« Back to Results
Hilton Atlanta, 314
Hosted By:
Society of Government Economists
In this paper we apply supervised machine-learning techniques to consumer-credit-report data and investigate the joint decision of consumer-loan borrowing and performance among different debt, including student loans, mortgage, credit card and auto debt. Specifically, we develop a machine learning model of multi-period competing non-performance risk and use it to examine households’ borrowing decisions incorporating labor income, house price, inflation, interest rate risk as well as the dynamics of various debt patterns and delinquency status. We implement the model with deep neural nets, in which the first-stage network prediction of student loan borrowing is augmented with instrument variables such as the opportunity cost measured by wage and share of low-skill sector jobs. Using simulations based on machine-learning forecasts, we examine the impact of exogenous shocks to local economic conditions on consumer loan debt and performance. Moreover, the time-series patterns of estimated default or delinquency rates from this model over the course of the Great Recession suggest that aggregated household borrowing behavior may have important applications in forecasting systemic risk.
The Haves and the Have Nots - Access, Opportunity, and Outcomes in Households, Businesses, and Life
Paper Session
Friday, Jan. 4, 2019 8:00 AM - 10:00 AM
- Chair: Austin Nichols, Abt Associates
Rural Hospital Closure Impacts on Mortality Rates
Abstract
Between January 2010 and July 2017, a total of 79 rural hospitals closed. The current rate of hospital closure has far surpassed the rate of hospital closures in the late 1980s and early 1990s. When a rural hospital closes it has immediate impacts on local employment, but it also creates a ripple effect that can be felt throughout the local community. Several studies (Wisner et al, 2015; Sorensen, 2008; Holmes et al, 2006) have stated that rural hospital closures affect unemployment rates, per capita incomes, and outmigration, Other studies (Burkey et al, 2017; Buchmueller et al, 2004; Rosenback and Dayhoff, 1995) examine effects of hosptial closures on geographical access to care. While there is a study in Canada (Lie et al, 2001) that takes mortality into account when examining rural hospitals in Saskatchewan, to the best of our knowledge no study has estimated the impact of the recent rural hospital closures on local mortality rates in the Untied States using a spatio-temporal difference-in-difference framework similar to Dubé et al (2017). We will to account for the time-varying aspect of the rural hospital closures, as well as, the spatial spillovers of the hospital closure on the neighboring county’s mortality rates. Data on hospital closures between 2010-2016 comes from a list we compiled and confirmed with the UNC Sheps Center. We also use the CDC National Center for Health Statistics’ Multiple Cause of Death files from 2009-2016. Our research findings should indicate a larger effect of hospital closures on rural mortality in more isolated isolated rural counties where the sole hospital closed. The effect should be minimal in rural counties that still had an operating hospital despite one hospital closing, as well as, in rural counties located adjacent to large urban counties.Remote Competition and Small Business Loans: Evidence from SBA Lending
Abstract
This paper examines the impact of entry by remote, specialized lenders in the market for Small Business Administration (SBA) guaranteed loans. Using data on all SBA loans from 2001-2017, we document an increase in remote lending, defined as lending to borrowers more than 100 miles away. Additionally, remote lenders tend to have portfolios that are more concentrated by industry and, consistent with building industry expertise, concentrated lenders have lower charge-off rates. To investigate the competitive effects, we then examine a case study of the entry a large, remote, specialized lender into specific industries. Exploiting their staggered entry into these industries, we find that entry generates significant growth in lending, with little evidence of substitution away from incumbent SBA lenders.Understanding Consumer Loan Performance Through Machine Learning
Abstract
The American aggregate household debt has recently surpassed the previous peak during the Great Recession. The share of housing debt, however, has significantly declined while the share of student loan and auto loan debt has increased. While mortgage delinquencies has declined, student loan performance has deteriorated. The relatively easy access to credit helps student-loan borrowers build credit for access to loans later in life while improving career and earning prospects. But some borrowers may fail to graduate or make enough money to pay back the loans. Debt position in earlier stage of life can have implications on borrowers’ financial decisions later in life.In this paper we apply supervised machine-learning techniques to consumer-credit-report data and investigate the joint decision of consumer-loan borrowing and performance among different debt, including student loans, mortgage, credit card and auto debt. Specifically, we develop a machine learning model of multi-period competing non-performance risk and use it to examine households’ borrowing decisions incorporating labor income, house price, inflation, interest rate risk as well as the dynamics of various debt patterns and delinquency status. We implement the model with deep neural nets, in which the first-stage network prediction of student loan borrowing is augmented with instrument variables such as the opportunity cost measured by wage and share of low-skill sector jobs. Using simulations based on machine-learning forecasts, we examine the impact of exogenous shocks to local economic conditions on consumer loan debt and performance. Moreover, the time-series patterns of estimated default or delinquency rates from this model over the course of the Great Recession suggest that aggregated household borrowing behavior may have important applications in forecasting systemic risk.
Discussant(s)
David Johnson
,
University of Michigan
Alice Zawacki
,
U.S. Census Bureau
Robert De Young
,
University of Kansas
Meta Brown
,
State University of New York-Stony Brook
JEL Classifications
- I0 - General
- G0 - General