1{"text": "{\"genre\":\"empirical\",\"subfield\":\"household_finance\",\"subfield_secondary\":\"real estate / mortgage default\",\"primary_method\":\"structural_estimation\",\"identification\":{\"estimand\":\"Distribution across borrowers of walk-away (threshold) negative equity / CLTV levels at which ruthless default occurs\",\"strategy\":\"Two steps: a discrete-time logit hazard model of default on flexible CLTV dummies, exploiting ZIP-level variation in the timing and size of house price declines among zero-down-payment borrowers; then maximum likelihood estimation of a gamma distribution of default costs, with defaults probabilistically classified as liquidity-driven or ruthless.\",\"key_assumptions\":[\"Default arises only from unwillingness to pay (an equity response) or inability to pay (liquidity)\",\"The effect of equity on default can be separated from liquidity shocks once loan-age fixed effects, quarter fixed effects and county economic conditions are controlled\",\"ZIP-code HPI accurately measures individual home equity\",\"Borrowers walk away when the financial benefits exceed the costs\"],\"threats_addressed\":[\"liquidity/income shocks confounding default timing\",\"endogenous down payment choice (all borrowers start with CLTV of 100)\",\"lender recourse\",\"measurement error in house values\",\"non-representativeness of the nonprime sample (external validity via LPS sample)\"],\"causal_language_strength\":3,\"uses_causal_claim\":true},\"abstract_moves\":{\"question\":{\"present\":true,\"sentence_index\":1},\"gap\":{\"present\":false,\"sentence_index\":-1},\"method\":{\"present\":true,\"sentence_index\":1},\"results\":{\"present\":true,\"sentence_index\":1},\"magnitude\":{\"present\":false,\"sentence_index\":-1},\"contribution\":{\"present\":true,\"sentence_index\":4}},\"introduction\":{\"moves\":[{\"paragraph_index\":0,\"move\":\"motivation\"},{\"paragraph_index\":0,\"move\":\"preview_of_approach\"},{\"paragraph_index\":1,\"move\":\"preview_of_results\"},{\"paragraph_index\":1,\"move\":\"literature_gap\"},{\"paragraph_index\":1,\"move\":\"contribution_statement\"},{\"paragraph_index\":2,\"move\":\"related_literature_positioning\"},{\"paragraph_index\":2,\"move\":\"tension_or_puzzle\"},{\"paragraph_index\":3,\"move\":\"policy_or_broader_implication\"},{\"paragraph_index\":4,\"move\":\"preview_of_approach\"},{\"paragraph_index\":5,\"move\":\"robustness_mention\"},{\"paragraph_index\":6,\"move\":\"identification_defense\"},{\"paragraph_index\":7,\"move\":\"mechanism_or_theory_sketch\"},{\"paragraph_index\":8,\"move\":\"policy_or_broader_implication\"},{\"paragraph_index\":9,\"move\":\"roadmap\"}],\"contribution_paragraph_index\":1,\"contribution_types\":[\"new_evidence\",\"new_method\"],\"roadmap_present\":true,\"literature_positioning_location\":\"inside_introduction\"},\"section_roles\":[{\"idx\":0,\"role\":\"introduction\"},{\"idx\":1,\"role\":\"theory\"},{\"idx\":2,\"role\":\"data\"},{\"idx\":3,\"role\":\"results\"},{\"idx\":4,\"role\":\"robustness\"},{\"idx\":5,\"role\":\"conclusion\"}],\"results_presentation\":{\"economic_significance_discussed\":true,\"statistical_significance_language\":\"unclear\",\"magnitude_statements_in_intro\":true},\"robustness\":{\"location\":\"mixed\",\"types\":[\"alternative_specification\",\"subsample\",\"other\"]},\"tables\":{\"convention_notes\":\"Notes describe the estimating equation, the number of observations and the fixed effects that are included but not shown. The logit table reports coefficients, standard errors and odds ratios. Walk-away distributions are reported as CLTV percentiles (25th, median, 75th) with N, and the baseline row is repeated across the mechanism, robustness and broader-sample tables.