# The Home Mortgage Interest Deduction and Migratory > Briefs, arguments, decisions, and more. URL: https://www.frixlaw.com/law-library/documents/agency%3Airs%3A33c5f43e56522367 ## Record - **Collection:** Agency decision - **Document type:** Agency decision ## Text The Home Mortgage Interest Deduction and Migratory Insurance over the Great Recession∗ Danny Yagan Abstract The home mortgage interest deduction (HMID) encourages homeownership and larger mortgages, which may impede migration when house prices fall. This paper investigates the degree to which the HMID reduced workers’ insurance against local variation in the employment effects of the Great Recession via impeded migration to strong local labor markets. Utilizing variation in the HMID at state borders and comparing similar individuals across space, I find that individuals lacked insurance against enduring employment effects of Great Recession local shocks, but I do not find significant evidence that state HMIDs hindered that insurance by impeding migration. I therefore do not find evidence in this context that “dynamic” distortions of the HMID via impeded migration magnified any of its “static” distortions to economic activity. However, estimates are uncertain and leave room for future work. 1 Introduction The home mortgage interest deduction (HMID) is the single largest U.S. tax expenditure other than the exclusion of employer-provided health insurance. Over 35 million households claim the federal HMID (Brady, Cronin and Houser 2003), and the HMID is estimated to cost the U.S. government over $75 billion in foregone tax revenue in 2016 (U.S. Treasury Department 2015). The HMID reduces households’ net cost of debt-financed home purchases, relative to other purchases of goods and services. The tax expenditure can increase home ownership, the size and quality of purchased homes, and the average share of each home purchase financed by debt rather than by down payments (Hendershott and Pryce 2006, Poterba and Sinai 2011). People who own homes with large mortgages may migrate at lower rates when house prices decline because they face larger moving costs: paying off “underwater mortgages” whose balance exceeds the home’s current market price (Ferreira, Gyourko and Tracy 2010). ∗ Email: yagan@berkeley.edu. The opinions expressed in this paper are those of the author alone and do not necessarily reflect the views of the Internal Revenue Service or the U.S. Treasury Department. This work is a component of a larger project examining the effects of tax expenditures on the budget deficit and economic activity. The tax data were accessed under IRS contract TIRNO-12-P-00374. 1 Analyses of the effects of the HMID on the U.S. economy have typically focused on static effects: its effect on a given year’s tax revenue (e.g. Poterba and Sinai 2011, U.S. Treasury Department 2015), distortions between housing and other forms of consumption (Aaron 1972, Rosen 1979, 1985, Mills 1987, Poterba 1984, 1992), and the quantity and price of housing at a given point in time (Hilber and Turner 2014). This paper estimates a dynamic effect of the HMID: the degree to which the HMID impeded adjustment to the Great Recession by impeding migration. The Great Recession had dramatically different effects across space; for example, America’s sixth largest city (Phoenix, Arizona) suffered a large decline in employment while America’s seventh largest city (San Antonio, Texas) suffered only a small decline. Migration is the primary way that the U.S. labor market adjusts to local employment shocks (Blanchard and Katz 1992, Bound and Holzer 2000), and the option to migrate to stronger labor markets is a primary way that the U.S. economy could have insured the original residents of places like Phoenix against especially adverse employment losses. But by encouraging people to buy houses and to buy houses with larger mortgages, the HMID may have impeded migration and thus impeded adjustment (Molloy, Smith and Wozniak 2011). I investigate this dynamic distortion channel with a novel empirical strategy. I first estimate the effect of the HMID on migration over the Great Recession using variation in home mortgage interest (HMI) deductibility at the state level. Specifically, I use selected de-identified data from U.S. tax records to examine workers living in 2007 in the 110 local labor markets—defined as Tolbert and Sizer’s (1996) Commuting Zones (CZ)—that straddle the borders of two or more states, 66 CZs of which allow for different degrees of HMI deductibility on either side of the border. For example, Arkansas allows HMI deductibility from state personal income taxes while Texas has no broad-based personal income tax, and the Texarkana CZ comprises counties on either side of the Arkansas-Texas border. I further attempt to hold all else equal by comparing very similar workers across borders: those who are the same age, earned the same amount, and worked in the same industry in 2006, just in different locales. I then compare the 2007-2015 migration rates of people who had claimed the federal HMID on the Arkansas side of the border in 2006 to the migration rates of people who claimed the federal HMID on the Texas side of the border in 2006 within firms (and similarly for other CZs). These people had very similar skills and lived in the same CZ and so were subject to similar labor market conditions and had revealed similar preferences for where in the United States to live, but faced very different incentives to purchases houses with large mortgages. After estimating the effect of state HMIDs on 2007-2015 migration, I then estimate the value of migration in escaping the incidence of especially severe Great Recession local shocks. This analysis utilizes cross-CZ variation in the severity of the 2007-2009 recession while continuing to utilize within-industry variation in similar workers’ locations to hold all else equal. I investigate the limited extent of “migratory insurance”—which I define as the degree to which 2 2015 employment differs across workers based on where they were living in 2007—given all existing adjustment mechanisms. I conduct several robustness checks to ensure the validity of the comparisons, as well as correlations that investigate a role for underwater mortgages and the HMID. I then assess the likelihood that extra migration unleashed by a hypothetical removal of the HMID could indeed have improved migratory insurance. The remainder of the paper is organized as follows. Section 2 presents background on the HMID and details the empirical strategy. Section 3 introduces the tax data. Section 4 presents estimates of the effect of the HMID on migration rates since the Great Recession. Section 5 presents an analogous analysis of mortgage holding, the key HMID effect channel. Section 6 documents the enduring need for migratory insurance in spite of other existing adjustment mechanisms. Section 7 assesses the degree to which the HMID hindered migratory insurance since the Great Recession. Section 8 concludes. 2 The HMID and the Empirical Strategy This section details the home mortgage interest deduction (HMID) and this paper’s empirical strategy based on state-level variation in the HMID and within-industry comparisons. 2.1 Background on the HMID Since the beginning of the federal income tax, Congress has permitted households to deduct interest payments on personal loans from their federal taxable income. The Tax Reform Act of 1986 eliminated the federal deductibility of many interest payments on consumer loans such as credit card payments but retained the deductibility of interest payments on mortgages with certain mild restrictions. To a first approximation at the federal level, homeowners can deduct interest payments on mortgages on first and second homes that total up to $1 million, as well as interest payments on home equity loans that total up to $100,000 (see IRS Publication 936 for full details). Hence, the vast majority of U.S. homeowners may deduct all of their interest payments from their federal taxable income. Home mortgage interest (HMI) deductibility at the state level varies considerably (ITEP 2011). Twenty-six states generally follow federal rules for HMI deductibility from state taxable income. Another five states and the District of Columbia follow federal rules but apply stricter limitations to the value and type of mortgage interest payments that can be deducted. Ten states do not allow any HMI deductions. Finally, nine states do not assess a broad-based personal income tax, instead raising revenue from other sources such as sales taxes that do not subsidize HMI. Table 1 lists all states in these categories and their associated top personal income tax rates, an easy-to-compare measure of the relative value of HMI deductibility across 3 states. The value of the HMID depends on the personal income tax rate that would otherwise be paid on the income that was deducted. To be concrete using an example that will surface later in the section, consider a married-filing-jointly household in Arkansas that earns $75,000 in gross income in a given year and pays $10,000 in deductible mortgage interest in that year. Arkansas allows the HMID and taxes residents’ income above $50,000 at a rate of 7%, so the household saves $10, 000 × 7% = $700 in taxes thanks to HMI deductibility.1 Hence, HMI deductibility is more valuable when tax rates are higher. 2.2 Empirical Strategy The central question of this paper is whether the HMID reduced Americans’ migration-based insurance (“migratory insurance”) against local variation in the Great Recession. Empirically identifying the effect of the HMID requires variation in HMI deductibility or the value of HMI deductibility and the ability to compare similar people subject to different deductibility while holding all else equal. Two leading sources of variation are problematic. First, because this paper focuses on a single time period, variation in the value of the HMID over time—such as when personal tax rates change or when inflation rates change as in Glaeser and Shapiro (2003)—is not useful. Second, taxpayers at different income levels are different in numerous ways that may affect migratory insurance over the Great Recession independent of the value of HMI deductibility, so comparing migratory insurance across taxpayers of different income levels would likely be problematic. Instead, this paper utilizes variation in HMI deductibility across state borders and holds all else equal by comparing workers within industries and firms. As detailed in the previous subsection, some states permit HMI deductibility while other states do not. Figure 1A displays this variation graphically. The states in white do not allow the HMID (either explicitly, or implicitly because they lack broad-based personal income taxes). The states in colors allow the HMID; the colors plot the top state personal tax rate in such states, so the HMID is more valuable in states with darker colors. Because migration rates can vary across states for reasons other than the HMID, I focus on small geographical areas that are economically connected but straddle state borders. The local area concept I use in this paper is called the Commuting Zone (CZ)—geographic units designed by Tolbert and Sizer (1996) to approximate U.S. local labor markets. Specifically, they used commuting patterns reported in the 1990 Census to divide the united States in 741 areas that 1 This example calculation ignores the fact that, in the absence of deducting HMI, the household could take the standard deduction of $4,000, so the first $4,000 in itemized deductions do not in fact represent net tax savings. This is an appropriate shortcut for this example because states allow numerous other deductions such as for real estate taxes that could exceed the $4,000 standard deduction on their own. 4 share strong commuting ties relative to nearby areas. Statistically, they are aggregations of counties; in rural areas, they may include only one or two counties, while in urban areas, they may include several. CZs have been used recently in economics research by Autor, Dorn and Hanson (2013) among others. I utilize variation within the 66 CZs in the continental United States that straddle the border of two states or more states that differ in HMID deductibility. Figure 1B displays those 66 CZs. They are distributed broadly across the United States. For concreteness, Figure 2 zooms in on the Texarkana CZ, which comprises Texarkana, Arkansas, Texarkana, Texas, and surrounding counties. Arkansas allows HMI deductibility while Texas does not. Using this CZ as an example, this compares 2007-2015 outcomes among people who at the beginning of 2007 (i.e. before the recession) lived in the Texarkana CZ on the Arkansas side of the border, relative to people who at the beginning of 2007 lived in the Texarkana CZ but on the Texas side of the border. By focusing on such within-CZ differences in migration rates, I hope to hold constant numerous factors that may independently influence migration. Upon estimating the effect of the HMID on migration, I then estimate the effect of migration on adjustment to the Great Recession. As discussed in the introduction and documented below, a striking feature of the Great Recession is that it yielded dramatic variation in employment shocks across space. Migration is the key adjustment mechanism by which workers can escape the incidence of large and enduring local shocks: workers in heavily-shocked places may be able to move to lightly-shocked places and compete for employment there. Since CZ’s approximate self-contained local labor markets, I focus on employment shock variation at the CZ level and estimate the degree to which the 2007 residents of heavily-shocked CZ’s were able to diffuse their CZ’s shocks across workers nationwide by migrating and finding employment in other CZs. To the extent that 2007 location affects 2015 employment and thus that migratory insurance is incomplete, I estimate the degree to which greater migration from a hypothetical removal of the HMID may have enabled greater insurance. It is instructive to note that, using the kinds of designs I have just described, one cannot estimate the direct effect of the HMID on adjustment to the Great Recession and instead must do so in the two specified stages. The effect of the HMID is identified only from within-CZ differences, while the effect of migration on adjustment to the Great Recession is identified only from cross-CZ differences. These differences stem from the sources of identifying variation. HMI deductibility varies across states, so the best hope of holding all else equal derives from narrow comparisons across state borders, such as within CZs. In contrast, local employment shocks are in principle shocks to entire local labor markets, so there is little credible variation in local employment shocks within CZ’s. 5 3 Data I implement this paper’s empirical design using selected de-identified data from federal income tax records spanning 1999-2015. All analyses were conducted at secure government facilities and on datasets stripped of unmasked personal identifiers. The sample construction is summarized as follows; additional details are listed in the Data Appendix. 