Diamond et al. (2020)
The economics of Housing & Homelessness
This is an augmented presentation of the original paper where I add both my own thoughts and ideas from other articles in addition to the paper under discussion. For a clear understanding of the original paper, please see the original paper.
Overview
Financial Distress
Financial Distress
Non-financial Distress
Landlord
Homeowner
Renter
Impact of Foreclosure
- "homeowners move to worse neighborhoods in terms of average income (16% lower after 5 years) and middle school test scores"
- "Over 5 years, owners are nearly 7 percentage points more likely to get divorced."
"Our most intriguing finding from the credit report data is that foreclosure causes additional financial distress."
Let's Discuss
Background
Which Foreclosure Is More Important?
"It turns out that foreclosure in the administrative judicial records and foreclosure in the credit report data are only loosely correlated."
"This may be because while payments and debt amounts are automatically filled from lenders computer systems into credit reports, foreclosures require manual entry on the part of lenders and thus may have more error. This is why many papers that use credit report data to study foreclosure use delinquency measures rather than the foreclosure flag."
Delinquency
Bank Foreclosure
Judge's Decision to Foreclose
Possible Setups
Impact of Judicial Foreclosure on Credit Score
Analysis
Instrumental Variables
Selection-on-Observables
Judge-IV
- "The key ingredient in this research design is that examiners with different tendencies will expose individuals to different treatments or interventions even when the individuals assigned to examiners are similar" - Chyn et al. (2023)
Concerns
- Monotonicity
- Multiple Treatments
- Clustering
Judges
Immediate
Outcomes
Downstream
Outcomes
Household
IV v.s. PSM
Context
- Authors believe IV is more credible than selection on observables
- IV captures LATE while PSM capture ATE
Proposal
- Difference in IV and PSM reflect treatment heterogeneity
Why this is a bad idea
- If this was true, you wouldn't need IV
Does Judge-IV Capture the relevant population in this setting?
- "Weighted average represents causal effects for individuals who would have received a different treatment status if they had been assigned to a different examiner." - Chyn et al. (2023)
"The average foreclosure, however, is in a worse neighborhood and does not have far to fall on the neighborhood quality ladder – and in fact may actually improve neighborhoods after moving out due to regression to the mean."
Let's Discuss
Yes ... but there's an uncertainty that the standard errors of the estimator don't capture
"We next turn to the landlords, for whom we mainly use PSM because of wide confidence intervals for IV."
"The confidence intervals in the reader's mind are generally larger than those reported under the coefficient in a regression table" - Anonymous*
Further Discussion
- I am not clear on the difference between (8a) and (8b)
- Why do the selected subsamples vary with the outcome?
Results
Re-read the interpretation - What do you think?
Comments/Thoughts
These are intended to be helpful to students reading the paper
"We ask whether existing estimates of the social costs of foreclosure are incomplete because they neglect significant non-pecuniary costs."
Comment
- This isn't a "real" question. The paper has already mentioned that prior work focuses only on the financial costs which is certainly an incomplete measure of the social costs.
Suggestion
- Don't phrase it this way. Be more direct. Say that in this paper we focus on the non-pecuniary costs of foreclosure.
"Second, our findings about treatment effect heterogeneity and the significant gaps between marginal and average households implies that different policies may have different implications."
Comment
- This sentence says nothing
Suggestion
- Cut the sentence
"For all owners, foreclosure causes housing instability, reduced homeownership, and financial distress including increased delinquency on other debts."
Comment
- These aren't results so much as they are a check on the quality of the data.
Suggestion
- Don't frame these as results. Use them as a check on the data, and dive deep on why there are no apparent non-financial costs for landlords and why there are limited negative outcomes for renters. That's interesting!
"The identifying assumption of the PSM approach is that (1) conditional on the propensity score fixed effects, the treatment and control groups have parallel trends"
Comment
- Is it a Diff-in-Diff strategy of a Selection on Observables?
"The main disadvantage of the PSM approach is that it assumes selection on the observables"
"In the judge IV context, montonicity is a concern if, for instance, minority judges are more lenient for minority defendants and non-minority judges are more lenient for non-minority defendants. We evaluate the monotonicity assumption extensively in Appendix B.2 and find no evidence of a violation of monotonicity."
Comment
- I don't think "find[ing] no evidence of a violation of monotonicity" is very informative.
- In practice, there is almost surely a violation of the monotonicty assumption.
- The aim should be to try and understand how the violation of monotonicity affects the interpretation