.16%
Unemployed (case 2)
3.15%
3.70%
2.73%
2.90%
2.98%
2.32%
Part-Time Worker (case 2)
3.66%
4.12%
3.19%
3.11%
3.67%
2.22%
Full-Time Worker (case 2)
3.82%
4.61%
2.77%
3.34%
3.47%
2.15%
Fig. 4 Construction Worker’s probability to out-migrate to other states (Case 2: B)
was noticeable during 2010, and this might be a response to the expanded job opportunity from shovel-ready projects under the ARRA (American Recovery & Reinvestment Act). In particular, the hike is most evident when individual A is in full-time employ-
123
Out-migration from the epicenters of the housing bubble…
ment status, and the second largest hike is observed when individual A is unemployed. Similar responses are also found from individual B with lagged hikes in 2011. Interestingly, when individual B is a full-time employee, the response to expanded opportunity elsewhere was quicker, as found in the large hike from 2009 to 2010. However, if individual B is either unemployed or NILF, he/she responded to the ARRA stimulus more sensitively in 2011. The biggest hike for individual B is when he/she is not in the labor force, potentially returning to the labor force to seek employment opportunities. The lowest probability to out-migrate was found when one was unemployed for both cases in the pre-recession period, whereas the lowest probability was found when one had NILF status in the midst of the recession (2009–2010). Interestingly, the lowest probability during recovery was found when one was a full-time employee, which proves that the stabilized local labor markets were retaining them in the 13 MSAs.
5 Discussion During the Great Recession in the USA, an unfavorable labor market status of an at-risk individual served as a pushing factor for out-migration from the epicenters of the housing bubble burst. However, after controlling for personal socioeconomic variables, the effects of labor market status were largely limited and varied by the following factors: (1) types of move based on destination choices, (2) labor status variables under consideration and (3) time period during and in the aftermath of the Great Recession. The largest effect of labor market status was found in intrastate out-migration. The effects of unemployed are the highest for intrastate out-migration during the pre-recession period (2007–2008) and early recession period (2008–2009). Intrastate migration of the unemployed reached a peak in 2008 but continuously declined afterward, reaching its lowest point in 2012. This is in line with speculative migration by Silvers (1977) since those who moved within the past 12 months but were still not employed can be classified as speculative migrants. As indicated by Burda (1993), speculative migration declines with the growing ‘wait and see’ attitude during the recession. In addition, the effects of NILF for intrastate out-migration were highest during the late recession and recovery period (2010–2012). The at-risk population classified as NILF was more willing to take risks in the “promised land” for job searching while minimizing relocation costs by moving within the same states. The probability of part-time workers for intrastate moves was weakest among the three labor status variables. These potentially speculative migrants during and in the aftermath of the Great Recession are much smaller in size compared to the other at-risk population. The employed at-risk population tended to be risk-aversive and reluctant to move during the Great Recession due to largely limited job opportunities for contracted migrants. The matching interstate out-migration pattern for the unemployed was quite similar, but the probability slightly increased from 2010 to 2011. Both intra- and interstate outmigrations show that speculative migration of the unemployed was larger during the early recession, but migration had decreased due to the growing ‘wait and see’ attitude in the midst of the recession and even further during the recovery. This confirms the findings on the ‘option value of waiting’ for the migration decision by Basile and
123
J. Lim
Lim (2016). The strongest speculative out-migration pattern was found among those below the poverty line for both intra- and interstate moves. Temporal fluctuation on the effects of labor market status might have been induced by the federal stimulus package under the ARRA for shovel-ready projects. Interestingly, those with lower socioeconomic status responded more sensitively to stimulus packages in other states, resulting in a higher probability of experiencing interstate migration. Due to the federal investment, interstate out-migration had temporarily increased. This is consistent with the findings by Partridge et al. (2012) as the responsiveness to the demand shock in the labor market was limited to certain groups of the population, and the ‘wait and see’ of the overall at-risk population had increased due to risk aversion with growing uncertainty during the recession.
