Research

Publications

2020 Draft - Longer version
Part of the Global Income Dynamics Project (by F. Guvenen, L. Pistaferri and G. Violante), which aims to produce a harmonized cross-country database containing detailed and relevant statistics on individual- and household-level wages, earnings, and related labor market measures. The STATA code used to produce the harmonized statistics in all countries is prepared by S. Salgado and myself.
Abstract: Using administrative data from Norway, we first present stylized facts on labor earnings dynamics between 1993 and 2017 and its heterogeneity across narrow population groups. We then investigate the parents' role in children's income dynamics—the intergenerational transmission of income dynamics. We find that children of high-income, high-wealth fathers enjoy steeper income growth over the life cycle and face more volatile but more positively skewed income changes, suggesting that they are more likely to pursue high-return, high-risk careers. Children of poorer fathers also face more volatile incomes, but theirs grow more gradually and are more left skewed. Furthermore, the income dynamics of fathers and children are strongly correlated. In particular, children of fathers with steeper life-cycle income growth, more volatile incomes, or higher downside risk also have income streams of similar properties. We also confirm that fathers' significant role in workers' income dynamics is not simply spurious because of omitted variables, such as workers' own permanent income. These findings shed new light on the determinants of intergenerational mobility.
Working paper Online appendix Downloadable data moments Parameter estimates for the income processes Replication files Bibtex file 2020 Draft
2019 Draft Downloadable data moments in 2019 draft Parameter estimates for the income processes Older Version (2016) (based on wage/salary + self employment income over 1994-2013): Data Appendix: Moments For Men Moments For WomenOlder Version (2015) (based on wage/salary income over 1978-2013): Data Appendix
Media Coverage: A nontechnical summary blog post on VoxEU Washington Post Slate The Telegraph Bloomberg/Business Bloomberg II CNBC

We study individual male earnings dynamics over the life cycle using panel data on millions of U.S. workers. Using nonparametric methods, we first show that the distribution of earnings changes exhibits substantial deviations from lognormality, such as negative skewness and very high kurtosis. Further, the extent of these non-normalities varies significantly with age and earnings level, peaking around age 50 and between the 70th and 90th percentiles of the earnings distribution. Second, we estimate nonparametric impulse response functions and find important asymmetries: Positive changes for high-income individuals are quite transitory, whereas negative ones are very persistent; the opposite is true for low-income individuals. Third, we turn to long-run outcomes and find substantial heterogeneity in the cumulative growth rates of earnings and the total number of years individuals spend nonemployed between ages 25 and 55. Finally, by targeting these rich sets of moments, we estimate stochastic processes for earnings that range from the simple to the complex. Our preferred specification features normal mixture innovations to both persistent and transitory components and includes state-dependent long-term nonemployment shocks with a realization probability that varies with age and earnings.

The Nature of Countercyclical Income Risk , Journal of Political Economy, 2014

(with F. Guvenen and J. Song)
Working Paper Downloadable data moments Code Bibtex fileWeb Appendices: Appendix A: Data (excel file) Appendix B: Sensitivity Analysis Appendix C: Parametric EstimationFEDS Notes NBER Digest Bloomberg Article Business Insider Article

This paper studies business cycle variation in individual earnings risk using administrative SSA dataset. We document two sets of results. First, contrary to past research, we find that the variance of idiosyncratic earnings shocks is not countercyclical. Instead, it is the left-skewness of shocks that is strongly countercyclical: during recessions, the upper end of the shock distribution collapses—large upward earnings movements become less likely—whereas the bottom end expands–-large drops in earnings become more likely. Thus, while the dispersion of shocks does not increase, shocks become more left skewed and, hence, risky during recessions. Second, we find that the fortunes during recessions are predictable by observable characteristics before the recession. For example, prime-age workers that enter a recession with high earnings suffer substantially less compared with those who enter with low earnings. Finally, the cyclical nature of earnings risk is dramatically different for the top 1% compared with all other individuals—even those in the top 2% to 5%.
Code & Data Working Paper with Online Appendix Bibtex file Summary in the Region Magazine Forbes Article