\",\"units_stated_in_notes\":true,\"standard_errors_description\":\"unclear\"},\"figures\":{\"uncertainty_shown\":\"unclear\"},\"conclusion\":{\"bounded_claims\":true,\"policy_implications\":false,\"open_questions\":true,\"restates_contribution\":true},\"hedging\":{\"level\":\"moderate\",\"typical_devices\":[\"suggests/appears\",\"probably\",\"we argue\",\"may be\",\"it is difficult to pin down\"]},\"claim_evidence\":[{\"claim\":\"Median borrower does not walk away until CLTV reaches about 174 (equity of -74%)\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table III; Figure 7\"},{\"claim\":\"CLTV alone does not drive default until it exceeds roughly 120\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table II; Figure 6\"},{\"claim\":\"Frictionless option model predicts certain default at about -20% equity\",\"evidence_type\":\"prior_literature\",\"evidence_ref\":\"Kau, Keenan, and Kim (1994)\"},{\"claim\":\"Lender recourse is not the main driver of high walk-away thresholds\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table IV\"},{\"claim\":\"High thresholds also hold in a broader sample including GSE and portfolio loans\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table VI\"},{\"claim\":\"House prices fell sharply in the four studied states\",\"evidence_type\":\"figure\",\"evidence_ref\":\"Figures 1-2\"}],\"structural_summary\":\"The introduction moves from the crisis facts to the ruthless-default debate, then previews the headline magnitude against model predictions. It defends a two-step method that separates liquidity-driven from ruthless defaults and closes with policy implications on moral hazard. A short conceptual section defines ruthless default, followed by the data section. The empirical section estimates a hazard model and then a maximum likelihood distribution of walk-away thresholds. A long discussion section weighs financial versus behavioral mechanisms, identification, external validity and robustness. The conclusion restates the findings and leaves the behavioral drivers to future research.\",\"writing_profile_keywords\":[\"walk-away threshold\",\"negative equity\",\"hazard model\",\"maximum likelihood\",\"behavioral explanation\",\"policy implication\",\"mechanism discussion\"],\"confidence\":0.8,\"notes\":\"Primary method could also be classed as reduced_form_regression, since step 1 is a logit hazard; it was labelled structural because step 2 estimates a parametric default-cost distribution by maximum likelihood. The Discussion section's role_rule says conclusion, but its content is mechanisms and robustness, so it was assigned robustness. The way statistical significance is reported is not visible in the excerpts.\"}", "json": {"genre": "empirical", "subfield": "household_finance", "subfield_secondary": "real estate / mortgage default", "primary_method": "structural_estimation", "identification": {"estimand": "Distribution across borrowers of walk-away (threshold) negative equity / CLTV levels at which ruthless default occurs", "strategy": "Two steps: a discrete-time logit hazard model of default on flexible CLTV dummies, exploiting ZIP-level variation in the timing and size of house price declines among zero-down-payment borrowers; then maximum likelihood estimation of a gamma distribution of default costs, with defaults probabilistically classified as liquidity-driven or ruthless.", "key_assumptions": ["Default arises only from unwillingness to pay (an equity response) or inability to pay (liquidity)", "The effect of equity on default can be separated from liquidity shocks once loan-age fixed effects, quarter fixed effects and county economic conditions are controlled", "ZIP-code HPI accurately measures individual home equity", "Borrowers walk away when the financial benefits exceed the costs"], "threats_addressed": ["liquidity/income shocks confounding default timing", "endogenous down payment choice (all borrowers start with CLTV of 100)", "lender recourse", "measurement error in house values", "non-representativeness of the nonprime sample (external validity via LPS sample)"], "causal_language_strength": 3, "uses_causal_claim": true}, "abstract_moves": {"question": {"present": true, "sentence_index": 1}, "gap": {"present": false, "sentence_index": -1}, "method": {"present": true, "sentence_index": 1}, "results": {"present": true, "sentence_index": 1}, "magnitude": {"present": false, "sentence_index": -1}, "contribution": {"present": true, "sentence_index": 4}}, "introduction": {"moves": [{"paragraph_index": 0, "move": "motivation"}, {"paragraph_index": 0, "move": "preview_of_approach"}, {"paragraph_index": 1, "move": "preview_of_results"}, {"paragraph_index": 1, "move": "literature_gap"}, {"paragraph_index": 1, "move": "contribution_statement"}, {"paragraph_index": 2, "move": "related_literature_positioning"}, {"paragraph_index": 2, "move": "tension_or_puzzle"}, {"paragraph_index": 