Working Papers

Time-Varying Risk Premia, Labor Market Dynamics, and Income Risk

with Maarten Meeuwis, Dimitris Papanikolaou, and Jonathan Rothbaum.  Revise and Resubmit at the American Economic Review. Updated December 2023

We show that time variation in risk premia leads to time-varying idiosyncratic income risk for workers. Using US administrative data on worker earnings, we show that increases in risk premia lead to lower earnings for low-wage workers; these declines are primarily driven by job separations. By contrast, productivity shocks affect the earnings mainly of highly paid workers. We build an equilibrium model of labor market search that quantitatively replicates these facts. The model generates endogenous time-varying income risk in response to changes in risk premia and matches several stylized features of the data regarding unemployment and income risk over the business cycle.

Changing income risk across the US skill distribution: Evidence from a generalized Kalman filter

with Carter Braxton, Kyle Herkenhoff, and Jonathan Rothbaum. Second round resubmission requested at the American Economic Review. Last updated in December 2023

For whom has earnings risk changed, and why? We answer these questions by combining the Kalman filter and EM-algorithm to estimate persistent and temporary earnings for every individual at every point in time. We apply our method to administrative earnings linked with survey data. We show that since the 1980s, persistent earnings risk rose by 20% for both employed and unemployed workers and the scarring effects of unemployment doubled. At the same time, temporary earnings risk declined. Using education and occupation codes, we show that rising persistent earnings risk is concentrated among high-skill workers and related to technology adoption. 

 Technology and Labor Displacement: Evidence from Linking Patents with Worker-Level Data

with Leonid Kogan, Dimitris Papanikolaou, and Bryan Seegmiller.  Updated November 2023

We develop measures of labor-saving and labor-augmenting technology exposure using textual analysis of patents and job tasks. Using US administrative data, we show that both measures negatively predict earnings growth of individual incumbent workers. While labor-saving technologies predict earnings declines and higher likelihood of job loss for all workers, labor-augmenting technologies primarily predict losses for older or highly-paid workers. However, we find positive effects of labor-augmenting technologies on occupation-level employment and wage bills. A model featuring labor-saving and labor-augmenting technologies with vintage-specific human capital quantitatively matches these patterns. We extend our analysis to predict the effect of AI on earnings.

Who Benefits from Retirement Saving Incentives in the U.S.? Evidence on Racial Gaps in Retirement Wealth Accumulation

with Taha Choukhmane, Jorge Colmenares, Cormac O'Dea, and Jonathan Rothbaum.  Updated November 2023

U.S. employers and the federal government devote more than 1.5% of GDP annually towards promoting Defined Contribution retirement saving. We study the distributional and lifetime impact of these savings incentives across racial groups using a new employer-employee linked data set covering millions of Americans. The average contribution rate of Black and Hispanic workers is roughly 40% lower than that of White workers. The rich and the children of the rich save more; racial differences in own and parental incomes account for a large share of the racial contribution gaps. Tax and employer matching subsidies further amplify these saving differences by channeling more resources to those who save more. We estimate that breaking the link between contribution choices and saving subsidies, through revenue-neutral reforms, would significantly reduce racial gaps and intergenerational persistence in wealth.

Climbing and Falling Off the Ladder: Asset Pricing Implications of Labor Market Event Risk

Job Market Paper, Revise and Resubmit at the Journal of Financial Economics. Winner of 2015 AQR Top Finance Graduate Award and 2015 Cubist Systematic Strategies Ph.D. Candidate Award for Outstanding Research

Administrative earnings data reveal that households are exposed to large, countercyclical idiosyncratic tail risks in labor earnings. I illustrate how these risks affect asset prices within an asset pricing framework with recursive preferences, heterogeneous agents and incomplete markets. Quantitatively, a model in which agents face a time-varying probability of experiencing a rare, idiosyncratic disaster, with parameters disciplined by data, matches the level and dynamics of the equity premium. Stock returns are highly informative about labor market event risk, and, consistent with model predictions, initial claims for unemployment, a proxy for labor market uncertainty, is a highly robust predictor of returns.  

Technological Innovation and Labor Income Risk

with Leonid Kogan, Dimitris Papanikolaou, and Jae Song.