Wage inequality has been significantly higher in the U.S. than in continental European countries (CEU) since the 1970s. Moreover, this inequality gap has further widened during this period as the U.S. has experienced a large increase in wage inequality, whereas the CEU has seen only modest changes. This article studies the role of labour income tax policies for understanding these facts, focusing on male workers. We construct a life cycle model in which individuals decide each period whether to go to school, work, or stay non-employed. Individuals can accumulate human capital either in school or while working. Wage inequality arises from differences across individuals in their ability to learn new skills as well as from idiosyncratic shocks. Progressive taxation compresses the (after-tax) wage structure, thereby distorting the incentives to accumulate human capital, in turn reducing the cross-sectional dispersion of (before-tax) wages. Consistent with the model, we empirically document that countries with more progressive labour income tax schedules have (i) significantly lower before-tax wage inequality at different points in time and (ii) experienced a smaller rise in wage inequality since the early 1980s. We then study the calibrated model and find that these policies can account for half of the difference between the U.S. and the CEU in overall wage inequality and 84% of the difference in inequality at the upper end (log 90–50 differential). In a two-country comparison between the U.S. and Germany, the combination of skill-biased technical change and changing progressivity of tax schedules explains all the difference between the evolution of inequality in these two countries since the early 1980s.
Code & Data Working Paper Bibtex file

How does the persistence of earnings change over the life cycle? Do workers at different ages face the same variance of idiosyncratic earnings shocks? This paper proposes a novel specification for residual earnings that allows for an age profile in the persistence and variance of labor income shocks. We show that the statistical model is identified, and we estimate it using Panel Study of Income Dynamics data. We find that shocks to earnings are only moderately persistent (around 0.75) for young workers. Persistence rises with age, up to unity, until midway through life. The variance of persistent shocks exhibits a U-shaped profile over the life cycle (with a minimum of 0.01 and a maximum of 0.05). These results suggest that the standard specification in the literature (with age-invariant persistence and variance) cannot capture the earnings dynamics of young workers. We also argue that a calibrated job turnover model can account for these nonflat profiles. The key idea is that workers sort into better jobs and settle down as they age; in turn, magnitudes of wage growth rates decline, thereby decreasing the variance of shocks. Furthermore, the decline in job mobility results in higher persistence. Finally, we investigate the implications of age profiles for consumption-savings behavior. The welfare cost of idiosyncratic risk implied by the age-dependent income process is up to 1.6 percent of lifetime consumption lower compared with its age-invariant counterpart. This difference is mostly due to a higher degree of consumption insurance for young workers, for whom persistence is moderate. These results suggest that age profiles of persistence and variances should be taken into account when calibrating life-cycle models.
Code & Data Online Appendix Downloadable data in figures Bibtex file
We use administrative data from the SSA to study not just the average scarring effects but investigate how the scarring effects vary with worker characteristics, such as age, skill level, and work history. Rather than focusing on involuntary job losses during mass layoffs, we consider more broadly the long lasting effects of spending at least one year nonemployment regardless of the reasons behind it. We find that the scarring effects of nonemployment vary greatly across workers with different recent earnings. In particular, low-recent-earnings workers as well as those in the top 5 percent suffer larger scarring effects of nonemployment. Furthermore, the large losses mostly result from the higher incidence of future nonemployment for the treatment group, rather than their lower earnings conditional on working. This effect is especially strong for low-income individuals.

Working Papers

NBER EF&G SLIDES


We study the determinants of lifetime earnings (LE) inequality in the U.S. by focusing on job ladder dynamics and on-the-job learning as sources of wage growth. Using administrative data, we document that i) lower LE workers change jobs more often, which is mainly driven by nonemployment; ii) average annual earnings growth for job stayers is similar, around 2% in the bottom two-thirds of the LE distribution, whereas for job switchers it rises with LE; iii) top LE workers enjoy around 10% average earnings growth regardless of job switching. We estimate a job ladder model with on-the-job learning featuring a rich set of worker types and firm heterogeneity. We find that the vast differences across worker types in job ladder risk—job loss, job finding, and contact rates—account for 80% of wage growth differences among workers below median LE. Above the median, almost all lifetime wage growth differences are a result of Pareto-distributed learning ability. We conclude that different economic forces are driving the inequality in different parts of the LE distribution.

SLIDES


This paper examines whether nonlinear and non-Gaussian features of earnings dynamics are caused by hours or hourly wages. Our findings from the Norwegian administrative and survey data are as follows: (i) Nonlinear mean reversion in earnings is driven by the dynamics of hours worked rather than wages since wage dynamics are close to linear while negative changes to hours are transitory and positive changes are persistent. (ii) Large earnings changes are driven equally by hours and wages, whereas small changes are associated mainly with wage shocks. (iii) Both wages and hours contribute to negative skewness and high kurtosis for earnings changes, although hour-wage interactions are quantitatively more important. (iv) When considering household earnings and disposable household income, the deviations from normality are mitigated relative to individual labor earnings: changes in disposable household income are close to symmetric and less leptokurtic.
Bibtex file (An earlier version of this paper was circulated under the title “Income Differences and Health Care Expenditures over the Life Cycle.”)