3, "move": "policy_or_broader_implication"}, {"paragraph_index": 4, "move": "preview_of_approach"}, {"paragraph_index": 5, "move": "robustness_mention"}, {"paragraph_index": 6, "move": "identification_defense"}, {"paragraph_index": 7, "move": "mechanism_or_theory_sketch"}, {"paragraph_index": 8, "move": "policy_or_broader_implication"}, {"paragraph_index": 9, "move": "roadmap"}], "contribution_paragraph_index": 1, "contribution_types": ["new_evidence", "new_method"], "roadmap_present": true, "literature_positioning_location": "inside_introduction"}, "section_roles": [{"idx": 0, "role": "introduction"}, {"idx": 1, "role": "theory"}, {"idx": 2, "role": "data"}, {"idx": 3, "role": "results"}, {"idx": 4, "role": "robustness"}, {"idx": 5, "role": "conclusion"}], "results_presentation": {"economic_significance_discussed": true, "statistical_significance_language": "unclear", "magnitude_statements_in_intro": true}, "robustness": {"location": "mixed", "types": ["alternative_specification", "subsample", "other"]}, "tables": {"convention_notes": "Notes describe the estimating equation, the number of observations and the fixed effects that are included but not shown. The logit table reports coefficients, standard errors and odds ratios. Walk-away distributions are reported as CLTV percentiles (25th, median, 75th) with N, and the baseline row is repeated across the mechanism, robustness and broader-sample tables.", "units_stated_in_notes": true, "standard_errors_description": "unclear"}, "figures": {"uncertainty_shown": "unclear"}, "conclusion": {"bounded_claims": true, "policy_implications": false, "open_questions": true, "restates_contribution": true}, "hedging": {"level": "moderate", "typical_devices": ["suggests/appears", "probably", "we argue", "may be", "it is difficult to pin down"]}, "claim_evidence": [{"claim": "Median borrower does not walk away until CLTV reaches about 174 (equity of -74%)", "evidence_type": "table", "evidence_ref": "Table III; Figure 7"}, {"claim": "CLTV alone does not drive default until it exceeds roughly 120", "evidence_type": "table", "evidence_ref": "Table II; Figure 6"}, {"claim": "Frictionless option model predicts certain default at about -20% equity", "evidence_type": "prior_literature", "evidence_ref": "Kau, Keenan, and Kim (1994)"}, {"claim": "Lender recourse is not the main driver of high walk-away thresholds", "evidence_type": "table", "evidence_ref": "Table IV"}, {"claim": "High thresholds also hold in a broader sample including GSE and portfolio loans", "evidence_type": "table", "evidence_ref": "Table VI"}, {"claim": "House prices fell sharply in the four studied states", "evidence_type": "figure", "evidence_ref": "Figures 1-2"}], "structural_summary": "The introduction moves from the crisis facts to the ruthless-default debate, then previews the headline magnitude against model predictions. It defends a two-step method that separates liquidity-driven from ruthless defaults and closes with policy implications on moral hazard. A short conceptual section defines ruthless default, followed by the data section. The empirical section estimates a hazard model and then a maximum likelihood distribution of walk-away thresholds. A long discussion section weighs financial versus behavioral mechanisms, identification, external validity and robustness. The conclusion restates the findings and leaves the behavioral drivers to future research.", "writing_profile_keywords": ["walk-away threshold", "negative equity", "hazard model", "maximum likelihood", "behavioral explanation", "policy implication", "mechanism discussion"], "confidence": 0.8, "notes": "Primary method could also be classed as reduced_form_regression, since step 1 is a logit hazard; it was labelled structural because step 2 estimates a parametric default-cost distribution by maximum likelihood. The Discussion section's role_rule says conclusion, but its content is mechanisms and robustness, so it was assigned robustness. The way statistical significance is reported is not visible in the excerpts."}, "model": "claude-opus-5-5", "stop_reason": "end_turn", "latency_s": 18.09, "usage": {"input": 7965, "output": 2234, "cache_read": 3968, "cache_write": 0}, "cost_usd": 0.077334, "request_id": "req_011CggpjbeyRwnQVvhd3o8eJ", "cached": false, "ts": "2026-11-03T22:59:19+00:00", "effort": "medium"}