Research2

Price Discrimination and Adverse Selection in the U.S. Mortgage Market

I document substantial price dispersion in the US mortgage market, with each given lender charging observably similar borrowers different rates. The welfare effects of conditional price dispersion are theoretically ambiguous: while individualized pricing can enhance competition and thus lower equilibrium prices, it may also harm certain consumers who are price discriminated against. To explain this conditional price dispersion, I build and estimate a structural model that features borrowers who choose a lender to finance a house, differentiated lenders who learn risk types about the borrowers, and flexible correlation between price signals, borrower shocks, and cost shocks. A novel component of the model is the way it captures adverse selection through the correlation between demand cost shocks and hence incorporating the interaction between price discrimination and adverse selection. With the model, I structurally decompose the conditional price dispersion into risk-based adjustments and non-risk-based price discrimination. I simulate equilibrium market outcomes under various counterfactual scenarios including varying the extent of price dispersion and information asymmetry. Preliminary results suggest that without information asymmetry, consumers may benefit from a certain degree of individualized pricing because it drives down equilibrium prices through competition. The effect is heterogeneous across income groups with the low- and moderate-income group benefiting more.

Cross-subsidies through Loan Guarantee: A Study of Shadow Bank Lending in the U.S. Mortgage Market

In recent decades, there has been a shift in the US mortgage market from traditional banks' balance sheet lending to shadow banks' originate-to-distribute lending. To evaluate the effect of this change in the mortgage market landscape, I show evidence of cross-subsidies: shadow banks charge lower interest rates compared to banks to observably similar borrowers and the difference is larger for the low income than the high income. Moreover, shadow banks lend disproportionately more to regions with lower average income and higher minority populations. To explain why shadow banks can offer lower prices, I study the loan guarantee scheme by government-sponsored entities (GSEs). I exploit a regression discontinuity design at the eligibility cutoff for GSEs' loan guarantees and find that interest rates jump discontinuously at the eligibility cutoff for nonbanks but not banks. I estimate a structural model of mortgage demand and lender competition in the US mortgage market. The results suggest that shadow banks charge similar mark-ups to banks, have lower marginal costs, and hence offer lower prices than banks across all income groups. Shadow banks' costs remain lower than banks across all income groups with a slightly higher gap for the lower income. I conclude by presenting suggestive evidence that shadow banks benefit from a lighter regulatory burden than banks so that they can exploit the loan guarantee scheme to sustain lower costs.

A Retrospective Analysis of U.S. Bank Mergers (with Tim Lee)

Anecdotal evidence suggests the antitrust agencies approved several "anti-competitive" bank mergers during the 2008 Financial Crisis to prevent the financial market from collapsing further. Using data from the Summary of Deposits (SOD) combined with mergers identification data, we revisit the bank mergers during the Financial Crisis. We begin by classifying bank mergers into normal and anti-competitive mergers using a threshold set by the Department of Justice (DOJ) and verify the leniency of antitrust enforcement during the Financial Crisis. We further exploit an event study model to quantify the impact of these mergers on equilibrium market outcomes including deposit amounts, deposit rates, and bank branch closure. Preliminary results suggest that mergers lead to fewer branches and lower deposit amounts, but no changes in deposit rates.

Pre-doctoral Writings

"Social Origin, Gender Differences, and Intergenerational Education Mobility in China," (April 2018)

Children of well-educated parents are often well-educated. This intergenerational association can be explained by selection (genetic transmission) or causation (education alters one's type, leading to higher education in offspring), with the latter being more important for policies. However, distinguishing the two channels is challenging due to endogeneity concerns. I estimate the causal intergenerational education mobility (IEM) for the peasant's social class in China using a natural experiment that exogenously affected people's educational attainment by their social origin and gender: the Maoist education reform in the 1950s. The Mao government implemented several targeted programs to boost the educational level of the poor peasants. A differences-in-differences (DID) estimator suggests that the reform increases peasants' education by 1.2 years on average relative to other social classes. The results are heterogeneous by gender with female peasants gaining 1.7 years of education and male peasants gaining 0.6 years. I further estimate that about 0.5 years of gain in education is transmitted to the next generation. Using the DID estimator as an instrumental variable for parents' education, I find that the causal IEM for the peasant group is 0.4 - 0.6 which is greater than the OLS estimate of 0.2 - 0.3 for the general population during the early 2000s. This suggests a greater pass-through of educational attainments to the next generation for the peasant group compared to the general population. Moreover, intergenerational education mobility has decreased in the post-reform period compared to the pre-reform period. I investigate the transition matrices to show that the reduction in educational mobility comes from the top and bottom education quintiles, which suggests an increase in educational disparity post-reform in China.

Resting Papers

"Dynamics of Consumer Payment Choices"

Using a panel data set from the Survey of Consumer Payment Choices, this paper studies the effects of uncertainty and learning on the adoption and usage of payment instruments in the US. Consumers are heterogeneous in their `match' values with alternative payment instruments. Understanding how quickly consumers learn about their true match values is important for banks and card networks in designing card reward policies. I begin by presenting stylized facts that consumers exhibit substantial switching in the adoption of alternative payment methods over the years and they use card payments more frequently as time passes since adoption. Motivated by the stylized facts, I develop a dynamic Bayesian learning model for consumer payment choice where consumers learn from experience about the usage values of alternative payment instruments. The structural estimates suggest that the adoption costs for all types of cards are positive with credit cards having greater costs than debit and prepaid cards, whereas the termination costs are heterogeneous across card types with only the credit cards having positive costs. The learning costs for credit cards than other types of card payments. This difference in costs could be due to stricter credit requirements for credit cards than other card types.