This paper studies differences in health care usage and health outcomes between low- and high-income individuals. Using data from the Medical Expenditure Panel Survey (MEPS) I find that early in life the rich spend significantly more on health care, whereas from midway through life until very old age the medical spending of the poor dramatically exceeds that of the rich. In addition, low-income individuals are less likely to incur any medical expenditures in a given year, yet, when they do incur medical expenditures, the amounts are more likely to be extreme. To account for these facts, I develop and estimate a life-cycle model of two distinct types of health capital: preventive and physical. Physical health capital determines survival probabilities, whereas preventive health capital governs the endogenous distribution of shocks to physical health capital, thereby controlling the life expectancy. Moreover, I incorporate important features of the U.S. health care system such as private health insurance, Medicaid, and Medicare. In the model, optimal expected life span is longer for the rich, which can only be achieved by greater investment in preventive health capital. Therefore, as they age, their health shocks grow milder compared to those of the poor, and in turn they incur lower curative medical expenditures. Public insurance for the elderly amplifies this mechanism by hampering the incentives of the poor to invest in preventive health capital. I use the model to examine a counterfactual economy with universal health insurance in which 75% of preventive medical spending is reimbursed on top of existing coverage in the benchmark economy. My results suggest that policies encouraging the use of health care by the poor early in life produce significant welfare gains, even when fully accounting for the increase in taxes required to pay for them.

Work In Progress

Why Are the Wealthiest So Wealthy? An Empirical-Quantitative Investigation of Life-Cycle Wealth Dynamics (Draft coming soon-Available upon request!) NBER SI Slides

(with E. Halvorsen, Joachim Hubmer, and S. Salgado )
We use administrative panel data from Norway between 1993 and 2015 on wealth and income to study lifecycle wealth dynamics. We investigate the roles of labor income, inheritances, rates of returns, and savings with a focus on the wealthiest. Our empirical results reveal strong persistence of wealth at the top of the distribution. On average, the wealthiest start their lives significantly richer relative to other households in the same cohort, invest more in private equity, earn higher returns, and save at higher rates. Furthermore, the lifetime incomes of the wealthy are mainly derived from dividends and capital gains on equity, whereas for the rest of the population labor income is the main source. Inheritances and inter vivos transfers constitute a negligible fraction of lifetime resources for most households except for a few wealthy ones. The wealthiest also receive these funds from parents earlier in the life cycle and more in the form of private equity. We further zoom in on the wealthiest and find large within-group heterogeneity: A significant fraction starts relatively poor with little private equity in their portfolios but experiences rapid wealth growth early in life due to high rates of return. With these facts on hand, we then develop and estimate an overlapping generations model with rich heterogeneity in bequests, labor income, and entrepreneurial ability, which households inherit from their parents imperfectly. We find inheritance and rate of return heterogeneity, and in particular, their positive correlation to be key for understanding the top wealth inequality. Our counterfactual analysis suggests that a tax on inheritances reduces wealth inequality but has a detrimental effect on aggregate output and wages.

High Risk Workers and High Risk Firms SED Slides

(with Marlène Koffi, Sergio Salgado, and Marco Weißler )
Recent literature documented large heterogeneity in earnings dynamics individuals experience, in particular, in average income profiles and higher order moments of income shocks as well as in unemployment risk and job finding rates. Using administrative social security data from Germany, we decompose heterogeneity in earnings dynamics into observable and unobservable worker and firm components. First, we document salient features of earnings risk conditional on observable worker and firm characteristics. Next, in order to identify unobservable heterogeneity we employ machine learning algorithms to cluster workers and firms by features of their earnings dynamics. Finally, we estimate an individual income process that allows for ex-ante worker and firm heterogeneity.
We find that workers in smaller and shrinking firms experience lower earnings growth with more volatile and left skewed earnings changes as well as higher unemployment risk. When we control for unobservable worker and firm types jointly, we conclude that person effects explain majority of differences in earnings dynamics while firm effects explain relatively little. These findings have important implications for public policy such as unemployment insurance..

Revisiting Gibrat’s Legacy: An Empirical Investigation of Nonlinear Firm Dynamics

(with B. Pugsley and S. Salgado )
Using firm-level administrative panel data from the US and ten other countries, we characterize nonlinear firms' growth dynamics in terms of sales, employment, and productivity. Using nonparametric methods we first show that the distribution of firms growth exhibits substantial deviations from log normality such as negative skewness and high kurtosis. Further, the extent of these non-normalities varies significantly with age and size. Second, we estimate nonparametric impulse response functions of firm growth and find important asymmetries between positive and negative changes and small and big firms. We then investigate whether a standard model of firm dynamics featuring capital and labor adjustment costs and financial frictions can explain these empirical patterns. Our results suggest that a non-gaussian productivity process is required to capture the nonlinear and non-Gaussian features of the firms' growth dynamics.