3.1 Analysis Samples I implement the paper’s empirical design using selected de-identified data from federal income tax records spanning 1999-2015. I construct five samples as follows. All five samples are balanced panels of individuals. Random Sample. The main sample comprises a 2% random sample from what I call the full sample. The full sample comprises all American citizens aged 30-49 (“working age”) on January 1, 2007, who had not died by December 31, 2015, and who had a valid payee ZIP code on at least one information return that indicates continental U.S. residence in January 2007. The age restriction confines the 1999-2015 employment analysis to those older than schooling age and younger than retirement age. Birth, death, and citizenship data are drawn from Social Security Administration (SSA) records housed alongside tax records.2 Restricting attention to those alive in 2015 excludes analysis of mortality effects, likely a conservative choice (Sullivan and Von Wachter 2009). I describe geocoded information returns in the next subsection. I randomly sample individuals from the full sample using the last two digits of the individual’s masked identification number, yielding the “full analysis sample” of 1,357,974 people working across 722 CZs. 3 Restricting to CZs that straddle a state border leaves me with a “random border analysis sample” of 233,530 workers in 115 CZs. Since only 66 CZs straddle state borders with different HMI deductability on either side, regressions in the border analysis sample derive identifying variation from 66 CZs, not 115. Retail Chain Sample. The retail chain sample comprises individuals in the full sample whose main employer in 2006 was a retail chain firm and who lived outside of the local area of the retail chain firm’s headquarters. It is constructed as follows. For every individual in the full sample with a 2006 W-2 form, I attempt to link the masked employer identification number (EIN) on the individual’s highest-paying 2006 W-2 to at least one business return in 2 Citizenship is recorded as of December 2016. Results are very similar when not conditioning on citizenship status. Conditioning on citizenship reduces the possibility that 2007 residents are employed in other countries but appear non-employed in U.S. tax data. 3 The sample is smaller than the universe of CZs for three main reasons: the age range restrictions, mismatches between W-2 EIN and business return EIN, and conservative removal of workers at firm headquarters and those not in the continental United States; see the Data Appendix for more details. 6 the universe of business income tax returns 1999-2007.4 I use the North American Industry Classification System (NAICS) code on the business income tax return to restrict attention to workers whose 2006 firms operated in the two-digit-NAICS retail trade industries (44 or 45), e.g. Walmart and Safeway.5 I further exclude employees living in 2007 in the CZ of their employer’s headquarters, using the workers’ payee ZIP codes across their information returns (see the next subsection) and the filing ZIP code on business income tax returns and mapping these ZIP codes to Commuting Zones (CZs, the local area concept defined in the next subsection). Then to identify CZs in which the 2006 firms operated, I further restrict to firms with at least ten 2006 employees living in each of at least five CZs and restrict to the firms’ employees living in 2007 in those CZs.6 This procedure yields a retail chain sample of 866,038 individuals at 524 retail firms.7 Then analogously to how I created the random border analysis sample, I create a “retail border analysis sample” comprising 147,334 individuals in 110 CZs. Unlike firms in manufacturing and other industries, retail firms employ workers to perform identical tasks in many different locales. I therefore assume that workers with similar demographics were as good as randomly assigned across 2007 local areas conditional on their 2006 retail firms and the amount they earned at their 2006 firms. Mass Layoffs Sample. The mass layoffs sample comprises individuals in the full sample who separated from an employer during a mass-layoff event in either 2008 or 2009, after having worked for the employer during the prior three calendar years inclusive of the separation year. It is constructed as follows, closely adhering to the sampling frame of Davis and Von Wachter (2011) except that I define an employer as an EIN-CZ pair rather than an EIN.8 Using the universe of W-2s and linking W-2 payee (residential) ZIP codes to CZs, I compute annual employment counts at the EIN-CZ level. For an employer to qualify as having a mass-layoff event in year t ∈ {2008, 2009}, the employer must satisfy the following conditions: it had at least 50 employees in t − 1; employment contracted by 30% to 99% from t − 1 to t + 1; employment in t − 1 was no greater than 130% of t − 2 employment; and t + 2 employment was less than 90% of t−1 employment.9 The mass layoffs sample comprises all 1,001,543 individuals in the full sample who received a W-2 with positive earnings in years t − 2 through year t from 4 Many firms’ workers cannot be linked to a business income tax return; see the next subsection. Accessed data lacked firm names. I do not know which specific firms survived the sample restrictions. These example firms and their industry codes were found on Yahoo Finance. 6 As in other U.S. administrative data (e.g. Census’s Longitudinal Employer Household Dynamics, see Walker 2013), specific establishments of multi-establishment firms are not directly identified in federal tax data. 7 The sample is smaller than the universe of retail chain workers for four main reasons: the age restriction, the de facto exclusion of workers at independently owned franchises, mismatches between W-2 EIN and business return EIN, and removal of workers at firm headquarters. 8 An EIN may be a firm or a division of a firm. 9 The 99% threshold protects against EIN changes yielding erroneous mass-layoff events. The last two criteria exclude temporary employment fluctuations. A firm that initially qualifies as having mass-layoff events in both 2008 and 2009 is assigned a 2008 event only. 5 7 a mass-layoff employer but not in t + 1. 3.2 Variable Definitions I now define variables. Year refers to calendar year unless otherwise specified. Variables are available 1999-2015. 1. Outcomes. 2007-2015 migration is defined as a worker possessing a 2015 CZ that is different from her 2007 CZ. 2007 CZ is the CZ corresponding to the payee (residential) ZIP code that appears most frequently for the individual in 2006 among the approximately thirty types of information returns (filed mandatorily by institutions on behalf of an individual, including W-2s).10 Information returns are typically issued in January of the following year, so the ZIP code on a individual’s 2006 information return typically refers to the individual’s location as of January 2007. 2015 CZ is defined analogously to 2015 CZ, except that if an individual lacks an information return in 2014, I impute CZ using information return ZIP code from the most recently preceding year in which the individual received an information return. 2007 state denotes the state with most or all of the 2007 CZ’s population. A mover is someone who migrated between 2007 and 2015. 2007 state deductability of mortgage interest equals zero if the worker’s 2007 state (defined analogously to 2007 CZ–i.e corresponding to the worker’s payee ZIP code that appears most frequently across the worker’s 2006 information returns) does not allow mortgage interest deductability from state personal income taxes or if the state lacks a personal income tax. It equals one if worker’s 2007 state allows full deductability of mortgage interest from the state personal income tax. When defined “inclusively”, I code partial-deductability states as one; when defined “exclusively”, I code partial-deductability states as zero. See Table 1 for deductability by state.11 2007 top rate deductability of mortgage interest equals 2007 state deductability of mortgage interest, multiplied by the worker’s 2007 state’s personal income tax rate. 2006 mortgage holder is a binary indicator for whether a Form 1098 information return was issued on the worker’s behalf by a mortgage servicer in 2006.12 Employment in a given year is an indicator for whether an individual has positive Form 10 Numerous activities trigger information returns including formal and independent contractor employment; SSA or UI benefit receipt; mortgage interest payment; business or other capital income; retirement account distribution; education and health savings account distribution; debt forgiveness; lottery winning; and college attendance. A comparison to external data suggests that 98.2% of the U.S. population appeared on some form submitted to the IRS in 2003 (Mortenson, Cilke, Udell and Zytnick 2009). 11 For standard errors, I cluster on the 2007 state with most or all of the worker’s 2007 CZ’s population, following earlier work. 12 A mortgage servicer is required to file a Form 1098 on behalf of any individual from whom the servicer receives at least $600 in mortgage interest on any one mortgage during the calendar year. 8 W-2 earnings or Form 1099-MISC independent contractor earnings (both filed mandatorily by the employer) in the year. Employment is thus a measure of having been employed at any time during the year. Note that this annual employment measure differs from the conventional point-in-time (survey reference week) measure used by the Bureau of Labor Statistics. Earnings in a given year represents labor income and equals the sum of an individual’s Form W-2 earnings and Form 1099-MISC independent contractor earnings. All dollar values are measured in 2015 dollars, adjusting for inflation using the headline consumer price index (CPI-U) and are top-coded at $500,000 after inflating. DI receipt is an indicator for whether the individual has positive Social Security Disability Insurance income (SSDI) in the year as recorded on Form 1099-SSA information returns filed mandatorily by the Social Security Administration. SSDI is the main disability insurance program in the United States. UI receipt is an indicator for whether the individual has positive unemployment insurance benefit income in the year as recorded on Form 1099-G information returns filed mandatorily by state governments. 2. Great Recession Local Shock. Each individual’s Great Recession local shock equals the percentage-point change in the individual’s 2007 CZ’s unemployment rate from 2007 to 2009. Annual CZ unemployment rates are computed by aggregating monthly population-weighted county-level unemployment rates from the monthly Bureau of Labor Statistics Local Area Unemployment Statistics series to the CZ-month level, then averaging evenly within CZ-years across months. 3. Covariates. Age is defined as of January 1 of the year, using date of birth from SSA records housed alongside tax records. Following Autor, Dorn, Hanson and Song (2014), an individual had high labor force attachment if she earned at least $10,382 in 2015 dollars—the compensation for 1,600 hours of work at the 2004 federal minimum wage in 2015 dollars—of earnings in each of the four years 2003-2006. An individual had no labor force attachment if she had zero earnings in any year 2003-2006. Female is an indicator for being recorded as female in SSA records. 1040 filer is an indicator for whether the individual appeared as either a primary or secondary filer on a Form 1040 tax return in tax year 2006. Married is an indicator for whether the individual was either the primary or secondary filer on a married-filing-jointly or married-filing-separately 1040 return in tax year 2006. Number of kids equals the number of children (zero, one, or two-or-more) living with the individual as recorded on the individual’s 2006 1040 if the individual was a 1040 filer and zero otherwise. Mortgage holder is an indicator for whether a Form 1098 information return was issued on the individual’s behalf by a mortgage servicer in 2006.13 Birth state is derived from SSA records and, for immigrants, equals the state of naturalization. 13 A mortgage servicer is required to file a Form 1098 on behalf of any individual from whom the servicer receives at least $600 in mortgage interest on any one mortgage during the calendar year. 9 2006 industry equals the four-digit NAICS industry code on the business income tax return of an individual’s highest-paying 2006 Form W-2, whenever a match can be made between the masked EIN on the W-2 and the masked EIN on the business income tax return. Fourdigit NAICS codes are quite narrow, distinguishing for example between restaurants and bars. As displayed below in summary stats and similar to parallel work (Kline, Petkova, Williams and Zidar 2017, Mogstad, Lamadon and Setzler 2017), almost half of all W-2 earners could not be matched—likely because the employer is a government entity (which does not file an income tax return, covering 15-20% of employment) or because the firm uses a different EIN (e.g. a non-tax-filing subsidiary) to pay workers from the one that appears on the firm’s tax return. For the construction of fixed effects, I assign individuals with missing industry to their own exclusive industry; I assign non-W-2-earning contractors to their own exclusive industry; and I assign the non-employed to their own exclusive industry. I show below that results are nearly unchanged when restricting the sample to the non-employed and those with a valid W-2 industry, for whom the correct industry is universally observed. 2006 age-earnings-industry fixed effects are interactions between age (measured in one-year increments), 2006 industry, and sixteen bins of the individual’s 2006 earnings (in 2015 dollars inflated by the CPI-U) from the individual’s highest-paying employer.14 2006 firm equals the masked employer identification number on the individual’s highest-paying 2006 W-2. 