6 Conclusion A high level of human capital encouraged mobility among UK university graduates, particularly repeated migration for better employment opportunities (Faggian et al. 2006). Better-educated German workers were also less reluctant to move over longer distance due to the risk-loving behavior with higher willingness to cross cultural boundaries (Bauernschuster et al., 2014). However, during and in the aftermath of the Great Recession, growing uncertainty in the US labor market significantly curbed contracted interregional migration among highly skilled workers. Migration costs for highly skilled workers tend to be higher mainly due to the higher social costs of migration. Higher migration costs for this group contribute to lower mobility facing growing uncertainty. Consequently, even with temporarily unfavorable labor market status (either unemployed or underemployed), highly skilled workers maintain a ‘wait and see’ attitude. By delaying migration decisions, they may avoid the cost of uncertainty with the expected recovery in current locations. However, speculative migrants with relatively lower socioeconomic status tend to move more frequently chasing better opportunities elsewhere even with growing uncertainty. This is evident for those with lower education attainment, living below the poverty line or being African-American in the proposed models. These marginally attached workers in local labor markets respond more sensitively to economic shocks than those with higher socioeconomic status. The expected return on their speculative migration over longer distances (e.g., interstate migration) hardly covers migration costs due to growing uncertainty during the Great Recession. The identified out-migration patterns from the epicenters of the housing bubble burst provide valuable implications for state/local economic development policymakers. First, state and local governments need to make further efforts to retain temporarily unemployed or underemployed highly skilled workers in local regions. Helping them in their local job search activities and connecting local employers to the locally available labor supply can contribute to retaining the workforce in the long run. This would prevent a potential supply shortage of highly skilled workers after the recovery. Second, local governments need to further expand workforce-training opportunities for marginally attached workers in partnership with state and federal programs. Most marginally attached workers, unemployed and/or underemployed during the reces-
123
Out-migration from the epicenters of the housing bubble…
sion are likely to relocate to other regions in the same states. They may easily return to the same metro areas for expanded employment opportunities with recovery. State and local governments’ various workforce-training programs can enhance the human capital for this group and in turn equip them with skills required for employment opportunities after the recession. A stable workforce supply is key for sustainable regional economic growth. If a devastating local labor market is unavoidable during the recession and the local workforce is stuck in such a situation due to limited mobility, state and local governments should utilize the economic recession as an opportunity to enhance the job readiness of the local workforce by retaining and training them. One of the key determinants for the migration decision is home ownership status. Unfortunately, this study could not control housing factors due to the data unavailability at a personal level in the ACS PUMS dataset. The at-risk population’s tenure choice in local housing markets will open a whole new chapter for understanding migration behavior. While those with unfavorable labor market status in the local labor market are the main victims of forced out-migration, individuals’ situation in local housing markets may retain them locally. This research can be further expanded to analyze the varying magnitudes of labor market status effects on migration by employing the subsets of the population with similar socioeconomic characteristics.
References Ashby NJ (2007) Economic freedom and migration flows between U.S. states. South Econ J 73(3):677–697 Basile R, Lim J (2016) Nonlinearities in interregional migration behavior evidence from the United States. Int Reg Sci Rev. doi:10.1177/0160017615626986 Brown RA (2009) The 2009 economic landscape: how the recession is unfolding across four US regions. FDIC Q 3(1):26–38 Burda MC (1993) The determinants of East-West German migration. Eur Econ Rev 37:452–461 Carrington W, Detragiache E, Vishwanath T (1996) Migration and endogenous moving costs. Am Econ Rev 86(4):909–930 Chi G, Voss PR (2005) Migration decision-making: a hierarchical regression approach. J Reg Anal Policy 35(2):11–22 Decressin JW (1994) Internal migration in West Germany and implications for East-West salary convergence. Weltwirtschaftliches Archiv 130:231–257 Faggian A, McCann P, Sheppard S (2006) An analysis of ethnic differences in UK graduate migration behaviour. Ann Reg Sci 40(2):461–471 Francis J (2007) Asymmetries in regional labor markets, migration, and economic geography. Ann Reg Sci 41(1):124–144 Franklin RS, Plane DA (2006) Pandora’s box: the potential and peril of migration data from American community survey. Int Reg Sci Rev 29(3):231–246 Gordon I (1985) The cyclical interaction between regional migration, employment and unemployment: a time-series analysis for Scotland. Scott J Polit Econ 32:135–158 Gordon I, Vickerman R (1982) Opportunity, preference and constraint: an approach to the analysis of metropolitan migration. Urban Stud 19:247–261 Greenwood MJ, Hunt G (2003) The early history of migration research. Int Reg Sci Rev 26(1):3–37 Harris J, Todaro M (1970) Migration, unemployment, and development: a two-sector analysis. Am Econ Rev 60:126–142 Hart R (1975) Inter-regional migration: some theoretical consideration (part 2). J Reg Sci 15(3):289–305 Horiba Y (2000) U.S. interregional migration and trade. Working paper, Tulane University Jeanty PW, Partridge M, Irwin E (2010) Estimation of a spatial simultaneous equation model of population migration and housing price dynamics. Reg Sci Urban Econ 40(5):343–352 Kancs DA (2011) The economic geography of labour migration: competition, competitiveness and development. Appl Geogr 31(1):191–200