Using administrative data from the United States, we document novel stylized facts regarding technological innovation and the riskiness of labor income. Higher rates of industry innovation are associated with significant increases in labor earnings for top workers. Decomposing this result, we find that own firm innovation is associated with a modest increase in the mean, but also variance, of worker earnings growth. Innovation by competing firms is related to lower, and more negatively skewed, future earnings. We construct a structural model featuring creative destruction and displacement of human capital that replicates these patterns. In the model, higher rates of innovation by competing firms increase the likelihood that both the worker and the incumbent producer are displaced. By contrast, a higher rate of innovation by the worker's own firm increases profits, but is a mixed blessing for workers, as it increases odds that the skilled worker is no longer a good match to the new technology. Estimating the parameters of the model using indirect inference, we find significant welfare losses and hedging demand against innovation shocks. Consistent with our model, we find that these left tail effects are more pronounced for process improvements, novel innovations, and are concentrated in movers rather than continuing workers.

Idiosyncratic financial risk and a reevaluation of the market risk-return tradeoff

with Sung Je Byun and Johnathan Loudis. 

A substantial amount of the variation in US stock market returns can be explained by a single dominant factor. However, market returns also have a component driven by factors which are particularly important for large stocks and unrelated to macroeconomic aggregates - an ``idiosyncratic financial factor" (IFF). We argue, both theoretically and empirically, that the IFF can contaminate tests of the risk-return tradeoff. We then reevaluate the current consensus for a weak market risk-return tradeoff in the US stock market using an alternative index unaffected by the IFF. In the time series, we find stronger evidence for a relation between the risk premium and variance of our alternative market index. In the cross-section, our index generates larger cross-sectional variation in market betas, and these exposures explain a much larger share of variation in expected returns. Finally, in line with our theory, using our index eliminates the ability of size factors to improve pricing within a large set of standard factor models. 

How Do Health Insurance Costs Affect Firm Labor Composition and Technology Investment?

with Janet Gao, Shan Ge, and Cristina Tello-Trillo

Employer-sponsored health insurance is a significant component of labor costs. We examine the causal effect of health insurance premiums on firms’ employment, both in terms of quantity and composition, and their technology investment decisions. To address endogeneity concerns, we instrument for insurance premiums using idiosyncratic variation in insurers' recent losses, which is plausibly exogenous to their customers who are employers. Using Census microdata, we show that following an increase in premiums, firms reduce employment. Relative to higher-income coworkers, lower-income workers see a larger increase in their likelihood of being separated from their jobs and becoming unemployed. Firms also invest more in information technology, potentially to substitute labor. 

National Experimental Wellbeing Statistics

with Adam Bee, Joshua Mitchell, Nikolas Mittag, Jonathan Rothbaum, Carl Sanders, and Matthew Unrath

Accurately measuring household income and poverty is essential to understanding the nation’s overall economic wellbeing. Many studies show that measurement error stemming from unit nonresponse, item nonresponse and misreporting biases key official statistics such as mean or median income and the official poverty rate. The direction of bias differs between these sources of measurement error. Since these error components are typically studied in isolation, their overall impact on the accuracy of survey estimates remains unclear. This paper summarizes the National Experimental Wellbeing Statistics (NEWS) Project, which integrates this research and address each of these sources of bias simultaneously in order to produce more accurate estimates of household income and poverty. The NEWS project makes three unique contributions. First, we address as many sources of measurement error as we can simultaneously – including unit and item nonresponse and underreporting in surveys as well as the various challenges in administrative data such as measurement error, conceptual misalignment, and incomplete coverage. Second, we bring together all of the available survey and administrative data, which allows to address many of the shortcomings of individual data sources. Third, we propose a model to combine survey and administrative earnings data given measurement error in both sources, replacing ad hoc assumptions that have been used in prior work.

Quantile spacings: an interpretable and scalable approach to distributional modeling

with Yinchu Zhu, Walter P. Heller Memorial Award Winner.  New draft coming soon!

We propose a simple alternative to linear-in-parameters quantile regressions for the modeling of conditional distributions. We parameterize the conditional quantile function in terms of a single "location" quantile (usually the median), to which we add or subtract sums of exponentially affine functions (quantile spacings) to obtain a finite number of other quantiles. Our generalized location-scale specification preserves the computational tractability of standard linear quantile regression, is not subject to the quantile crossing problem, and the separability restriction we impose on scale can be motivated by the non-parametric generalization of differences-in-differences of Athey and Imbens (2006). Thus, under some assumptions, an application of our method extends nonlinear differences-in- differences to allow for many, potentially continuous covariates. We illustrate the utility of the method by considering impacts of mass layoffs on the distribution of displaced workers’ earnings using employer-employee matched data from the US. We find that average effects, which are in line with established literature, are driven by a substantial fattening of the left tail, a phenomenon which is further exacerbated during macroeconomic downturns, suggesting that the welfare costs of cyclical variation in income losses from job displacement are even higher than considering average effects alone.