2006 age-earnings-firm fixed effects are constructed analogously to 2006 age-earnings-industry fixed effects. Other controls are used only for robustness checks and are defined when used. 3.3 Summary Statistics Table 2 reports summary statistics for the five data samples used in both the main analyses and the robustness checks: a random 2% sample of the full population and satisfy the restrictions described above, a random 2% “border” sample that restricts the full population sample to those who live in CZs that straddle two states, the retail chain sample (all non-headquarters workers for identifiable retail chain firms in 2006), the border retail sample, and the mass layoffs sample (all workers who separated from a firm in a 2008 or 2009 mass layoff). By the characteristics described in the table, the border samples and the full analysis samples are broadly similar. Compared to the random sample of the full population, the retail sample is poorer, more female, is less likely to get married, have kids, or own homes. 15 The mass layoffs 14 The main result below is nearly identical when using Local CPI 2—the more aggressive of the Moretti (2013) local price deflators—to locally deflate 2006 earnings before binning. Chosen to create roughly even-sized bins, the bin minimums are: $0, $2,000, $4,000, $6,000, $8,000, $10,000, $15,000, $20,000, $25,000, $30,000, $35,000, $40,000, $45,000, $50,000, $75,000, and $100,000. 15 The mortgage holder shares in the border and full analysis samples are lower than the U.S. adult home ownership rate: the sample is younger and poorer than the U.S. as a whole, the mortgage holder share excludes home owners without a mortgage, and mortgages held only in the name of a worker’s spouse or other third 10 sample, in contrast, is richer and more male than the random sample, but still less likely to get married or have kids. 4 Effect of the HMID on Migration In this section, I use the state-border empirical strategy in the random 2% border sample to estimate the effect of the HMID on 2007-2015 migration. I first estimate the effect of the HMID on migration using a reduced-form specification that puts no structure on the nature or strength of the mechanism by which a tax subsidy to mortgage interest affects migration. I then apply a series of robustness checks, first applying rich controls to the reduced-form estimates, allowing partial HMID states to be misclassified and allowing mortgage interest tax subsidies to have a linear effect on migration. As a further robustness check, I rerun this analysis on the retail border sample, which has the additional benefit of comparing especially similar workers. 4.1 Main Results To estimate the effect of the HMID on migration using the state-border empirical strategy, I estimate regressions in the border analysis sample of the form: M IGRAT EDi = βHM IDST AT Es(i2007) + Xi2007c(i2007) γ, (4.1) where M IGRAT EDi is an indicator for whether worker i migrated across CZs between 2007 and 2015, HM IDs(i2007) is an indicator for whether i was living in 2007 in an HMID-eligible state, and Xi2007r(i,2007) is a vector of individual-level and CZ-level covariates. The coefficient β̂ is the coefficient of interest: the estimated effect of living in an HMID-eligible state in 2007 on whether the worker migrated 2007-2015. Table 3A displays the main results: estimated effects of living in an HMID-eligible state in 2007 on 2007-2015 migration, under successively larger sets of controls. HMID-eligible states are defined inclusively: partial deductibility states are classified as allowing HMI deductibility. Column 1 has no controls. Column 2 adds age fixed effects, Column 3 uses fixed effects of age interacted with earnings bins, and Column 4 further interacts these age-earnings with industry fixed effects based on NAICS code. These age-earnings-industry fixed effects in Column 4 represent my preferred specification, and all subsequent columns control for these effects while adding additional controls that one might concerned about, such as gender and marital status. Column 1 shows that, in the cross section, workers who lived in 2007 in a state that permits HMID deductibility are estimated to have been 2.425 percentage points more likely to have party are not included here. 11 migrated 2007-2015. This estimate is statistically significant and has the unexpected sign, which would be consistent with the HMID increasing migratory insurance. The point estimate is fairly insensitive to additional controls, except for CZ size, which is added in Column 9. Controlling for CZ size reduces the coefficient to 1.739, which implies that populous CZs are disproportionally located in states with HMID and that people in populous CZs were more likely to have moved. The additional controls in Table 3A do, however, highlight the weakness of the main result. Several columns, including my preferred estimate in Column 4, are statistically insignificant at the 95% confidence level, meaning that one cannot say with confidence that there is an effect of being in an HMID state on migration. In fact, the data do not reject a negative effect of the HMID on 2007-2009 migration under reasonable specifications. 4.2 Robustness to State Misclassification The main results in Table 3A classify states as HMI deductible or nondeductible states inclusively: partial deductibility states are classified as allowing HMI deductibility. However, it is possible that the near-zero estimates of Table 3A are attenuated toward zero because of misclassification: perhaps partial deductibility states have such muted effects of home ownership and mortgage leverage that they are effectively the same as non-deductible states. Table 3B therefore defines deductibility exclusively: partial deductibility states are classified as not allowing HMI deductibility and thus lumped in with states that either do not allow HMI deductibility in their personal income tax or do not have a personal income tax at all. Table 3B shows that the exclusive definition does indeed somewhat alter the point estimates, though no more so than adding controls did in Table 3A. It also points to a cautious interpretaion of the point estimates based on the size of the 95% confidence intervals. The confidence interval on the coefficient in Column 15, for instance, is [−0.26, 4.52], which again means that living in a (full) HMID state could increase migration, decrease migration or have no impact on migration at all. 4.3 Robustness to Linear Specifications The results of Table 3 use a simple binary classification of HMI deductibility. Despite its simplicity, the binary specification could lack statistical power relative to a specification that allows for larger effects of among states with larger HMI deductibility. Table 4 therefore replicates Table 3 using the continuous measure of HMI deductibility of the worker’s 2007 top rate deductibility of mortgage interest.16 16 Though many workers are not in the top state income tax bracket, this measure is simple and readily available. 12 Table 4 yields point estimates that are intermittently significant at the 95% level. Consistent with Table 3, there is little difference in the estimates in Panel A and Panel B, implying that the classification of partial HMID states is not particularly important. Figure 3A non-parametrically presents the result in Table 4A Column 4. It is constructed by regressing 2007-2015 migration and 2007 top rate deductibility of mortgage interest on the controls underlying Column 4, computing residuals, adding back their means for interpretation, and plotting means of the 2007-2015 migration residuals within twenty equal-sized bins of the 2007 top rate deductibility residuals. Overlaid is the best-fit line estimated by regressing the 2007-2015 migration residuals on the 2007 top rate deductibility residuals, whose slope of course equals the 0.295 reported in Table 4A Column 4. As one can see from the graph, the marginally significant result under the linear specification does not appear to be masking a visually obvious non-linear relationship, further suggesting that there is indeed no statistically significant relationship between these two variables. Table 4B Column 15 presents analogous results using the exclusively defined measure of state HMI deductibility. Like Table 4A, Table 4B reports positive estimates are that are still statistically insignificant. Therefore the null results of Table 3 are robust to the linear specifications presented in Table 4. 5 Effect of the HMID on Mortgage Holding The previous section found no conclusive evidence of the HMID on 2007-2015 migration due to substantial statistical noise. If it had, one would have expected this effect to flow through the mechanism of home ownership. The HMID encourages people to take on mortgages and buy homes, and owning a home in a given area increases the likelihood that people will stay in that area. This section focuses on determining whether the first link in that logical chain holds, looking at whether the people living in states with HMID were indeed more likely to own homes in 2006. As in the previous section, I use a wide range of robustness checks on both the 2% random border sample and the retail border sample in order to present a complete picture of the data. Table 5 replicates Table 3 for the outcome of owning a mortgage in 2006.17 Estimates using the inclusive definition of HMI deductibility reveal a near-zero relationship between deductibility and mortgage holding. Depending on controls, the point estimate of this effect was either positive or negative. In the preferred specification, workers living in 2007 in a state offering deductibility of mortgage interest were insignificantly 0.426 percentage points less likely to hold 17 Recall that although location is measured in 2007 and mortgage holding is measured in 2006, these outcomes are actually simultaneous: both 2007 location and 2006 mortgage holding are measured using 2006 information returns. 13 a mortgage. The standard error 1.213 is substantial, implying a substantial 95% confidence interval [−2.88, 2.03]. Thus subject to statistical uncertainty, I find no evidence that residents of HMI deductible states have higher rates of holding a mortgage. Table 5B shows similar results when using the exclusive definition of HMI deductibility, though the point estimate for my preferred specification is now positive (although still statistically insignificant). Residents of states with HMI deductibility are 0.266 percentage points more likely to hold a mortgage. However, the standard error remains large at 0.731 percentage points. I therefore fail to find a statistically significant positive relationship between binary HMI deductibility and mortgage holding. Turning to the continuous measure of HMI deductibility, Table 6 replicates Table 4 for the mortgage holding outcome. Like the previous tables, I continue to find near-zero and statistically insignificant results. Defining HMI deductibility inclusively, Column 4 reports that residents of states with one-percentage-point higher HMI deductibility were 0.043 percentage points less likely to hold a mortgage, with a standard error of 0.164. Defining HMI deductibility exclusively, Column 15 reports that residents of states with one-percentage-point higher HMI deductibility were 0.093 percentage points more likely to hold a mortgage, with a standard error of 0.088. Thus the two panels of Table 6 both find no statistically significant relationship. Figure 3B non-parametrically presents the result in Table 6A Column 4, using the same method as Figure 3A. Here, one can see that the slightly negative relationship, but the standard error shows that this result is completely statistically insignificant. Thus across specifications, I find no statistically significant relationship between HMI deductibility and mortgage holding. This sheds light on interpreting the previous section’s lack of a statistically significant effect of HMI deductibility on migration. It may indeed be the case that a tax policy that causes people to buy a house or take out a larger mortgage also causes them to migrate less. However, it appears that the HMID may not in fact be such a mortgage-holding-inducing policy. 6 Robustness to Inter-firm Heterogeneity in Workers Although I have thus far failed to find conclusive effects in either direction of the impact of HMID on migration in response to the Great Recession, it is possible that these effects are masked by inter-firm heterogeneity in workers. It is possible, for instance, that different firms within the same industry hire workers of different average skill level, or that some workers develop specialized skills that make them valuable only to a specific firm, and that that firm is located only in a certain area. To address potential selection issues, I used my retail border analysis sample and repeated the analysis that I did for the random 2% border sample, with the full suite of controls. 14 The results are displayed in Tables 7 through 10, with Table 7 corresponding to Table 3, Table 8 corresponding to Table 4 and so on. The results are broadly similar, both for the main analysis and the robustness checks. As with the 2% random sample, the point estimates for the coefficients of interest in Tables 7 and 8 are all positive for Panels A and B, though with greater statistical precision that rejects a substantial negative effect. Tables 9 and 10, like Tables 5 and 6 for the 2% random sample, show statistically insignificant point estimates. Figure 4 is analogous to Figure 3. Figure 4A corresponds to Table 8 Column 5 and Figure 4B corresponds to Tablel 10 Column 5. As with its companion figure, Figure 4A serves to show that there are no non-linearities masking a more significant relationship, and 4B shows a nominally negative, but statistically insignificant relationship. 7 The Enduring Need for Migratory Insurance The previous sections found no significant negative effect of HMI deductibility on 2007-2015 migration rates. However, these effects were estimated with error, failing to reject the possibility that the effect could be negative and substantial. In particular, I estimated a 95%-confidence upper bound of the reduction in migration rates due to HMI deductibility (on average across states with HMI deductibility versus states without HMI deductibility) equal to 0.22 percentagepoints-lower migration rates in the preferred specification (recall Table 3A, Column 4). If migration was exceptionally valuable in avoiding the incidence of local variation in the Great Recession, then the 95%-confidence upper bound effect of HMI deductibility on migratory insurance may yet be large. This section investigates whether there was an enduring need for migratory insurance in the first place, or whether employment rates had converged across space through existing mechanisms. 