Monetary Policy, Heterogeneity and Housing Channel

(with A. Hedlund, K. Mitman and F. Karahan )
We investigate the role of housing and mortgage debt in the transmission and effectiveness of monetary policy. First, monetary policy induced-movements in house prices translate into consumption changes because of wealth effects. Second, a contractionary monetary shock raises the cost of borrowing which reduces the demand and as a result the liquidity of the housing market, further depressing house prices and further increases the cost of borrowing. Furthermore, nominal long-term mortgage debt implies that changes in monetary policy result in redistribution between lenders and borrowers and generate cash-flow effects that are larger for borrowing constrained households. We build a heterogenous agent New Keynesian model with a frictional housing market to quantify the various mechanisms. The model is able to match the rich empirical heterogeneity in home ownership, leverage and MPC across households. In particular, our model is consistent with the significant difference in MPC between low- and high-LTV households that we document in the data. Our quantitative findings are as follows: First, we find that about 20% of the drop in aggregate consumption against a contractionary monetary shock is due to declining house prices. Second, we find asymmetric responses of the economy to shocks, with contractionary shocks yielding a larger response of all variables. Finally, we investigate how the transmission of monetary policy depends on the distribution of mortgage debt and find that monetary policy is more effective in stimulating the economy in an high-LTV environment.

Wealth Inequality under Nonlinear Income Dynamics

(with F. Guvenen)
We study the consumption savings implications of non-Gaussian nonlinear earnings dynamics. First, the distribution of earnings changes displays substantial deviations from lognormality—the standard assumption in the incomplete markets literature. In particular, earnings changes display strong negative skewness and extremely high kurtosis. Second, these non-Gaussian features vary significantly both over the life cycle and with the earnings level of individuals. Third, earnings changes have "asymmetric mean reversion:" For high income individuals positive earnings shocks are quite transitory, whereas negative shocks are very long-lasting, and vice versa for low income workers. In this paper, we study consumption-savings implications of these features of the data. For this purpose we solve and simulate a life-cycle consumption-savings model that allows for non-Gaussian income risk. The idiosyncratic income fluctuations we document generate large welfare costs, and the implications for wealth inequality and partial insurance differ from those of a Gaussian process in important ways.

Code & Data

A Rich Set of Moments on Labor Income Dynamics from the US SSA's Master Earnings File


  • Very efficient and flexible global optimization algorithm.

  • This algorithm is evolved out of Fatih Guvenen's joint projects with me and Tony Smith, Fatih Karahan, Tatjana Kleineberg, and Antoine Arnaud.

  • You can read the description of this algorithm in this paper and find its applications in my papers here, here and there.

  • You can find my version on GitHub.

  • Great care was taken to make it as compliant with Fortran 90 as possible, but there may be a couple of invocations to Fortran 95 intrinsics.

  • Parallel performance. Here is an example of the scaling performance of TikTak on an SMM estimation problem with 1200+ moments and 7 parameters (taken from Guvenen, Karahan, Ozkan, Song (2021, specification in Table IV, column 2).

This figure shows the completion time of TikTak against the number of cores used in parallel, with the condition that the objective attained is always within 1% of the single-core value. The log-log plot is almost linear from 1 to 32 cores, with a slope of -0.976 showing almost linear scaling. Doubling the number of cores from 32 to 64 yields a final objective value that exceeds the 1% threshold.

The estimation required about N=1000 restarts (or local optimizations) in the global stage, so these results suggest a heuristic: #core ≤ sqrt(N) (sqrt(1000) ≈ 32), for the number of cores that can be used in parallel with linear scaling and no performance degradation. (Although we found that in this example, using 50 cores still delivered linear scaling without slowdown, so further experimentation is recommended).

Stata Programs for the Global Income Dynamics Project

(S. Salgado )
Being prepared for a special issue of the Quantitative Economics edited by F. Guvenen, L. Pistaferri and G. Violante.
Part of the Global Income Dynamics Project, which aims to produce a harmonized cross-country database containing detailed and relevant statistics on individual- and householdlevel wages, earnings, and related labor market measures. The STATA code used to produce the harmonized statistics in all countries is prepared by S. Salgado and myself.