123
J. Lim LeSage J, Pace RK (2005) Spatial econometric modeling of origin-destination flows, mimeo, University of Toledo Lim J (2011) Does wage differential driven migration continue to exist? Tests on the role of regional economic structure in wage differential driven migration. Ann Reg Sci 47(1):213–233 Little JS, Rogers A (2007) What can the age composition of the population tell us about the age composition of its out-migrants? Popul Space Place 13:23–39 Molho I (1984) A dynamic model of inter-regional migration flows in Great Britain. J Reg Sci 24:317–337 Molloy R, Smith CL, Wozniak A (2011) Internal migration in the United States. J Econ Perspect 25(3):173– 196 Oswald AJ (1990) The housing market and Europe’s unemployment: a non-technical paper, mimeo University of Warwick Parrado EA, Kandel W (2011) Industrial change, Hispanic immigration, and the internal migration of low-skilled native male workers in the United States, 1995–2000. Soc Sci Res 40(2):626–640 Partridge MD, Rickman DS (2006) An SVAR model of fluctuations in U.S. migration flows and state labor markets. South Econ J 72(4):958–980 Partridge MD, Rickman DS, Olfert MR, Ali K (2012) Dwindling US internal migration: evidence of spatial equilibrium or structural shifts in local labor markets? Reg Sci Urban Econ 42(1):375–388 Pendergrass S (2013) Perceptions of race and region in the black reverse migration to the south. Du Bois Rev: Soc Sci Res Race 10(01):155–178 Pissarides CA, McMaster I (1990) Regional migration, wages and unemployment: empirical evidence and implications for policy. Oxf Econ Pap 42:812–831 Plane DA (1981) Estimation of place-to-place migration flows from net migration totals: a minimum information approach. Int Reg Sci Rev 6(1):33–51 Plane DA, Rogerson PA (1994) The geographical analysis of population: with applications to planning and business. Wiley, New York Plantinga AJ, Detang-Dessendre C, Hunt GL, Piguet V (2013) Housing prices and inter-urban migration. Reg Sci Urban Econ 43(2):296–306 Renas SM (1978) The cost of living, labor market opportunities, and the migration decision. Ann Reg Sci 15(1):95–104 Rogers A (1973) Estimating internal migration from incomplete data using model multiregional life tables. Demography 10(2):277–287 Rogers A (1990) Requiem for the net migrant. Geogr Anal 22:283–300 Rogers A, Little JS (1994) Parameterizing age patterns of demographic rates with the multiexponential model schedule. Math Popul Stud 4(3):175–195 Rogers A, Raymer J (1998) The spatial focus of US interstate migration flows. Int J Popul Geogr 4:63–80 Rogers A, Willekens FJ, Raymer J (2001) Modeling interregional migration flows: continuity and change. Math Popul Stud 9:231–263 Rogers A, Raymer J (2005) Origin dependence, secondary migration, and the indirect estimation of migration flows from population stocks. J Popul Res 22(1):1–19 Rogers A, Raymer J, Newbold KB (2003a) Reconciling and translating migration data collected over time intervals of differing widths. Ann Reg Sci 37(4):581–601 Rogers A, Willekens FJ, Raymer J (2003b) Imposing age and spatial structures on inadequate migration-flow datasets. Prof. Geogr. 55(1):56–69 Rogers A, Willekens FJ, Raymer J (2002) Capturing the age and spatial structures of migration. Environ Plan A 34:341–359 Rogerson PA, Plane DA (1984) Modeling temporal change in flow matrices. Pap Reg Sci 54:147–164 Saltz IS (1998) State income tax policy and geographic labour force mobility in the United States. Appl Econom Lett 5(10):599–601 Silvers AL (1977) Probabilistic income-maximizing behavior in regional migration. Int Reg Sci Rev 2:19–40 Sjaastad L (1962) The costs and returns of human migration. J Polit Econ 70(2):80–93 Taylor P, Morin R, Cohn D, Wang W (2008) American mobility–who moves? Who stays put? Where’s home?. Pew Research Center, Washington, DC Treyz GI, Rickman DS, Hunt GL, Greenwood MJ (1993) The dynamics of US internal migration. Rev Econ Stat 75(2):209–214 Vedder RK, Gallaway LE, Graves PE, Sexton R (1986) An historical perspective ininterregional migration in the US. In: Krumm RJ (ed) Housing and migration. Mount Pleasant, Blackstone, pp 101–124
123
Out-migration from the epicenters of the housing bubble… Vigdor JL (2002) The pursuit of opportunity: explaining selective black migration. J Urban Econ 51(3):391– 417 White MJ, Mueser PR (1994) Changes in the demographic determinants of US population mobility: 1940– 80. Rev Reg Stud 24(3):246–264 Wilson BA, Berry EH, Toney MB, Kim YT, Cromartie JB (2009) A panel based analysis of the effects of race/ethnicity and other individual level characteristics at leaving on returning. Popul Res Policy Rev 28(4):405–428 Zabel JE (2012) Migration, housing market, and labor market responses to employment shocks. J Urban Econ 72(2):267–284
123