7.1 Main Effects Figure 5A plots the time series of estimated effects of living in 2007 in a relatively severely shocked CZ, conditional on the main controls in the full analysis sample. The plotted 2015 data point is this subsection’s main result and equals β̂ estimated in: EM P LOY EDi2015 = βSEV EREc(i2007) + Xi2007c(i2007) γ, (7.1) where EM P LOY EDi2015 is an indicator for whether worker i was employed in 2015, SEV EREc(i2007) is an indicator for whether i was living in 2007 in a relatively severely shocked CZ, and Xi2007r(i,2007) is 2006 age-x-earnings-x-industry fixed effects. For other years t, plotted 15 data points equal the same coefficient from a regression of EM P LOY EDit on the exact same right-hand-side values in the exact same sample. 95% confidence intervals are plotted in vertical lines unadjusted for multiple hypotheses, based on standard errors clustered at the 2007-state level. The 2015 data point shows that living in 2007 in a relatively severely shocked CZ is estimated to have caused a 0.393 percentage-point reduction in employment rates, relative to those who in 2007 were living in a relatively mildly shocked CZ (see Table 11 Column 4). The estimate is very significantly different from zero. The mean 2015 employment rate in this sample is 79.1%, so this estimated effect is equal to a 0.41% difference in employment rates. The plotted time series of estimated zero effects 1999-2007 constitute placebo tests corroborating the identifying assumption that conditional on controls, severe- CZ status is as good as randomly assigned. Panels B and C of Figure 5 re-enforce the conclusion of a significant and enduring employment impact from Great Recession local shocks. Panel B shows that there is a roughly linear relationship between Great Recession local shocks and relative employment in 2015 net of controls. Panel C repeats the analysis in Panel A for earnings and shows a similar enduring drop. Table 11 Column 4 displays this main 2015 effect plotted in the Panel A, along with similar effects under different controls. 18 All specifications in columns 1-8 display similarly negative and significant results, regardless of controls. Column 6 shows that the employment impacts were monotonically worse with increasing local shocks, and Columns 9-11 show consistent results using alternate measures of employment impact. As a robustness check, Table 12 repeats the analysis for the retail sample. These estimates are also uniformly negative and statistically significant. Figure 6A is the retail analog for Figure 5A, and similarly, the 2015 data point corresponds to Table 12 Column 5.19 7.2 Robustness Table 13 presents several robustness checks of the impact of Great Recession local shocks on employment in 2015, with all its implications for HMID and migratory insurance. Taking Table 13 Column 4 as a starting point, Columns 2 through 5 add a suite of individual level controls. Columns 6-9 control for CZ-level characteristics. Column 6 controls an individual’s 2007 CZ’s size, equal to the CZ’s total employment in 2006 as reported in Census’s County Business Patterns (CBP). Column 7 controls an individual’s 2007 CZ’s size growth, equal to the CZ’s log change in CBP employment from 2000 to 2006. Column 8 controls for an 18 See Appendix Table 2 for Great Recession local shock by CZ. Table 2 showed that the main and retail chain samples differ demographically, and I find impact heterogeneity across demographic groups. I therefore reweight the retail chain sample to match the main sample as in DiNardo, Fortin and Lemieux (1996) along 2007 CZ, gender, five-year age bin, and 2006 earnings bins. 19 16 individual’s 2007 CZ’s share of workers who work outside of the CZ, computed from the 20062010 American Community Surveys. Column 9 controls for an individual’s 2007 CZ’s state’s maximum unemployment insurance duration over years 2007-2015. Column 10 controls for the individual’s 2007 state’s 2015 minimum wage minus that state’s 2007 minimum wage. Column 11 restricts the sample to the 2006 non-employed and 2006 workers with a valid industry code (i.e. excluding contractors and W-2 earners without industry). Column 12 further restricts the sample by excluding individuals employed in construction or manufacturing in 2006. Column 13 instruments the individual’s Great Recession local shock using the mean of the Great Recession local shock in the individual’s birth state. Throughout all of this, the estimates remain negative and significant. Table 14 provides a finer look at the year-by-year impacts of Great Recession local shocks in 2007. It displays results for migration and various labor market outcomes. These outcomes include those in Column 8, the individual’s unemployment insurance benefits, and Column 9, the individual’s Social Security Disability Insurance benefits in year t. This pair of outcomes is discussed more in the next subsection. The estimates for migration are positive, but insignificant and the estimates for the basic employment measure are negative and significant. The estimates for other measures vary in significance. Migration rates were slightly higher out of severely shocked CZs than other CZs: 18.2% out of most-shocked-quintile CZs and 16.5% out of leastshocked-quintile CZs. However, Columns 4 and 7 reveal no statistically significant evidence that migration enabled individuals in severely shocked areas to find employment and earnings at higher levels in other CZs. This suggests that any extra migration unleashed by a hypothetical removal of the HMID may not have substantially improved migratory insurance. 20 7.3 Tests for Worker-Scarring Mechanisms If living in 2007 in a relatively severely shocked CZ “scarred” workers by reducing their human capital or raising their reservation wages, their employment may be persistently low even if they were to move to a stronger local labor market. That is, the enduring employment impact of 2007 location could be specific to the worker rather than specific to the worker’s 2015 location, meaning that migration would not help the worker escape incidence. I therefore test for leading candidates of worker-specific effects; if I find strong effects, then additional migration likely could not have provided substantially more insurance. A first potential worker-specific channel is disability insurance. Severe Great Recession local shocks may have induced workers to supplement their income with Social Security Disability Insurance (“DI”)—a typically permanent location-independent income stream—thereby permanently raising their reservation wages and reducing their employment independent of 20 Appendix Table 1 replicates Table 14 in the years leading up to the great recession. 17 current location (Autor and Duggan 2003, Maestas, Mullen and Strand 2013). One can estimate an upper bound on the contribution of DI receipt to the main employment result, under the weak monotonicity assumption that the treatment (living in 2007 in a relatively severely shocked area) did not make anyone in the analysis sample less likely to go on DI. The estimated upper bound on the DI mechanism equals the estimated effect on an indicator for 2015 employment (the main result) minus the estimated effect on an indicator for whether the worker was employed in 2015 or was on DI in 2015. Table 15 Column 2 displays the result: living in 2007 in a relatively severely shocked area is estimated to have caused workers to be 0.27 percentage-points less likely to be employed or on DI on 2015. Subtracting this effect from the main −0.393 percentage-point effect on employment, 32.6% of the incrementally non-employed relatively severely shocked natives were on DI by 2015, and thus 32.6% is the estimated upperbound contribution of transition to DI to the enduring employment impact. To the extent that incremental transition to DI was a response to a lack of employment rather than a cause of it, transition to DI explains no more than 32.6% of the employment impact and potentially less. The tight upper-bound on the DI contribution is reflected in the statistically zero impact of living in 2007 in a relatively severely shocked area on 2015 DI receipt (Column 1). A second potential worker-specific channel is more workers being laid off—following a long line of work documenting long-term earnings losses after layoff (Ruhm 1991, Jacobson, LaLonde and Sullivan 1993, Neal 1995, Couch and Placzek 2010). I proxy for layoff using unemployment insurance (UI) receipt.21 Table 15 Column 6 shows that living in 2007 in a relatively severely shocked area caused workers to be 1.43 percentage points more likely to have received UI by 2015 (i.e. at any point 2007-2015), but this effect is significant, but small relative to the sample-wide mean of 25.6 percentage points. This suggests that higher rates of layoff cannot explain relatively severely shocked natives’ lower 2015 employment rates. Column 8 shows that controlling for UI receipt by 2015 barely changes the employment effect estimate. This is of course not quasi-experimental since layoff is endogenous. But if one assumes that the laid-off relatively severely shocked natives were equal or stronger on unobservables than laid-off relatively mildly shocked CZ natives—as would be expected of incremental layoffs in a layoffsand-lemons model (Gibbons and Katz 1991)—then these columns indicate that higher layoffs do not explain the employment results. A third potential worker-specific channel is general human capital decay after long nonemployment spells. Table 15 column 6 indicates that Great Recession local shocks caused individuals to be more likely to spend at least one year 2007-2014 completely non-employed. Column 10 shows that the individual’s employment history 2007-2012 explains nearly the entire 2015 employment impact. Columns 3-5 and Figure 6B reveal similar impacts in the mass 21 Kawano and LaLumia (2017) show that UI-tax-data-based unemployment rates are close in both level and trend to official Bureau-of-Labor-Statistics unemployment rates 1999-2011 (correlation 0.94). 18 layoffs sample, which comprises quite similar workers who were laid off into local labor markets that were more likely (severely shocked CZs) or less likely (other CZs) to lead to a long nonemployment spell. These results are consistent with the worker-specific channel of general human capital decay. Overall, the findings rule out some candidate worker-scarring mechanisms, but general human capital decay via prolonged non-employment is consistent with the results. However, this worker-scarring mechanism is not the only mechanism consistent with the results. The results are also consistent with persistently low local labor demand, under which laid-off workers have not experienced human capital decay but have experienced either a decline in their local wages or cannot obtain desired employment at prevailing local wages. Under persistently low local labor demand, more migration could indeed have provided more insurance: the local areas rather than the workers can be thought of as scarred. Further distinguishing mechanisms is a valuable area for future work. 7.4 Heterogeneity of Effects I close by estimating whether subgroups of workers that had higher migration rates also had attenuated employment effects and thus greater insurance. Figure 7A plots point estimates and 95% confidence intervals for several worker subgroups defined by pre-2007-determined characteristics in the full analysis sample. Each row reports results from estimating equation 7.1 with the main controls on a different subsample: the full analysis sample, by gender, by 2006 earnings bin, by labor force attachment, by 2006 age group, by 2006 marital status, by 2006 number of kids, and by 2006 mortgage holding status. Rates of 2007-2015 migration of the analyzed subsample are listed in the far right of each row. Comparison of subgroup migration rates to subgroup differences presents a surprising result: the effect of 2007 location is not smaller for more mobile subgroups. The finding is most salient for mortgage-holding versus non-mortgage-holding comparison. Mortgage holders had migration rates of only 13% while non-mortgage-holders had 18% migration rates. Yet the two subgroups experienced similar (such that their confidence intervals overlap) 2015 employment effects of living in 2007 in a relatively severely shocked CZ, and if anything, it would appear that the mortgage holders experienced slightly smaller employment effects. These subgroups could of course be different along other dimensions, but this is suggestive evidence that greater migration may not have provided greater insurance against local variation in Great Recession local shocks. Figure 7B plots point estimates and 95% confidence intervals for the same groups, only this time for 2006 earnings. Results are broadly similar. 19 8 Conclusion This paper has investigated whether the home mortgage interest deduction (HMID)—the second-largest U.S. tax expenditure—substantially impeded insurance against Great Recession local labor demand shocks by impeding residents’ migration. Utilizing a novel empirical design based on variation in home mortgage interest deductibility across state borders and comparing similar workers within firms, I find no significant effect of the HMID on migration 2007-2015. However, the statistical uncertainty permits considerable negative effects if affected residents indeed lacked insurance and if existing migration was indeed a beneficial insurance mechanism. I find substantial under-insurance against local variation in the Great Recession: the 2007 residents of severely affected areas were substantially less like to be employed in 2015 than the 2007 residents of mildly affected areas. However, a direct analysis of the insurance benefit of migration reveals no statistically significant evidence that out-migration from severely affected areas was a beneficial insurance mechanism, though with large standard errors. Hence, it remains possible that the HMID hindered adjustment to the Great Recession by hindering migration, but the analysis failed to find significant evidence of it. The results nevertheless inform future work. 20 References Aaron, HJ. 1972. “Shelter and Subsidies: Who benefits from Federal Housing Subsidies?’