 Click here for sample R code which uses our spacing estimator

Above link includes slides with discussion of D-i-D and new Census results. Click here for a link to the older draft


Selling Fast and Buying Slow: Heuristics and Trading Performance of Institutional Investors

with Klakow Akepanidtaworn, Rick Di Mascio, and Alex Imas. Journal of Finance, August 2023

Are market experts prone to heuristics, and if so, do they transfer across closely related domains---buying and selling? We investigate this question using a unique dataset of institutional investors with portfolios averaging $573 million. A striking finding emerges: while there is clear evidence of skill in buying, selling decisions underperform substantially---even relative to random selling strategies. This holds despite the similarity between the two decisions in frequency, substance and consequences for performance. Evidence suggests that an asymmetric allocation of cognitive resources such as attention can explain the discrepancy: we document a systematic, costly heuristic process when selling but not when buying.

Pockets of Predictability

with Leland Farmer and Allan Timmermann, Journal of Finance, April 2023

For many benchmark predictor variables, short-horizon return predictability in the U.S. stock market is local in time as short periods with significant predictability (‘pockets’) are interspersed with long periods with little or no evidence of return predictability. We document this result empirically using a flexible time-varying parameter model which estimates predictive coefficients as a nonparametric function of time and explore possible explanations of this finding, including time-varying risk-premia for which we only find limited support. Conversely, pockets of return predictability are consistent with a sticky expectations model in which investors only slowly update their beliefs about a persistent component in in the cash flow process. 

Note: A minor coding error impacted some of the results using the original method in the paper. In this note, we show that a simple adjustment to the estimation procedure restores the key results of the published paper. 

Robust Comparative Statics for the Elasticity of Intertemporal Substitution

with Joel Flynn and Alexis Toda, Theoretical Economics, January 2023

We study a general class of consumption-savings problems with recursive preferences. We characterize the sign of the consumption response to arbitrary shocks in terms of the product of two sufficient statistics: the elasticity of intertemporal substitution between contemporaneous consumption and continuation utility (EIS), and the relative elasticity of the marginal value of wealth (REMV). Under homotheticity, the REMV always equals one, so the propensity of the agent to save or dissave is always signed by the relationship of the EIS with unity. We apply our results to derive comparative statics in classical problems of portfolio allocation, consumption-savings with income risk, and entrepreneurial investment. Our results suggest empirical identification strategies for both the value of the EIS and its relationship with unity.

Measuring Document Similarity with Weighted Averages of Word Embeddings

with Dimitris Papanikolaou and Bryan Seegmiller, Explorations in Economic History, January 2023

We detail a methodology for estimating the textual similarity between two documents while accounting for the possibility that two different words can have a similar meaning. We illustrate the method's usefulness in facilitating comparisons between documents with very different formats and vocabularies by textually linking occupation task and industry output descriptions with related technologies as described in patent texts; we also examine economic applications of the resultant document similarity measures. In a final application we demonstrate that the method also works well relative to alternatives for comparing documents within the same domain by showing that pairwise textual similarity between occupations' task descriptions strongly predicts the probability that a given worker will transition from one occupation to another. Finally, we offer some suggestions on other potential uses and guidance in implementing the method.

Working Remotely and the Supply-Side Impacts of COVID-19

with Dimitris Papanikolaou.  Review of Asset Pricing Studies, March 2022

We analyze the supply-side disruptions associated with Covid-19 across firms and workers. To do so, we exploit differences in the ability of workers across industries to work remotely using data from the American Time Use Survey (ATUS). We find that sectors in which a higher fraction of the workforce is not able to work remotely experienced significantly greater declines in employment, significantly more reductions in expected revenue growth, worse stock market performance, and higher expected likelihood of default. In terms of individual employment outcomes, lower-paid workers, especially female workers with young children, were significantly more affected by these disruptions. Last, we combine these ex-ante heterogeneous industry exposures with daily financial market data to create a stock return portfolio that most closely replicates the supply-side disruptions resulting from the pandemic. 

Click here for additional data and resources related to the paper

Investor Information Acquisition and Money Market Fund Risk Rebalancing During the 2011-12 Eurozone Crisis

with Emily Gallagher, Allan Timmermann, and Russ Wermers, Review of Financial Studies, July 2019.

