Home » Posts tagged 'Clustering'
Tag Archives: Clustering
Heterogeneity in Mortgage Refinancing
By Julia Wu
Abstract
Many households who would benefit from and are eligible to refinance their mortgages fail to do so. A recent literature has demonstrated a significant degree of heterogeneity in the propensity to refinance across various dimensions, yet much heterogeneity is left unexplained. In this paper, I use a clustering regression to characterize heterogeneity in mortgage refinancing by estimating the distribution of propensities to refinance. A key novelty to my approach is that I do so without relying on borrower characteristics, allowing me to recover the full degree of heterogeneity, rather than simply the extent to which the propensity to refinance varies with a given observable. I then explore the role of both observed and unobserved heterogeneity in group placement by regressing group estimates on a set of demographic characteristics. As a complement to my analysis, I provide evidence from a novel dataset of detailed information on borrower perspectives on mortgage refinancing to paint a more nuanced picture of how household characteristics and behavioral mechanisms play into the decision to refinance. I find a significant degree of heterogeneity in both the average and marginal propensity to refinance across households. While observables such as education, race and income do significantly correlate with group heterogeneity, it is clear that much heterogeneity may still be attributed to the presence of unobservable characteristics.
David Berger, Faculty Advisor
Michelle Connolly, Faculty Advisor
JEL codes: D9, E52, G21
Peer Effects & Differential Attrition: Evidence from Tennessee’s Project STAR
By Sanjay Satish
Abstract
This paper explores the effects of attrition on student development in early education.
It aims to provide evidence that student departure in elementary schools has educational
impacts on the students they leave behind. Utilizing data from Tennessee’s Project STAR
experiment, this paper aims to expand upon the literature of peer effects, as well as attrition,
in public elementary schools. It departs from previous papers by utilizing survival analysis to
determine which characteristics of students prolonged participation in the experiment. Clustering
analysis is subsequently employed to group departed students to better understand
the various channels of attrition present in STAR. It finds that students who left Project
STAR were more likely to be of lower income and lower ability than their peers. This paper
then uses these findings to estimate the peer effects of attrition on students who remained
in the experiment and undertakes a discussion of potential sources of bias in this estimation
and their effects on the explanatory power of peer effects estimates.
Professor Robert Garlick, Faculty Advisor
Professor Michelle Connolly, Faculty Advisor
JEL Classification: I, I21, I26, H4, J13