.” Washington, DC: The Brookings Institution. Autor, David, David Dorn, Gordon H Hanson, and Jae Song. 2014. “Trade Adjustment: Worker-Level Evidence.” The Quarterly Journal of Economics, 129(4): 1799–1860. Autor, David H, and Mark G Duggan. 2003. “The Rise In The Disability Rolls And The Decline In Unemployment.” The Quarterly Journal of Economics, 118(1): 157–205. Autor, David H, David Dorn, and Gordon H Hanson. 2013. “The China Syndrome: Local Labor Market Effects of Import Competition in the United States.” The American Economic Review, 103(6): 2121–2168. Blanchard, Olivier J, and Lawrence F Katz. 1992. “Regional Evolutions.” Brookings papers on economic activity, 1–75. Bound, John, and Harry J Holzer. 2000. “Demand shifts, population adjustments, and labor market outcomes during the 1980s.” Journal of labor Economics, 18(1): 20–54. Brady, Peter, Julie-Anne Cronin, and Scott Houser. 2003. “Regional differences in the utilization of the mortgage interest deduction.” Public Finance Review, 31(4): 327–366. Couch, Kenneth A, and Dana W Placzek. 2010. “Earnings Losses of Displaced Workers Revisited.” The American Economic Review, 572–589. Davis, Steven J, and Till Von Wachter. 2011. “Recessions and the Costs of Job Loss.” Brookings Papers on Economic Activity, 1–55. Department, US Treasury. 2015. “Tax Expenditures.” Analytical Pespectives. DiNardo, John, Nicole M Fortin, and Thomas Lemieux. 1996. “Labor Market Institutions and the Distribution of Wages, 1973-1992: A Semiparametric Approach.” Econometrica, 64(5): 1001–1044. Ferreira, Fernando, Joseph Gyourko, and Joseph Tracy. 2010. “Housing Busts and Household Mobility.” Journal of Urban Economics, 68(1): 34–45. Gibbons, Robert, and Lawrence F Katz. 1991. “Layoffs and Lemons.” Journal of Labor Economics, 9. Glaeser, Edward L, and Jesse M Shapiro. 2003. “The Benefits of the Home Mortgage Interest Deduction.” In Tax Policy and the Economy, Volume 17. 37–82. MIT Press. Hendershott, Patric H, and Gwilym Pryce. 2006. “The sensitivity of homeowner leverage to the deductibility of home mortgage interest.” Journal of Urban Economics, 60(1): 50–68. Hilber, Christian AL, and Tracy M Turner. 2014. “The mortgage interest deduction and its impact on homeownership decisions.” Review of Economics and Statistics, 96(4): 618–637. 21 ITEP. 2011. “State Treatment of Itemized Deductions.” Institute on Taxation and Economic Policy Policy Brief. Jacobson, Louis S, Robert J LaLonde, and Daniel G Sullivan. 1993. “Earnings Losses of Displaced Workers.” The American Economic Review, 685–709. Kawano, Laura, and Sara LaLumia. 2017. “How Income Changes During Unemployment: Evidence from Tax Return Data.” Journal of Human Resources. Kline, Patrick, Neviana Petkova, Heidi Williams, and Owen Zidar. 2017. “Who Profits from Patents? Rent-sharing at Innovative Firms.” mimeo. Maestas, Nicole, Kathleen J Mullen, and Alexander Strand. 2013. “Does Disability Insurance Receipt Discourage Work? Using Examiner Assignment to Estimate Causal Effects of SSDI Receipt.” The American Economic Review, 103(5): 1797–1829. Mills, Edwin S. 1987. “Dividing up the investment pie: have we overinvested in housing?” Business Review, , (Mar): 13–23. Mogstad, Magne, Thibaut Lamadon, and Bradley Setzler. 2017. “Rent Sharing, Earnings Dynamics and Transmission of Firm and Market Level Shocks.” mimeo. Molloy, Raven, Christopher L Smith, and Abigail Wozniak. 2011. “Internal Migration in the United States.” Journal of Economic Perspectives, 25(3): 173–196. Moretti, Enrico. 2013. “Real Wage Inequality.” American Economic Journal. Applied Economics, 5(1): 65. Mortenson, Jacob A, James Cilke, Michael Udell, and Jonathon Zytnick. 2009. “Attaching the Left Tail: A New Profile of Income for Persons Who Do Not Appear on Federal Income Tax Returns.” Vol. 102, 88–102, JSTOR. Neal, Derek. 1995. “Industry-Specific Human Capital: Evidence from Displaced Workers.” Journal of labor Economics, 653–677. Poterba, James, and Todd Sinai. 2011. “Revenue costs and incentive effects of the mortgage interest deduction for owner-occupied housing.” Poterba, J M. 1984. “Tax subsidies to owner-occupied housing: an asset-market approach.” Quarterly Journal of Economics, 99(4): 729. Poterba, J M. 1992. “Taxation and Housing: Old Questions, New Answers.” American Economic Review, 82(2): 237–242. Rosen, Harvey S. 1985. “Housing behavior and the Experimental Housing-Allowance Program: What have we learned?” In Social experimentation. 55–94. University of Chicago Press. Rosen, Sherwin. 1979. “Wage-based Indexes of Urban Quality of Life.” Current Issues in Urban Economics. Edited by Peter Mieszkowski and Mahlon Straszheim, Johns Hopkins Univ. Press. 22 Ruhm, Christopher J. 1991. “Are Workers Permanently Scarred by Job Displacements?” The American Economic Review, 319–324. Sullivan, Daniel, and Till Von Wachter. 2009. “Job Displacement and Mortality: An Analysis Using Administrative Data.” The Quarterly Journal of Economics, 124(3): 1265– 1306. Tolbert, Charles M, and Molly Sizer. 1996. “US Commuting Zones and Labor Market Areas: A 1990 Update.” US Department of Agriculture Economic Research Service Staff Paper, , (9614). Walker, W Reed. 2013. “The Transitional Costs of Sectoral Reallocation: Evidence from the Clean Air Act and the Workforce.” The Quarterly Journal of Economics, 128: 1787–1835. 23 Data Appendix This appendix section provides additional data details. First, the universe of business tax returns used is the universe of C-corporate (Form 1120), S-corporate (Form 1120S), and partnership (Form 1065) tax returns. Businesses that file other types of tax returns employ a small share of U.S. workers. Second, Form 1099-MISC data on independent contractor employment are missing in 1999. Results are very similar when omitting 1999 data. Third, many retail chain firms are missing from the retail border and chain samples, both because of subsidiaries and franchises and also because a (likely small) number of firms outsource their W-2 administration to third-party payroll administration firms that list their own EINs on W-2s. Nevertheless, the retail chain sample includes very large nationwide chains. Fourth and also specific to the retail border and chain samples, the filing ZIP code on a firm’s business income tax return typically but not always refer to the business’s headquarters ZIP code. Excluding workers at the business’s headquarters is useful because headquarters workers may perform systematically different tasks than workers at other establishments and thus may possess different human capital even conditional on baseline earnings. I therefore conservatively exclude firms’ workers living in the CZ with the largest number of the firm’s workers living there, as well as the CZ with the largest number of the firm’s workers living there as a share of the total number of workers living there. Fifth and also specific to the retail border and chain samples, I consider a firm to have operated in a CZ in 2006 if it employed at least ten stably located workers who lived in the CZ—defined as individuals of any age and citizenship with a W-2 from the firm in all years 2005-2007 and the same residential CZ in all years 2005-2007 based on those W-2s’ payee (residential) ZIP codes. It is necessary to define CZ operations using more than one year of W-2 data because W-2 payee ZIP code refers to the worker’s ZIP code in January of the year after employment. That feature implies that almost all firms would appear to have operations in every large CZ if one were to use only 2006 W-2s to identify CZ operations, since many workers move to large cities. 24 Figure 1: HMI Tax Subsidy Rates across U.S. States A. HMI Tax Subsidy Rates by State B. CZs that Straddle States with Different HMI Subsidies Notes: Panel A plots HMI (home mortgage interest) tax subsidy rates by state, equal to zero for states that do not allow HMI deductions or lack a personal income tax, and equal to the top state marginal personal income tax rate for states that do allow HMI deductions. Panel B highlights the 66 Commuting Zones (CZs) that straddle borders between at least two states with different HMI subsidies. 25 Figure 2: Example of a CZ Utilized in the HMID Cross-Border Design Notes: Commuting Zones (CZs) are collections of counties that correspond to relatively self-contained local labor markets. This paper estimates the effect of the HMI deduction on migration rates by comparing migration rate differences across state borders in the CZs that straddle the border between at least two states that have different HMI subsidy rates. The Texarkana CZ is one such CZ. The main cities in the Texarkana CZ are Texarkana, Texas, and Texarkana, Arkansas. The CZ encompasses these two cities’ counties and nearby counties. Arkanasas allows for HMI deductibility from state personal income taxes, while Texas does not have a state personal income tax from which HMI could be deducted. 26 Figure 3: 2% Random Sample Visualizations 13 14 Migration rate (%) 15 16 17 A. 2007-2015 Migration Rates versus HMID Tax Subsidy Rates 0 2 4 6 HMID tax expenditure subsidy 8 10 38 Mortgage holding rate (%) 39 40 41 B. 2006 Mortgage Holding versus HMID Tax Subsidy Rates 0 2 4 6 HMID tax expenditure subsidy 8 10 Notes: Panel A non-parametrically depicts the relationship between workers’ 2007-2015 migration rates and their 2007 state’s HMID tax subsidy. It does so by regressing migration rates and HMID tax subsidy on the main controls, computing residuals, added back their means for interpretation, and plotting means of the migration rate residuals within twenty equal-sized bins of the HMID tax subsidy residuals. Overlaid is the best-fit line (slope 0.295, standard error 0.152). Panel B is the analogous figure showing the relationship between workers’ 2006 mortgage holding and their 2007 state’s HMID tax subsidy. Overlaid is the best-fit line (slope −0.043, standard error 0.164). 27 Figure 4: Retail Sample Visualizations 16 17 Migration rate (%) 18 19 20 A. 2007-2015 Migration Rates versus HMID Tax Subsidy Rates 0 5 HMID tax expenditure subsidy 10 25 Mortgage holding rate (%) 26 27 28 B. 2006 Mortgage Holding versus HMID Tax Subsidy Rates 0 5 HMID tax expenditure subsidy 10 Notes: Panel A non-parametrically depicts the relationship between workers’ 2007-2015 migration rates and their 2007 state’s HMID tax subsidy. It does so by regressing migration rates and HMID tax subsidy on the main controls, computing residuals, added back their means for interpretation, and plotting means of the migration rate residuals within twenty equal-sized bins of the HMID tax subsidy residuals. Overlaid is the best-fit line (slope 0.328, standard error 0.127). Panel B is the analogous figure showing the relationship between workers’ 2006 mortgage holding and their 2007 state’s HMID tax subsidy. Overlaid is the best-fit line (slope −0.147, standard error 0.156). 28 Figure 5: Employment and Earnings Impacts of Great Recession Local Shocks .25 0 -.25 -.5 2015 2014 2013 2012 2011 2010 C. Earnings Impact of Great Recession Local Shocks 0 -500 2015 2014 2013 2012 2011 2010 2009 2008 2007 2006 8 2005 7 2004 6 2003 5 2002 4 Great Recession local shock 2001 3 2000 2 1999 -1500 -1000 -7 -7.5 -8 -8.5 -9 2015 relative employment (pp) -6.5 Effect of Great Recession local shock ($) 500 B. Non-Parametric Visualization of the 2015 Impact 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 -.75 Effect of Great Recession local shock (pp) .5 A. Employment Impact of Great Recession Local Shocks Notes: Panel A plots regression estimates of the effect of Great Recession local shocks on annual relative employment conditional on 2006 age-earnings-industry fixed effects in the main sample (a 2% random sample). Each year t’s outcome is year-t relative employment: the individual’s year-t employment (binary employment status) minus the individual’s mean 1999-2006 annual employment. 95% confidence intervals are plotted around estimates, clustering on 2007 state. For reference, the 2015 data point (the paper’s main estimate) implies that a 1-percentage-point higher Great Recession local shock caused individuals to be 0.393 percentage points less likely to be employed in 2015. Panel B non-parametrically depicts the relationship underlying the Panel A 2015 data point. It is produced by regressing Great Recession local shocks on 2006 age-earnings-industry fixed effects, computing residuals, adding back their means for interpretation, and plotting means of the 2015 relative employment within twenty equal-sized bins of the shock residuals. Overlaid is the best-fit line, whose slope is equal to panel A 2015 data point. Panel C replicates panel A for the outcome of year-t relative earnings: the individual’s year-t earnings minus the individual’s mean 1999-2006 annual earnings. 29 Figure 6: Employment Impacts in Special Samples .2 0 -.2 -.4 2011 2012 2013 2014 2015 2011 2012 2013 2014 2015 2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 -.6 Effect of Great Recession local shock (pp) .4 A. Retail Chain Sample 0 -.5 -1 2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 -1.5 1999 Effect of Great Recession local shock (pp) .5 B. Mass Layoffs Sample Notes: Panel A replicates Figure 5A in the retail chain sample (all non-headquarters workers for identifiable retail chain firms in 2006). Panel B replicates Figure 5A in the mass layoffs sample (all workers who separated from a firm in a 2008 or 2009 mass layoff). See the notes to Figure 5A for specification details. 