Summary

RQ: Can a Google-search “Strategic Default Risk Index” (SDRI) predict opportunistic mortgage default?

Main empirical results

• +0.39 log-odds (≈ +47 % odds) and +0.20 pp default probability per +1 σ SDRI – applies to underwater (>100 % CLTV) high‑FICO loans
• Effect only present in non-recourse states
• SDRI tracks well strategic default rates

Identification strategy

• Treatment (shock): monthly, state-level SDRI built from 5 “walk-away” keywords
  
• Outcome: loan-level default indicator, 3.2 M Fannie Mae 30-yr FRM loan-months, 2008-13
  
• Econometric design: panel logit with loan, month & ZIP FEs
  • Triple interaction SDRI × Negative-Equity × High-FICO (defines “strategic”)
    
  • IV: population-weighted SDRI in other states to address simultaneity

My comments: clarify channels (+ think more about IV, external validity)

Summary

RQ: Can a Google-search “Strategic Default Risk Index” (SDRI) predict opportunistic mortgage default?

#1. Main comment — theoretical benchmark

When does a homeowner walk?

\[ \underbrace{-m_t}_{\text{payment}} \;+\;\beta\,\mathbb{E}_t[V^{\text{keep}}_{t+1}] \;<\; \underbrace{-\color{red}{\lambda_t}}_{\text{legal / credit}} \;-\;\underbrace{\color{orange}{\kappa_t}}_{\text{moving}} \;-\;\underbrace{\color{purple}{\psi_t}}_{\text{stigma}} \;+\;\beta\,\mathbb{E}_t[V^{\text{rent}}_{t+1}] \]

Q: What is the incremental value of SDRI beyond observables? Paper will benefit from clearer story

Competing stories (I)

• Missing fundamentals (\(m_t\), \(\mathbb{E}_t[HP]\))
  • ZIP-level HPI & borrower income unobserved → plausible
    
  • Paper: state HPI & unemployment controls don’t kill SDRI
    
  • Try: add UI claims or “housing-crash” sentiment index
• Legal-cost learning (\(\lambda_t\))
  • Some evidence that many HHs mis-perceive legislation → plausible
    • Aggregate shocks could = attention shocks, leading people to search
  • Paper: effect only in non-recourse states
  • Try: use variation in education, financial literacy, or legal knowledge (from surveys)
    • More data-intensive: exploit news shocks?

Competing stories (II)

• Stigma shift (\(\psi_t\))
  • Social norms can move: media trends, generational, immigration…
  • Paper: is this main story?
    
  • Try: could try interact SDRI with county religiosity or other proxy for social norms
• Neighbour contagion
  • Defaults cluster spatially; herd effect plausible
    
  • Paper: not tested
    
  • Try: add lagged neighbourhood defaults; if SDRI fades, contagion dominates
    • Harder: social networks data? (Bailey et al, etc.)

Further comments

#1. Identification strategy — IV

• Instrument = population-weighted SDRI in other states
  
• Potential issue: nationwide news shocks
  • National news shocks (e.g., CFPB rule, viral CNBC segment) that hit all states ­simultaneously would lead to failure of exclusion restriction
• Suggestion: Lag-structure IV – use yesterday search intensity in time-zone-shifted states à-la Choi-Varian

External validity — beyond conforming GSE loans

Sample: Fannie 30-yr FRMs → excludes jumbo / subprime.

• But some evidence from 2008 shows strategic default also common in larger non-conforming loans:
  • Survey evidence: strategic share ≈ 35 % of defaults overall (Guiso-Sapienza-Zingales 2013).
  • PSID micro (Gerardi et al. 2018): 38 % of defaults have ability-to-pay characteristics
• Suggestion: try find some data (CoreLogic ?) to capture jumbo & investor properties

Smaller points

• IV table: report exact re-sampling draws for Google Trends (replicability).
  
• Introduction needs preview of results
• Map figure: show only included states (drop D.C., territories).
  
• Minor typo p. 23 footnote 7 (dates reversed).