We study investor redemptions and portfolio rebalancing decisions of prime money market mutual funds (MMFs) during the Eurozone crisis. We find evidence that investors selectively acquire and act upon information about MMFs' risk exposures. In turn, this provides strong incentives for managers to withdraw funding from issuers whose debt becomes information-sensitive. Consistent with this, we show that MMF managers, particularly those serving the most sophisticated investors, selectively adjust their portfolio risk exposures to avoid information-sensitive European risks, while maintaining or increasing risk exposures to other regions. This mechanism helps to explain the occurrence of selective dry-ups in debt markets where delegation is common and returns to information production are often low.

Runs on Money Market Mutual Funds

with Allan Timmermann and Russ Wermers, American Economic Review, September 2016

We study daily money market mutual fund flows at the individual share class level during September 2008. This fine granularity of data facilitates new insights into investor and portfolio holding characteristics conducive to run risk in cash-like asset pools. Empirically, we find that cross-sectional flow data observed during the week of the Lehman failure are consistent with key implications of a simple model of coordination with incomplete information and strategic complementarities. Similar conclusions follow from daily models fitted to capture dynamic interactions between investors with differing levels of sophistication within the same money fund, holding constant the underlying portfolio. 

An Empirical Test of Pricing Kernel Monotonicity

with Brendan Beare, Journal of Applied Econometrics, March 2016

A large class of asset pricing models predicts that securities which have high payoffs when market returns are low tend to be more valuable than those with high payoffs when market returns are high. More generally, we expect the projection of the stochastic discount factor on the market portfolio--that is, the discounted pricing kernel evaluated at the market portfolio--to be a monotonically decreasing function of the market portfolio. Numerous recent empirical studies appear to contradict this prediction. The nonmonotonicity of empirical pricing kernel estimates has become known as the pricing kernel puzzle. In this paper we propose and apply a formal statistical test of pricing kernel monotonicity. We apply the test using seventeen years of data from the market for European put and call options written on the S&P 500 index. Statistically significant violations of pricing kernel monotonicity occur in a substantial proportion of months, suggesting that observed nonmonotonicities are unlikely to be the product of statistical noise. 

On The Dimensionality of Bounds Generated by the Shapley-Folkman Theorem

Journal of Mathematical Economics, January 2012

The Shapley-Folkman Theorem places a scalar upper bound on the distance between a sum of non-convex sets and its convex hull. We observe that some information is lost when a vector is converted to a scalar to generate this bound and propose a simple normalization of the underlying space which removes this loss of information. As an example, we apply this result to the Anderson (1978) core convergence theorem, and demonstrate how our normalization leads to an intuitive, unitless upper bound on the discrepancy between an arbitrary core allocation and the corresponding competitive equilibrium allocation.

Older Working Papers

Layoff Risk, the Welfare Cost of Business Cycles, and Monetary Policy

with David Berger, Ian Dew-Becker, and Yuta Takahashi

The strongest predictor of changes in the Fed Funds rate in the period 1982–2008 was the layoff rate. That fact is puzzling from the perspective of representative-agent models of the economy, which imply that the welfare gains of stabilizing employment fluctuations are small. This paper augments a standard New Keynesian model with a labor market featuring countercyclical layoffs that lead to large,uninsurable, and permanent idiosyncratic wage declines. In our benchmark calibration, welfare may be increased by 1 percent of lifetime consumption or more when the central bank’s policy rule responds to the layoff rate instead of purely targeting inflation.


"Heterogeneity and Asset Prices: A Different Approach"

by Nicolae Gârleanu and Stavros Panageas, Asset Pricing Meeting, NBER Summer Institute,  July 2018

"Does Precautionary Savings Drive the Real Interest Rate? Evidence from the Stock Market"

by Carolin Pflueger, Emil Siriwardane, and Adi Sunderam, NBER Asset Pricing Meeting,  November 2017

"The Tail that Wags the Economy: Beliefs and Persistent Stagnation"

by Julian Kozlowski, Laura Veldkamp, and Venky Venkateswaran, ASSA meetings, Midwest Finance Association,  March 2017

"Self-Fulfilling Runs: Evidence from the U.S. Life Insurance Market"

by Nathan Foley-Fisher, Borghan Narajabad, and Stéphane Verani, ASSA meetings, International Banking, Economics and Finance Association session on "Measuring and Managing Financial Stability",  January 2016