30 Figure 7: Impact Heterogeneity A. Employment Subgroup Migration Rate Overall 16% Earnings $0 Earnings $1-$15k Earnings $15k-$45k Earnings $45k+ 16% 20% 16% 15% No LF attachment Low LF attachment High LF attachment 18% 21% 14% Age 30-34 Age 35-39 Age 40-44 Age 45-49 21% 17% 14% 13% Men Women 17% 16% Single Married 19% 14% 0 kids 1 kid 2+ kids 20% 15% 13% Mortgage holder Non-mortgage-holder 13% 18% -1 -.75 -.5 -.25 0 .25 .5 .75 1 Estimated employment impact of Great Recession local shocks (pp) B. Earnings Subgroup Migration Rate Overall 16% Earnings $0 Earnings $1-$15k Earnings $15k-$45k Earnings $45k+ 16% 20% 16% 15% No LF attachment Low LF attachment High LF attachment 18% 21% 14% Age 30-34 Age 35-39 Age 40-44 Age 45-49 21% 17% 14% 13% Men Women 17% 16% Single Married 19% 14% 0 kids 1 kid 2+ kids 20% 15% 13% Mortgage holder Non-mortgage-holder 13% 18% -12 -8 -4 0 4 8 12 Estimated earnings impact of Great Recession local shocks (%) Notes: Panel A plots coefficients and 95% confidence intervals of the impact of Great Recession local shocks on 2015 relative employment—overall (equal to the 2015 data point in Figure 5A) and by subgroup. All estimates derive from the specification underlying the 2015 data point in Figure 5A. Subgroup estimates restrict the sample to the specified subgroup defined by gender, 2006 earnings, 2007 age, 2006 marital status, 2006 number of kids, or 2006 mortgage holding. Non-1040-filers are classified here as single and childless. Standard errors are clustered by 2007 state. Subgroup migration rates are superimposed on the right, where migration is defined as one’s 2015 CZ being different from one’s 2007 CZ. Panel B replicates panel A for 2015 earnings expressed in multiples of mean annual earnings 1999-2006: 2015 earnings divided by mean annual 1999-2006 earnings. This quantity is top-coded at the 99th percentile, and individuals with zero 1999-2006 earnings are assigned the top code if 2015 earnings were positive and assigned 0 otherwise. The overall estimate is -0.0355 (standard error 0.0094), implying that a 1-percentage-point-higher Great Recession local shock reduced the average individual’s 2015 earnings by 3.55% of her pre-recession earnings. 31 TABLE 1 Home Mortgage Interest Deductability at the State Level State Home Mortgage Interest Deductability Top Personal Income Tax Rate Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware D.C. Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming Full None Full Full Partial Full None Full Partial None Full Partial Full None None Full Full Full Full Full Full None None Full Full Full Full Full None None None Full Partial Full Full None Full Full None None Full None None None Partial Full Full None None Partial None 5.00% 0.00% 4.54% 7.00% 12.30% 4.63% 6.70% 6.75% 8.95% 0.00% 6.00% 11.00% 7.40% 5.00% 3.40% 8.98% 4.90% 6.00% 6.00% 8.00% 5.75% 5.25% 4.25% 7.85% 5.00% 6.00% 6.90% 6.84% 0.00% 0.00% 8.97% 4.90% 8.82% 7.75% 3.99% 5.93% 5.25% 9.90% 3.07% 5.99% 7.00% 0.00% 0.00% 0.00% 5.00% 8.95% 5.75% 0.00% 6.50% 7.75% 0.00% Notes - For each state, this table specifies whether the state allows full (i.e. equal to the federal level), partial (less than the federal level but still positive), or no deductibility of home mortgage interest. This table also lists each state's top personal income tax rate. TABLE 2 Summary Statistics Random 2% Border Sample Outcomes (in 2015) Employed (%) Earnings (2015 $) DI receipt (%) UI receipt sometime 2007-2014 (%) Personal characteristics (in 2006, 2007) Female (%) Retail Border Sample Random 2% Sample Retail Chain Sample Mass Layoffs Sample Mean (1) Std. Dev. (2) Mean (3) Std. Dev. (4) Mean (5) Std. Dev. (6) Mean (7) Std. Dev. (8) Mean (9) Std. Dev. (10) 80.0 47,089 6.7 25.7 40.0 61,180 25.0 43.7 81.9 32,253 7.2 27.9 38.5 41,901 25.9 44.9 79.1 47,587 6.2 25.6 40.7 63,784 24.2 43.6 81.8 33,381 6.9 28.3 38.5 44,557 25.3 45.0 84.1 48,204 6.0 52.2 36.5 62,830 23.8 50.0 49.4 50.0 61.2 48.7 49.3 50.0 60.8 48.8 44.5 49.7 Earnings (2015 $) 44,808 52,182 32,219 35,266 45,652 55,122 33,424 36,708 52,511 55,336 Age Aged 30-34 (%) Aged 35-39 (%) Aged 40-44 (%) Aged 45-49 (%) 40.0 21.9 24.4 26.0 27.6 5.7 41.4 43.0 43.9 44.7 39.2 26.7 25.0 24.4 24.0 5.8 44.2 43.3 42.9 42.7 39.9 22.2 24.5 26.0 27.3 5.7 41.5 43.0 43.9 44.6 39.2 27.0 25.0 24.3 23.6 5.8 44.4 43.3 42.9 42.5 39.7 23.8 25.0 25.5 25.7 5.7 42.6 43.3 43.6 43.7 22.7 14.9 62.8 41.9 35.6 48.4 15.1 24.3 60.6 35.8 42.9 48.9 10.0 16.4 73.6 29.9 37.1 44.1 No Labor Force Attachment (%) Low Labor Force Attachment (%) High Labor Force Attachment (%) Married (%) 91.4 28.0 93.2 25.2 62.8 48.3 52.2 50.0 52.9 49.9 0 kids (%) 1 kid (%) 2+ kids (%) 3954.0 5.3 12.2 4889.4 22.4 32.7 2607.5 100.0 0.0 4390.5 0.0 0.0 36.2 22.6 41.2 48.0 41.8 49.2 41.8 22.9 35.4 49.3 42.0 47.8 40.6 23.4 36.0 49.1 42.3 48.0 1040 filer (%) 25.3 43.5 0.0 0.0 91.2 28.3 93.1 25.3 93.8 24.1 Mortgage holder (%) 0.0 0.2 0.0 0.0 38.3 48.6 25.9 43.8 38.4 48.6 Retail trade (NAICS 44,45) (%) Construction/manufacturing (NAICS 23,31-33) (%) Other observed industry (%) Contractor (%) Non-employed (%) 10.9 439.1 0.0 0.0 0.0 31.1 144.6 0.0 0.0 0.0 0.0 442.1 0.0 0.0 0.0 0.0 138.9 0.0 0.0 0.0 5.2 11.9 25.9 4.2 11.5 22.3 32.4 43.8 20.1 31.9 100.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 4.7 17.7 34.9 0.0 0.0 21.2 38.2 47.7 0.0 0.0 Great Recession local shock (pp) 0.0 0.0 0.0 0.0 4.6 1.5 4.8 1.5 5.0 1.5 Number of individuals Number of 2007 CZs 233,530 115 147,334 110 1,357,974 722 865,954 655 1,001,543 668 Notes - This table lists summary statistics for the paper's five samples: the random 2% border sample, the retail border sample, the 2% random sample, the retail chain sample (all non-headquarters workers for identifiable retail chain firms in 2006), and the mass layoffs sample (all workers who separated from a firm in a 2008 or 2009 mass layoff). Earnings is the sume of W-2 wage earnings and 1099-MISC independent contractor earnings in the calendar year, in 2015 dollars and top-coded at $500,000. Employed is an indicator for having positive earnings. DI receipt is an indicator for having positive 1099-SSA disability insurance income in the calendar year. UI receipt sometime 2007-2014 is an indicator for having positive 1099-G unemployment insurance benefit income at some point 2007-2014. Age is measured on January 1, 2007. Married is an indicator for filing a married-filing-jointly or married-filing-separately 1040 for tax year 2006. Number of kids is the number of current dependent kids currently living with the worker as listed on the filed 1040. 1040 filer is an indicator for having appeared as a primary or secondary filer on a Form 1040 for tax year 2006. Displayed marriage and number of kids statistics are restricted to 1040 filers; in regressions controlling for marriage or number of kids fixed effects, non-1040-filers are included as a separate group. Mortgage holder is an indicator for having positive mortgage payment listed on a Form 1098 in 2006 (mortgages held only in the name of a worker's spouse or other third party are not included here). Industry categories are based on the North American Industrial Classification System code on the business income tax return based on matching the individual's highest-paying 2006 W2 to the universe of business returns. Almost half of W-2 earners could not be matched and individuals who only had 1099-MISC independent contractor earnings are not matched; in fixed effect regressions, unmatched 2006 W-2 earners, contractors, and the non-employed are assigned to three separate industries. 2007 CZ derives from the worker's January 2007 residential location as reflected most commonly on her 2006 information returns. The Great Recession local shock equals the 2009 unemployment rate in the individual's 2007 CZ minus the 2007 unemployment rate in that CZ as reported by the Bureau of Labor Statistics Local Area Unemployment Statistics. TABLE 3 Effect of Home Mortgage Interest Deduction on Migration Binary Indicator of Home Mortgage Interest Deductibility at the State Level in 2% Random Border Sample A. Inclusive Definition of HMI Deductibility States Outcome: 2007 state deductability of mortgage interest (pp) (1) (pp) (2) 2.425 (1.170) 2.311 (1.147) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N Indicator for individual migration between 2007 and 2014 (pp) (pp) (pp) (pp) (pp) (pp) (pp) (3) (4) (5) (6) (7) (8) (9) (pp) (10) (pp) (11) 2.311 (1.138) 2.210 (1.201) 2.218 (1.193) 2.131 (1.190) 2.169 (1.207) 2.193 (1.201) 1.739 (1.024) 2.253 (1.137) 2.379 (1.043) X X X X X X X X X X X X X X X X X R2 233,530 0.00 233,530 0.01 233,530 233,530 233,530 233,530 233,530 233,530 233,530 0.02 0.18 0.18 0.18 0.18 0.18 0.18 233,530 0.18 233,530 0.18 Migration rate (%) Estimate divided by migration rate (%) 15.1 16.07 15.1 15.31 15.1 15.31 15.1 11.52 15.1 14.93 15.1 15.76 (pp) (12) (pp) (13) Indicator for individual migration between 2007 and 2014 (pp) (pp) (pp) (pp) (pp) (pp) (pp) (14) (15) (16) (17) (18) (19) (20) (pp) (21) (pp) (22) 2.317 (1.159) 2.219 (1.136) 2.230 (1.126) B. Exclusive Definition of HMI Deductibility States Outcome: 2007 state deductability of mortgage interest Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N R2 Migration rate (%) Estimate divided by migration rate (%) 15.1 14.64 15.1 14.69 15.1 14.12 15.1 14.37 15.1 14.53 2.130 (1.181) 2.143 (1.172) 2.105 (1.177) 2.122 (1.190) 2.140 (1.185) 1.797 (0.887) 2.212 (1.091) 2.301 (1.005) X X X X X X X X X X X X X X X X X 233,530 0.00 15.1 15.35 233,530 0.01 15.1 14.70 233,530 233,530 233,530 233,530 233,530 233,530 233,530 0.02 0.18 0.18 0.18 0.18 0.18 0.18 15.1 15.1 15.1 15.1 15.1 15.1 15.1 14.77 14.11 14.20 13.94 14.06 14.18 11.90 Notes - This table estimates the effect of state deductibility of the home mortgage interest (HMI) deduction on 2007-2014 migration. Panel A categorizes state HMI deductibility inclusively: states that permit only partial deductibility are categorized as allowing deductions when defining the 2007 state deductability of mortgate interest binary indicator. Panel B categorizes state HMI deductibility exclusively: states that permit only partial deductibility are categorized as not allowing deductions. See the text for additional details. 233,530 0.18 15.1 14.66 233,530 0.18 15.1 15.25 TABLE 4 Effect of Home Mortgage Interest Deduction on Migration Continuous Measure of Home Mortgage Interest Deductibility at the State Level in 2% Random Border Sample A. Inclusive Definition of HMI Deductibility States Outcome: 2007 top rate deductability of mortgage interest (pp) (1) (pp) (2) Indicator for individual migration between 2007 and 2014 (pp) (pp) (pp) (pp) (pp) (pp) (pp) (3) (4) (5) (6) (7) (8) (9) (pp) (10) (pp) (11) 0.324 (0.148) 0.307 (0.145) 0.305 (0.144) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N 2 R Migration rate (%) Estimate divided by migration rate (%) 2 R Migration rate (%) Estimate divided by migration rate (%) 0.278 (0.152) 0.288 (0.153) 0.293 (0.152) 0.225 (0.127) 0.292 (0.153) 0.315 (0.130) X X X X X X X X X X X X X X X X 233,530 0.00 233,530 233,530 233,530 233,530 233,530 233,530 233,530 233,530 0.01 0.02 0.18 0.18 0.18 0.18 0.18 0.18 15.1 1.95 15.1 1.96 15.1 1.84 15.1 1.91 15.1 1.94 233,530 0.18 233,530 0.18 15.1 2.15 15.1 2.04 15.1 2.02 15.1 1.49 15.1 1.93 15.1 2.09 (pp) (12) (pp) (13) Indicator for individual migration between 2007 and 2014 (pp) (pp) (pp) (pp) (pp) (pp) (pp) (14) (15) (16) (17) (18) (19) (20) (pp) (21) (pp) (22) 0.321 (0.155) 0.305 (0.152) 0.306 (0.151) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N 0.295 (0.150) X B. Exclusive Definition of HMI Deductibility States Outcome: 2007 top rate deductability of mortgage interest 0.295 (0.152) 0.296 (0.158) 0.297 (0.157) 0.289 (0.160) 0.295 (0.160) 0.299 (0.159) 0.251 (0.114) 0.305 (0.155) 0.318 (0.133) X X X X X X X X X X X X X X X X X 233,530 0.00 15.1 2.13 233,530 233,530 233,530 233,530 233,530 233,530 233,530 233,530 0.01 0.02 0.18 0.18 0.18 0.18 0.18 0.18 15.1 15.1 15.1 15.1 15.1 15.1 15.1 15.1 2.02 2.03 1.96 1.97 1.91 1.96 1.98 1.66 233,530 0.18 15.1 2.02 233,530 0.18 15.1 2.10 Notes - This table replicates Table 3 except that it replaces the binary independent variable measuring HMI deductibility with a continuous measure. The continuous measure equals the binary measure times the state's top marginal income tax rate. See the text for additional details. TABLE 5 Effect of Home Mortgage Interest Deduction on Mortgage Holding Binary Indicator of Home Mortgage Interest Deductibility at the State Level in 2% Random Border Sample A. Inclusive Definition of HMI Deductibility States Outcome: 2007 state deductability of mortgage interest (pp) (1) (pp) (2) (pp) (3) Indicator for individual mortgage holding in 2006 (pp) (pp) (pp) (pp) (pp) (4) (5) (6) (7) (8) (pp) (9) (pp) (10) (pp) (11) 0.345 (0.920) 0.491 (0.906) -0.310 (1.279) -0.426 (1.213) -0.331 (0.976) -0.289 (1.175) -0.316 (1.163) 0.000 (0.000) -0.837 (0.949) -0.362 (0.933) -0.364 (0.971) X X X X X X X X X Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N 2 R Migration rate (%) Estimate divided by migration rate (%) X X X X X X X X 233,530 0.00 233,530 233,530 233,530 233,530 233,530 233,530 233,530 233,530 0.01 0.19 0.32 0.36 0.33 0.33 1.00 0.32 39.5 0.87 39.5 1.24 39.5 -0.79 (pp) (12) (pp) (13) 2007 state deductability of mortgage interest 1.149 (1.229) 1.271 (1.186) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N 2 R Migration rate (%) Estimate divided by migration rate (%) 39.5 -0.84 39.5 -0.73 39.5 -0.80 233,530 0.32 39.5 0.00 39.5 -2.12 39.5 -0.92 39.5 -0.92 (pp) (14) Indicator for individual mortgage holding in 2006 (pp) (pp) (pp) (pp) (pp) (15) (16) (17) (18) (19) (pp) (20) (pp) (21) (pp) (22) 0.364 (0.785) 0.266 (0.731) 0.418 (0.547) 0.321 (0.780) 0.301 (0.775) 0.000 (0.000) -0.059 (0.560) 0.299 (0.545) 0.422 (0.525) X X X X X X X X X B. Exclusive Definition of HMI Deductibility States Outcome: 39.5 -1.08 233,530 0.32 X X X X X X X X 233,530 0.00 39.5 2.90 233,530 233,530 233,530 233,530 233,530 233,530 233,530 233,530 0.01 0.19 0.32 0.36 0.33 0.33 1.00 0.32 39.5 39.5 39.5 39.5 39.5 39.5 39.5 39.5 3.22 0.92 0.67 1.06 0.81 0.76 0.00 -0.15 233,530 0.32 39.5 0.76 233,530 0.32 39.5 1.07 Notes - This table replicates Table 3 for the outcome of whether a worker held a mortgage in 2006, defined as the worker having received a Form 1098 in 2006. See Table 3 and the text for additional details. TABLE 6 Effect of Home Mortgage Interest Deduction on Mortgage Holding Continuous Measure of Home Mortgage Interest Deductibility at the State Level in 2% Random Border Sample A. Inclusive Definition of HMI Deductibility States Outcome: 2007 top rate deductability of mortgage interest (pp) (1) (pp) (2) (pp) (3) Indicator for individual mortgage holding in 2006 (pp) (pp) (pp) (pp) (pp) (4) (5) (6) (7) (8) (pp) (9) (pp) (10) (pp) (11) 0.015 (0.097) 0.037 (0.097) -0.027 (0.171) -0.043 (0.164) -0.036 (0.140) -0.017 (0.155) -0.025 (0.153) 0.000 (0.000) -0.113 (0.140) -0.053 (0.138) -0.027 (0.135) X X X X X X X X X Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N 2 R Migration rate (%) Estimate divided by migration rate (%) X X X X X X X X 233,530 0.00 233,530 233,530 233,530 233,530 233,530 233,530 233,530 233,530 0.01 0.19 0.32 0.36 0.33 0.33 1.00 0.32 39.5 0.04 39.5 0.09 39.5 -0.07 (pp) (12) (pp) (13) 2007 top rate deductability of mortgage interest 0.178 (0.151) 0.197 (0.145) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N 2 R Migration rate (%) Estimate divided by migration rate (%) 39.5 -0.09 39.5 -0.04 39.5 -0.06 233,530 0.32 39.5 0.00 39.5 -0.29 39.5 -0.13 39.5 -0.07 (pp) (14) Indicator for individual mortgage holding in 2006 (pp) (pp) (pp) (pp) (pp) (15) (16) (17) (18) (19) (pp) (20) (pp) (21) (pp) (22) 0.108 (0.092) 0.093 (0.088) 0.112 (0.070) 0.104 (0.093) 0.096 (0.093) 0.000 (0.000) 0.035 (0.074) 0.076 (0.074) 0.124 (0.078) X X X X X X X X X B. Exclusive Definition of HMI Deductibility States Outcome: 39.5 -0.11 233,530 0.32 X X X X X X X X 233,530 0.00 39.5 0.45 233,530 233,530 233,530 233,530 233,530 233,530 233,530 233,530 0.01 0.19 0.32 0.36 0.33 0.33 1.00 0.32 39.5 39.5 39.5 39.5 39.5 39.5 39.5 39.5 0.50 0.27 0.23 0.28 0.26 0.24 0.00 0.09 233,530 0.32 39.5 0.19 233,530 0.32 39.5 0.31 Notes - This table replicates Table 4 for the outcome of whether a worker held a mortgage in 2006, defined as the worker having received a Form 1098 in 2006. See Table 4 and the text for additional details. TABLE 7 Effect of Home Mortgage Interest Deduction on Migration Binary Indicator of Home Mortgage Interest Deductibility at the State Level in Retail Border Sample A. Inclusive Definition of HMI Deductibility States Outcome: 2007 state deductability of mortgage interest (pp) (1) (pp) (2) (pp) (3) 2.917 (0.927) 2.881 (0.924) 2.854 (0.959) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Age-Earnings-Firm FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N Indicator for individual migration between 2007 and 2014 (pp) (pp) (pp) (pp) (pp) (pp) (4) (5) (6) (7) (8) (9) (3) (4) (5) 2.884 2.589 2.587 2.445 2.449 2.537 (1.023) (0.957) (0.964) (0.990) (0.978) (0.955) (pp) (10) (pp) (11) (pp) (12) 2.475 (0.780) 2.577 (0.925) 2.782 (0.778) X X X X X X X X X X X X X X X X X X R 147,334 0.00 147,334 147,334 147,334 147,334 147,334 147,334 147,334 147,334 0.01 0.02 0.06 0.27 0.27 0.27 0.27 0.27 147,334 0.27 147,334 0.27 147,334 0.27 Migration rate (%) Estimate divided by migration rate (%) 17.6 16.56 17.6 16.35 17.6 16.20 17.6 16.37 17.6 14.69 17.6 14.68 (pp) (13) (pp) (14) (pp) (15) (pp) (16) (pp) (17) 3.319 (0.992) 3.269 (1.001) 3.222 (1.026) 3.218 (1.072) 2 17.6 13.88 17.6 13.90 17.6 14.40 17.6 14.05 17.6 14.63 17.6 15.79 (pp) (18) (pp) (19) (pp) (20) (pp) (21) (pp) (22) (pp) (23) (pp) (24) 2.837 (1.006) 2.847 (1.013) 2.731 (1.031) 2.756 (1.024) 2.813 (1.000) 2.918 (0.879) 2.928 (0.933) 3.084 (0.860) X X X X X X X X X B. Exclusive Definition of HMI Deductibility States Outcome: 2007 state deductability of mortgage interest (6) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Age-Earnings-Firm FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N R2 Migration rate (%) Estimate divided by migration rate (%) X X X X X X X X X 147,334 0.00 17.6 18.84 147,334 147,334 147,334 147,334 147,334 147,334 147,334 147,334 0.01 0.02 0.06 0.27 0.27 0.27 0.27 0.27 17.6 17.6 17.6 17.6 17.6 17.6 17.6 17.6 18.55 18.29 18.26 16.10 16.16 15.50 15.64 15.97 Notes - This table replicates Table 3 in the retail border sample. See Table 3 and the text for additional details. 147,334 0.27 17.6 16.56 147,334 0.27 17.6 16.62 147,334 0.27 17.6 17.51 TABLE 8 Effect of Home Mortgage Interest Deduction on Migration Continuous Measure of Home Mortgage Interest Deductibility at the State Level in Retail Border Sample A. Inclusive Definition of HMI Deductibility States Outcome: 2007 top rate deductability of mortgage interest (pp) (1) (pp) (2) (pp) (3) Indicator for individual migration between 2007 and 2014 (pp) (pp) (pp) (pp) (pp) (pp) (4) (5) (6) (7) (8) (9) (pp) (10) (pp) (11) (pp) (12) 0.363 (0.128) 0.357 (0.126) 0.360 (0.129) 0.370 (0.135) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Age-Earnings-Firm FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N 2 R Migration rate (%) Estimate divided by migration rate (%) 0.328 (0.127) 0.327 (0.128) 0.306 (0.134) 0.308 (0.132) 0.322 (0.128) 0.296 (0.099) 0.309 (0.132) 0.351 (0.105) X X X X X X X X X X X X X X X X X X 147,334 0.00 147,334 147,334 147,334 147,334 147,334 147,334 147,334 147,334 0.01 0.02 0.06 0.27 0.27 0.27 0.27 0.27 17.6 2.06 17.6 2.03 17.6 2.04 17.6 2.10 17.6 1.86 17.6 1.86 (pp) (13) (pp) (14) (pp) (15) (pp) (16) (pp) (17) 0.459 (0.145) 0.451 (0.146) 0.450 (0.146) 0.454 (0.149) 147,334 0.27 17.6 1.75 17.6 1.83 17.6 1.68 17.6 1.76 17.6 1.99 (pp) (18) (pp) (19) (pp) (20) (pp) (21) (pp) (22) (pp) (23) (pp) (24) 0.390 (0.145) 0.391 (0.146) 0.373 (0.151) 0.380 (0.149) 0.388 (0.145) 0.399 (0.127) 0.399 (0.145) 0.421 (0.124) X X X X X X X X X (6) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Age-Earnings-Firm FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N R2 Migration rate (%) Estimate divided by migration rate (%) 147,334 0.27 17.6 1.74 B. Exclusive Definition of HMI Deductibility States Outcome: 2007 top rate deductability of mortgage interest 147,334 0.27 X X X X X X X X X 147,334 0.00 17.6 2.61 147,334 147,334 147,334 147,334 147,334 147,334 147,334 147,334 0.01 0.02 0.06 0.27 0.27 0.27 0.27 0.27 17.6 17.6 17.6 17.6 17.6 17.6 17.6 17.6 2.56 2.55 2.58 2.21 2.22 2.12 2.15 2.20 Notes - This table replicates Table 4 in the retail border sample. See Table 4 and the text for additional details. 147,334 0.27 17.6 2.26 147,334 0.27 17.6 2.26 147,334 0.27 17.6 2.39 TABLE 9 Effect of Home Mortgage Interest Deduction on Mortgage Holding Binary Indicator of Home Mortgage Interest Deductibility at the State Level in Retail Border Sample A. Inclusive Definition of HMI Deductibility States Outcome: 2007 state deductability of mortgage interest (pp) (1) (pp) (2) (pp) (3) Indicator for individual mortgage holding in 2006 (pp) (pp) (pp) (pp) (pp) (pp) (4) (5) (6) (7) (8) (9) (pp) (10) (pp) (11) (pp) (12) -0.311 (1.450) -0.268 (1.469) -0.788 (1.636) -1.052 (1.461) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Age-Earnings-Firm FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N -1.204 (1.259) -1.219 (1.330) -0.994 (1.162) -0.948 (1.109) 0.000 (0.000) -1.412 (1.028) -1.262 (0.845) -1.243 (1.077) X X X X X X X X X X X X X X X X X X R 147,334 0.00 147,334 147,334 147,334 147,334 147,334 147,334 147,334 147,334 0.01 0.21 0.25 0.42 0.43 0.42 0.42 1.00 147,334 0.42 147,334 0.42 147,334 0.42 Migration rate (%) Estimate divided by migration rate (%) 26.1 -1.19 26.1 -1.03 26.1 -3.02 26.1 -4.04 26.1 -4.62 26.1 -4.67 (pp) (13) (pp) (14) (pp) (15) (pp) (16) (pp) (17) 0.530 (0.971) 0.586 (0.997) -0.250 (1.104) -0.495 (0.951) 2 26.1 -3.81 26.1 -3.63 26.1 0.00 26.1 -5.42 26.1 -4.84 26.1 -4.77 (pp) (18) (pp) (19) (pp) (20) (pp) (21) (pp) (22) (pp) (23) (pp) (24) -0.566 (0.793) -0.497 (0.808) -0.411 (0.753) -0.433 (0.759) 0.000 (0.000) -0.688 (0.791) -0.699 (0.522) -0.535 (0.664) X X X X X X X X X B. Exclusive Definition of HMI Deductibility States Outcome: 2007 state deductability of mortgage interest (6) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Age-Earnings-Firm FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N R2 Migration rate (%) Estimate divided by migration rate (%) X X X X X X X X X 147,334 0.00 26.1 2.03 147,334 147,334 147,334 147,334 147,334 147,334 147,334 147,334 0.01 0.21 0.25 0.42 0.43 0.42 0.42 1.00 26.1 26.1 26.1 26.1 26.1 26.1 26.1 26.1 2.25 -0.96 -1.90 -2.17 -1.91 -1.58 -1.66 0.00 Notes - This table replicates Table 5 in the retail border sample. See Table 5 and the text for additional details. 147,334 0.42 26.1 -2.64 147,334 0.42 26.1 -2.68 147,334 0.42 26.1 -2.05 TABLE 10 Effect of Home Mortgage Interest Deduction on Mortgage Holding Continuous Measure of Home Mortgage Interest Deductibility at the State Level in Retail Border Sample A. Inclusive Definition of HMI Deductibility States Outcome: 2007 top rate deductability of mortgage interest (pp) (1) (pp) (2) (pp) (3) 0.015 (0.200) 0.022 (0.201) -0.078 (0.206) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Age-Earnings-Firm FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N 2 R Migration rate (%) Estimate divided by migration rate (%) Indicator for individual mortgage holding in 2006 (pp) (pp) (pp) (pp) (pp) (pp) (4) (5) (6) (7) (8) (9) (3) (4) (5) -0.119 -0.147 -0.155 -0.115 -0.111 0.000 (0.182) (0.156) (0.167) (0.144) (0.136) (0.000) (pp) (12) -0.203 (0.142) -0.170 (0.124) -0.142 (0.132) X X X X X X X X X X X X X X X X X 147,334 0.00 147,334 147,334 147,334 147,334 147,334 147,334 147,334 147,334 0.01 0.21 0.25 0.42 0.43 0.42 0.42 1.00 26.1 0.06 26.1 0.08 26.1 -0.30 26.1 -0.46 26.1 -0.56 26.1 -0.59 (pp) (13) (pp) (14) (pp) (15) (pp) (16) (pp) (17) 0.175 (0.129) 0.184 (0.131) 0.021 (0.136) -0.019 (0.115) 147,334 0.42 147,334 0.42 147,334 0.42 26.1 -0.44 26.1 -0.43 26.1 0.00 26.1 -0.78 26.1 -0.65 26.1 -0.54 (pp) (18) (pp) (19) (pp) (20) (pp) (21) (pp) (22) (pp) (23) (pp) (24) -0.033 (0.098) -0.025 (0.100) -0.008 (0.092) -0.018 (0.094) 0.000 (0.000) -0.078 (0.113) -0.073 (0.078) -0.018 (0.099) X X X X X X X X X (6) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Age-Earnings-Firm FEs Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting N R2 Migration rate (%) Estimate divided by migration rate (%) (pp) (11) X B. Exclusive Definition of HMI Deductibility States Outcome: 2007 top rate deductability of mortgage interest (pp) (10) X X X X X X X X X 147,334 0.00 26.1 0.67 147,334 147,334 147,334 147,334 147,334 147,334 147,334 147,334 0.01 0.21 0.25 0.42 0.43 0.42 0.42 1.00 26.1 26.1 26.1 26.1 26.1 26.1 26.1 26.1 0.71 0.08 -0.07 -0.13 -0.10 -0.03 -0.07 0.00 Notes - This table replicates Table 6 in the retail border sample. See Table 6 and the text for additional details. 147,334 0.42 26.1 -0.30 147,334 0.42 26.1 -0.28 147,334 0.42 26.1 -0.07 TABLE 11 2015 Impacts of Great Recession Local Shocks A. 2015 Employment Outcome relative to pre-2007 mean: Great Recession local shock Employed in 2015 (pp) (1) (pp) (2) (pp) (3) (pp) (4) -0.412 (0.112) -0.425 (0.112) -0.417 (0.099) -0.393 (0.097) (pp) (5) Most severely shocked quintile -1.746 (0.471) Fourth shock quintile -1.144 (0.434) Third shock quintile -0.793 (0.356) Second shock quintile -0.181 (0.320) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Unemployment persistence in 2007 CZ Unemployment persistence in 2015 CZ (pp) (6) (pp) (7) -0.366 (0.089) -0.364 (0.089) X X X X X X X X N R2 (6) 0.00 (7) 0.00 (8) 0.01 1,357,974 0.07 1,357,974 0.07 1,357,974 0.07 1,357,974 0.07 Outcome mean Absolute outcome mean Std. dev. of Great Recession local shocks Interquartile range of G.R. local shocks -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.44 B. Additional Outcomes and Controls Cumulative employment 2009Outcome relative to pre-2007 mean: 2015 Earnings in 2015 Cumulative earnings 20092015 (pp) (8) ($) (9) ($) (10) (pp) (11) (pp) (12) (pp) (13) -2.700 (0.516) -997 (168) -6,212 (919) -0.364 (0.100) -0.480 (0.133) -0.378 (0.112) Rust CZ × Great Recession local shock 0.067 (0.192) -0.035 (0.148) Other CZ × Great Recession local shock 0.094 (0.250) Great Recession local shock Age-Earnings-Industry FEs Manufacturing share N R2 Outcome mean Absolute outcome mean Std. dev. of Great Recession local shocks Interquartile range of G.R. local shocks Employed in 2015 X X X X X X X 1,357,974 0.07 1,357,974 0.11 1,357,974 0.13 1,357,974 0.07 1,357,974 0.07 1,357,974 0.07 -40.5 563.9 1.49 2.31 6,249 47,587 1.49 2.31 27,646 317,011 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 Notes – All columns except column 5 report coefficient estimates of the effect of Great Recession local shocks on post-recession outcomes in the main sample. Column 5 divides indviduals into quintiles based on their Great Recession local shocks and reports coefficients on indicators of shock quintiles, relative to the least shocked quintile. Age fixed effects are birth year indicators. Earnings fixed effects are indicators for sixteen bins in the individual's 2006 earnings. Industry fixed effects are indicators for the individual's 2006 four-digit NAICS industry. Local unemployment persistence equals the 2015 LAUS unemployment rate minus the 2007 LAUS unemployment rate in either the individual's 2007 CZ or the individual's 2015 CZ. The columns-1-7 outcome is 2015 relative employment: the individual's 2015 employment (indicator for any employment in 2015) minus the individual's mean 1999-2006 employment. The column 8 outcome equals the sum of the individual's 2009-2015 employment minus seven times the individual's mean 1999-2006 employment. The column 9 outcome equals the individual's 2015 earnings minus the individual's mean 1999-2006 earnings. The column 10 outcome equals the sum of the individual's 2009-2015 earnings minus seven times the individual's mean 1999-2006 earnings. Column 11 controls for the 2000 manufacturing share of employment in the individual's 2007 CZ, computed in County Business Patterns. Columns 11 and 12 control for indicators (not shown) and interactions of a rust-CZ indicator and an other-CZ indicator, based on the individual's 2007 CZ. A rust CZ is a CZ with an above-median manufacturing share; an other CZ is a CZ with a below-median manufacturing share and an above median 2006-2009 change in housing net worth from Mian and Sufi (2014). The coefficient for column 11 is -0.358 when controlling for a quartic in the manufacturing share. The absolute outcome mean equals the outcome mean before subtracting the pre-recession mean. Standard errors are clustered by 2007 state. For reference, column 4 indicates that a 1percentage-point higher Great Recession local shock caused individuals to be 0.393 percentage points less likely to be employed in 2015. TABLE 12 2015 Impacts of Great Recession Local Shocks ‒ Retail Chain Sample A. Main Specifications Outcome relative to pre-2007 mean: Great Recession local shock Employed in 2015 (pp) (1) (pp) (2) (pp) (3) (pp) (4) (pp) (5) -0.414 (0.118) -0.426 (0.119) -0.398 (0.102) -0.407 (0.094) -0.359 (0.080) (pp) (6) Most severely shocked quintile -1.496 (0.308) Fourth shock quintile -1.416 (0.298) Third shock quintile -0.968 (0.288) Second shock quintile -0.366 (0.298) Age FEs Age-Earnings FEs Age-Earnings-Industry FEs Age-Earnings-Firm FEs Unemployment persistence in 2007 CZ Unemployment persistence in 2015 CZ (pp) (7) (pp) (8) -0.340 (0.063) -0.339 (0.061) X X X X X X X X X N R2 865,954 0.00 865,954 0.00 865,954 0.02 865,954 0.03 865,954 0.07 865,954 0.15 865,954 0.15 865,954 0.15 Outcome mean Absolute outcome mean Std. dev. of Great Recession local shocks Interquartile range of G.R. local shocks -9.80 81.8 1.49 2.31 -9.80 81.8 1.49 2.31 -9.80 81.8 1.49 2.31 -9.80 81.8 1.49 2.31 -9.80 81.8 1.49 2.31 -9.80 81.8 1.49 2.31 -9.80 81.8 1.49 2.31 -9.80 81.8 1.49 2.44 Earnings in 2015 ($) (10) Cumulative earnings 20092015 ($) (11) (pp) (12) Employed in 2015 (pp) (13) (pp) (14) -422 (134) -2,777 (606) -0.365 (0.081) -0.589 (0.111) -0.405 (0.095) Rust CZ × Great Recession local shock 0.258 0.144 0.069 0.115 Other CZ × Great Recession local shock 0.414 0.251 B. Additional Outcomes and Controls Cumulative employment 2009Outcome relative to pre-2007 mean: 2015 (pp) (9) Great Recession local shock Age-Earnings-Firm FEs Manufacturing share -2.504 (0.430) X X X X X X X N R2 865,954 0.15 865,954 0.21 865,954 0.23 865,954 0.15 865,954 0.15 865,954 0.15 Outcome mean Absolute outcome mean Std. dev. of Great Recession local shocks Interquartile range of G.R. local shocks -51.5 590.0 1.49 2.31 2,356 33,381 1.49 2.31 8,381 225,554 1.49 2.31 -9.80 81.8 1.49 2.31 -9.80 81.8 1.49 2.31 -9.80 81.8 1.49 2.31 Notes – This table replicates Table 11 in the retail chain sample. See the notes to that table for details. Firm is an indicator for the individual's 2006 firm (a retail chain firm). TABLE 13 Robustness of the 2015 Employment Impacts Outcome relative to pre-2007 mean: Great Recession local shock Main controls Gender Number of kids Married Home ownership CZ size CZ pre-2007 size growth Cross-CZ commuting Max UI duration 2007-2015 Minimum wage change 2007-2015 Employed in 2015 (pp) (1) (pp) (2) (pp) (3) (pp) (4) (pp) (5) (pp) (6) (pp) (7) (pp) (8) (pp) (9) (pp) (10) (pp) (11) (pp) (12) (pp) (13) -0.393 (0.097) -0.394 (0.096) -0.344 (0.090) -0.344 (0.093) -0.397 (0.098) -0.439 (0.093) -0.412 (0.098) -0.381 (0.095) -0.404 (0.095) -0.399 (0.096) -0.400 (0.115) -0.397 (0.129) -0.477 (0.125) X X X X X X X X X X X X X X X X X X X X X X X X X Exclude if invalid industry code Exclude if construction/manufacturing Instrumented with birth state shock N R2 Outcome mean Absolute outcome mean Std. dev. of G.R. local shocks Interquartile range of G.R. local shocks X 1,357,974 (6) 1,357,974 0.08 1,357,974 (7) 1,357,974 0.08 1,357,974 (8) 1,357,974 0.07 1,357,974 0.07 1,357,974 0.07 1,357,974 0.07 1,357,974 0.07 741,165 0.11 579,553 0.10 1,357,974 0.07 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.23 79.1 1.49 2.31 -7.20 73.9 1.49 2.31 -6.69 70.8 1.49 2.31 -7.23 79.1 1.49 2.31 Notes – This table adds controls, sample restrictions, or instruments to the specification underlying Table 11 column 4, reprinted here in column 1. Column 2 controls for the individual's gender. Column 3 controls for the individual's 2006 number of kids (fixed effets for 0, 1, or 2+ kids). Column 4 controls for the individual's 2006 marital status. Column 5 controls for individual's 2006 home ownership status. Colums 6-10 control for CZ-level characteristics. Column 6 controls for the individual's 2007 CZ's size, equal to the CZ's total employment in 2006 as reported in Census's County Business Patterns (CBP). Column 7 controls for the individual's 2007 CZ's size growth, equal to the CZ's log change in CBP employment from 2000 to 2006. Column 8 controls for the individual's 2007 CZ's share of workers who work outside of the CZ, computed from the 2006-2010 American Community Surveys. Column 9 controls for the individual's 2007 state's maximum unemployment insurance duration over years 2007-2015. Column 10 controls for the individual’s 2007 state’s 2015 minimum wage minus that state’s 2007 minimum wage. Column 11 excludes 2006 W-2 earners without an industry code and 2006 contractors and thus restricts the sample to those for whom 2006 industry is correctly measured: 2006 W-2 earners with a valid industry code and the 2006 nonemployed. Column 12 further excludes individuals employed in construction or manufacturing in 2006. Column 13 instruments the individual's Great Recession local shock using the mean of the Great Recession local shock in the individual's birth state. Standard errors are clustered by 2007 state. TABLE 14 Time Series of Adjustment Margins Employed Migrated outside 2007 CZ Employed in 2007 CZ Employed outside 2007 CZ Earnings Earnings in 2007 CZ Earnings outside 2007 CZ (pp) (1) (pp) (2) (pp) (3) (pp) (4) UI income SSDI income ($) (5) ($) (6) ($) (7) ($) (8) ($) (9) Effect in 2007 0.087 (0.106) 0.000 0.087 (0.106) 0.000 -205 (115) -205 (115) 0 14.3 (12.2) -3.8 (8.9) Effect in 2008 -0.099 (0.079) 0.036 (0.119) -0.098 (0.077) -0.001 (0.006) -508 (108) -480 (108) -28 (11) 36.3 (15.6) -3.4 (9.4) Effect in 2009 -0.349 (0.080) 0.109 (0.208) -0.321 (0.077) -0.028 (0.012) -750 (129) -687 (127) -63 (18) 94.0 (32.1) 0.5 (11.8) Effect in 2010 -0.403 (0.074) 0.209 (0.272) -0.367 (0.074) -0.037 (0.017) -807 (131) -736 (125) -71 (30) 83.9 (32.1) 2.6 (13.0) Effect in 2011 -0.387 (0.072) 0.248 (0.296) -0.339 (0.071) -0.048 (0.022) -840 (119) -745 (105) -94 (35) 43.1 (24.3) 7.5 (13.7) Effect in 2012 -0.373 (0.075) 0.244 (0.334) -0.324 (0.076) -0.049 (0.026) -890 (141) -784 (117) -106 (45) 19.9 (20.1) 9.8 (15.2) Effect in 2013 -0.434 (0.089) 0.180 (0.382) -0.365 (0.091) -0.069 (0.032) -960 (147) -810 (108) -149 (56) 7.7 (16.3) 13.0 (17.0) Effect in 2014 -0.360 (0.100) 0.134 (0.422) -0.322 (0.101) -0.038 (0.031) -968 (197) -799 (135) -169 (84) -3.0 (9.9) 15.7 (17.7) Effect in 2015 -0.393 (0.097) 0.073 (0.456) -0.338 (0.100) -0.055 (0.042) -997 (168) -786 (108) -211 (101) -5.9 (8.9) 19.6 (18.8) Outcome (relative or absolute): Notes – This table expands on the specifications of Table 11 column 4 and column 9, whose results are reprinted here in the bottom rows of columns 1 and 5. Each cell reports the coefficient on the Great Recession local shock variable from a separate regression in which the outcome uses the post-recession year indicated in the row, instead of exclusively using 2015 as in Table 2. Every regression uses the same 1,357,974 observations underlying Table 2. The column 1 outcome of relative employment is defined in Table 2, varying the postrecession year between 2007 and 2015. The column 2 outcome is an indicator for out-migration, equal to the individual's year-t CZ being different from her 2007 CZ. Columns 3 and 4 separate the column 1 outcome for year t into two outcomes: employment in year t in the individual's 2007 CZ and employment in year t outside the individual's 2007 CZ, each minus mean 1999-2006 employment. The column 5 outcome is defined in Table 2. Columns 7 and 8 separate the column 5 outcome analogously to columns 3-4. The column 8 outcome is the individual's unemployment insurance benefits in year t . The column 9 outcome is the individual's Social Security Disability Insurance benefits in year t . Standard errors are clustered by 2007 state. See Appendix Table 1 for pre-trends. TABLE 15 Additional Outcomes A. Disability Insurance Receipt in Main Sample B. Employment Impacts in Mass Layoffs Sample SSDI receipt in 2015 (pp) (1) 2015 relative employment-orSSDI-receipt (pp) (2) 2015 relative employment (pp) (3) 2015 relative employment (pp) (4) 2015 relative employment (pp) (5) Great Recession local shock 0.071 (0.145) -0.265 (0.099) -0.577 (0.126) -0.628 (0.118) -0.605 (0.125) Main controls Exclude if invalid industry code Exclude if construction/manuf. X X X X X X X X 1,357,974 0.12 1,357,974 0.09 1,001,543 0.11 573,493 0.17 396,377 0.15 6.22 6.22 -2.28 84.06 -10.12 84.12 -10.45 83.11 -9.85 83.20 UI receipt sometime 2007-2014 2015 relative employment-or-UIreceipt-sometime2007-2014 2015 relative employment Relative nonemployment 20072014 2015 relative employment 2013-2015 relative employment (pp) (6) (pp) (7) (pp) (8) (pp) (9) (pp) (10) (pp) (11) 1.431 (0.418) -0.019 (0.121) -0.354 (0.099) 0.487 (0.122) -0.057 (0.111) -0.285 (0.101) Outcome: N R2 Outcome mean Absolute outcome mean C. Layoffs and Nonemployment in Main Sample Outcome: Great Recession local shock UI receipt sometime 2007-2014 Main controls Employment 2007-2012 N 2 R Outcome mean Absolute outcome mean -2.734 (0.142) X X X X X X X 1,357,974 0.16 1,357,974 0.06 1,357,974 0.08 1,357,974 0.08 1,357,974 0.26 1,357,974 0.07 25.6 25.6 -2.8 83.6 -7.2 79.1 3.0 3.0 -7.2 79.1 -1.2 85.2 Notes – The table reports estimates of the specification in Table 11 column 4 with alternative outcomes, samples, and/or controls. Column 1 replicates the main specification using the outcome of an indicator for 2015 receipt of Social Security Disability Insurance. Column 2 replicates the main specification using the outcome of an indicator for 2015 employment or 2015 SSDI receipt, minus the individual's mean employment 1999-2006. Column 3 replicates the main specification and columns 45 replicate Table 13 columns 11-12 in the mass layoffs sample. Column 6 replicates the main specification using the outcome of an indicator for unemployment insurance (UI) benefit receipt at some point 2007-2014. Column 7 replicates column 2 but uses UI receipt 2007-2014 in place of 2015 SSDI receipt. Column 8 replicates the main specification, controlling for UI receipt sometime 2007-2014. Column 9 replicates the main specification using the outcome of an indicator for having any year of nonemployment 2007-2014, minus an indicator for having any year of nonemployment 1999-2006. Column 10 replicates the main specification while controlling for indicators of employment in each year 2007-2012. Column 11 replicates the main specification using the outcome of an indicator for employment in any year 2013-2015, minus the individual's mean employment 1999-2006. Standard errors are clustered by 2007 state. APPENDIX TABLE 1 Adjustment Margins Pre-Trends Employed Migrated outside 2000 CZ Employed in 2000 CZ Employed outside 2000 CZ Earnings Earnings in 2000 CZ Earnings outside 2000 CZ (pp) (1) (pp) (2) (pp) (3) (pp) (4) UI income SSDI income ($) (5) ($) (6) ($) (7) ($) (8) ($) (9) Effect in 2000 -0.036 (0.083) 0.000 -0.036 (0.083) 0.000 72 (66) 72 (66) 0 0.6 (6.2) 0.4 (3.9) Effect in 2001 -0.094 (0.045) -0.036 (0.166) -0.093 (0.044) -0.001 (0.003) -74 (46) -67 (42) -7 (6) 10.6 (9.2) 0.3 (4.6) Effect in 2002 -0.068 (0.018) -0.047 (0.280) -0.060 (0.017) -0.008 (0.004) -56 (26) -47 (23) -9 (5) 9.7 (15.4) 0.7 (5.4) Effect in 2003 -0.044 (0.029) -0.005 (0.361) -0.036 (0.031) -0.008 (0.007) -26 (21) -17 (20) -10 (6) 10.2 (15.9) 0.6 (5.8) Effect in 2004 0.031 (0.039) 0.077 (0.437) 0.030 (0.036) 0.001 (0.009) 1 (36) 6 (30) -5 (12) 11.0 (11.9) 0.3 (6.6) Effect in 2005 0.083 (0.061) 0.151 (0.526) 0.070 (0.053) 0.013 (0.010) 40 (66) 42 (57) -1 (22) 6.8 (11.5) -0.1 (7.3) Effect in 2006 0.228 (0.135) 0.252 (0.612) 0.189 (0.116) 0.039 (0.020) -15 (118) 1 (94) -16 (36) 9.1 (13.1) 0.4 (9.1) Outcome (relative or absolute): Notes – This table replicates Table 14 for years 2000-2006 and where each individual's Great Recession local shock equals the 2007-2009 percentage-point unemployment rate change in the individual's 2000 CZ. See the notes to Table 14 for details. APPENDIX TABLE 2 CZ-Level Means and Estimates CZ Mortgage holding rate Migration rate 2015 employment rate 100 200 301 302 401 402 500 601 602 700 800 900 1001 1002 1100 1201 1202 1203 1204 1301 1302 1400 1500 1600 1701 1702 1800 1900 2000 2100 2200 2300 2400 2500 2600 2700 2800 2900 3001 3002 3003 3101 3102 3201 3202 3203 3300 3400 3500 3600 3700 3800 3901 4001 4002 4003 4102 4103 4200 4301 4302 4401 4402 4501 4502 4601 4602 4701 4702 4800 4901 4902 34.934 34.886 27.682 38.913 39.444 29.195 38.540 37.500 32.298 35.556 37.182 42.845 38.393 36.771 40.668 33.333 28.516 35.283 31.818 29.968 32.553 35.011 36.235 34.004 41.714 30.405 32.075 38.431 39.601 33.276 33.183 37.132 40.964 37.074 29.131 35.963 27.613 33.982 27.545 26.531 37.332 32.553 33.516 35.294 27.155 30.383 36.196 33.655 37.803 30.160 31.339 31.074 32.692 32.558 31.515 27.036 28.892 30.000 38.292 30.612 37.118 31.838 35.889 25.126 24.839 37.875 34.266 24.924 29.250 23.395 36.588 35.501 11.180 17.443 9.343 15.027 15.229 15.862 14.953 16.327 22.671 16.167 16.242 17.138 25.000 16.147 17.109 23.188 20.703 19.686 21.970 14.196 15.603 23.210 19.369 13.894 16.850 15.878 19.707 23.098 22.794 21.160 16.066 15.625 14.919 23.694 15.332 18.469 17.160 21.401 19.760 20.408 13.416 16.159 20.330 17.647 17.241 20.649 14.145 12.824 13.533 15.615 16.451 11.211 14.194 14.729 15.557 20.847 22.754 17.714 16.159 23.129 23.362 10.999 17.129 15.578 10.921 12.807 10.490 14.286 17.733 23.602 19.206 18.428 75.990 73.158 65.398 77.826 79.646 74.943 79.081 70.918 77.329 74.333 76.690 80.479 74.405 72.049 79.003 71.015 74.219 75.416 72.727 71.609 75.638 74.730 76.659 79.250 80.467 78.378 77.149 77.569 81.098 76.792 78.529 81.526 81.558 81.328 79.557 74.896 73.373 74.838 76.048 70.068 79.372 77.752 73.077 70.588 64.655 74.041 76.507 73.907 78.179 75.401 77.702 74.549 79.945 75.969 78.811 77.199 75.749 79.143 78.734 78.912 72.489 65.991 73.083 61.307 57.602 74.659 67.133 75.684 75.868 74.741 76.824 77.778 2015 mean earnings Residualized shock Employment effect 33,701.93 30,354.65 26,237.57 41,178.64 41,551.73 29,813.81 38,866.46 25,289.45 27,103.95 33,756.25 36,041.41 50,136.20 36,232.13 29,371.88 38,007.98 31,773.86 28,322.27 35,401.41 30,493.29 28,407.33 31,431.28 33,138.24 38,650.96 33,887.00 51,477.81 30,266.29 34,604.89 36,087.00 43,301.13 31,021.75 28,581.81 38,249.03 48,695.93 42,796.29 29,324.41 35,246.99 33,179.74 36,895.31 28,728.07 24,490.89 40,242.86 36,948.13 31,640.96 38,192.68 23,403.13 31,708.43 44,511.75 40,941.56 44,732.08 32,976.11 39,078.48 39,676.41 36,072.76 36,195.82 36,859.53 35,118.54 34,151.31 34,774.08 43,187.26 30,270.42 33,534.48 27,415.66 32,517.84 23,662.71 22,266.31 27,777.81 25,412.81 30,100.32 31,403.92 27,854.33 33,095.92 33,433.75 5.40 7.00 5.89 5.17 5.52 7.89 6.11 7.00 5.23 6.67 8.39 6.45 5.11 7.87 8.01 6.08 5.78 5.07 6.86 7.79 6.12 4.67 5.95 6.80 4.83 6.90 4.58 5.07 3.75 5.31 4.94 4.07 4.43 3.85 5.85 2.80 3.55 2.99 7.04 4.08 2.59 3.94 3.89 3.72 3.80 3.77 2.88 2.36 2.46 2.45 2.53 2.75 3.20 3.08 2.71 2.98 2.70 2.68 2.30 4.95 2.16 4.67 4.71 3.96 3.27 5.19 5.83 4.76 3.92 3.95 6.24 9.20 -2.71 -2.82 -4.93 -1.37 0.24 -4.01 -0.64 -5.44 0.55 -5.08 -3.18 0.26 -2.94 -4.08 -1.19 -1.91 -3.25 -2.52 -3.16 -6.00 -3.14 -4.11 -0.79 -1.21 0.53 -2.04 -1.34 -1.81 -0.78 0.42 -1.64 -0.18 -0.47 -0.14 0.06 -3.05 -4.16 -3.03 -1.26 -8.58 -1.18 -1.50 -2.18 -7.74 -7.47 -4.54 -0.98 -4.10 -1.83 -1.61 -0.18 -2.25 0.80 -4.82 -1.81 -0.10 -4.99 -1.59 -2.30 0.13 -4.41 -5.26 -3.55 -7.90 -9.67 0.64 -7.31 -0.70 -2.09 -1.73 -1.83 -0.34 Percentage Earnings effect earnings effect -1591.9 -2221.3 -1881.4 -657.3 623.5 -3451.4 -1827.0 -5864.3 -1736.3 -3825.5 -2816.9 -1132.8 554.7 -1664.6 -500 [Text truncated at 120,000 characters. The full text is on the page linked above.] --- Source: Frix Law Library, https://www.frixlaw.com/law-library/documents/agency%3Airs%3A33c5f43e56522367. Public record. Not legal advice.