1{"text": "{\"genre\":\"empirical\",\"subfield\":\"financial_intermediation_banking\",\"subfield_secondary\":\"macro_finance\",\"primary_method\":\"reduced_form_regression\",\"identification\":{\"estimand\":\"Whether bank-specific CD funding dry-ups predict future bank performance conditional on public information, and whether funds shift toward high-quality banks under stress, as a test of adverse-selection versus informed-lender theories\",\"strategy\":\"Predictive panel regressions of future changes in ROA, impaired loans and size on dry-up dummies with current controls, plus event studies of stock returns, maturity shortening and rating downgrades, and probit models of reallocation\",\"key_assumptions\":[\"Future performance conditional on current public information proxies for quality unobserved by the market\",\"Impaired loans reflect loans extended before the dry-up and so are less prone to reverse causality\",\"CDs are unsecured and lenders are mostly money market funds, so asymmetric information concerns only issuer creditworthiness and liquidity hoarding by lenders is unlikely\"],\"threats_addressed\":[\"reverse causality from dry-ups to performance via fire sales or forgone lending\",\"self-fulfilling runs\",\"demand-driven switching to central bank funding\",\"heavy reliance on CD funding driving the results\"],\"causal_language_strength\":2,\"uses_causal_claim\":false},\"abstract_moves\":{\"question\":{\"present\":true,\"sentence_index\":0},\"gap\":{\"present\":false,\"sentence_index\":-1},\"method\":{\"present\":true,\"sentence_index\":0},\"results\":{\"present\":true,\"sentence_index\":1},\"magnitude\":{\"present\":false,\"sentence_index\":-1},\"contribution\":{\"present\":true,\"sentence_index\":5}},\"introduction\":{\"moves\":[{\"paragraph_index\":0,\"move\":\"motivation\"},{\"paragraph_index\":0,\"move\":\"policy_or_broader_implication\"},{\"paragraph_index\":1,\"move\":\"preview_of_approach\"},{\"paragraph_index\":1,\"move\":\"mechanism_or_theory_sketch\"},{\"paragraph_index\":1,\"move\":\"tension_or_puzzle\"},{\"paragraph_index\":2,\"move\":\"preview_of_results\"},{\"paragraph_index\":3,\"move\":\"identification_defense\"},{\"paragraph_index\":4,\"move\":\"preview_of_results\"},{\"paragraph_index\":4,\"move\":\"mechanism_or_theory_sketch\"},{\"paragraph_index\":5,\"move\":\"contribution_statement\"},{\"paragraph_index\":6,\"move\":\"contribution_statement\"},{\"paragraph_index\":6,\"move\":\"related_literature_positioning\"},{\"paragraph_index\":7,\"move\":\"related_literature_positioning\"},{\"paragraph_index\":8,\"move\":\"roadmap\"}],\"contribution_paragraph_index\":6,\"contribution_types\":[\"new_evidence\",\"new_data_or_measure\"],\"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\":\"results\"},{\"idx\":5,\"role\":\"results\"},{\"idx\":6,\"role\":\"results\"},{\"idx\":7,\"role\":\"conclusion\"},{\"idx\":8,\"role\":\"appendix\"}],\"results_presentation\":{\"economic_significance_discussed\":true,\"statistical_significance_language\":\"unclear\",\"magnitude_statements_in_intro\":false},\"robustness\":{\"location\":\"mixed\",\"types\":[\"alternative_measure\",\"subsample\",\"falsification\"]},\"tables\":{\"convention_notes\":\"Notes open by naming the equation estimated and the dependent variable. Panels split partial and full dry-ups. Timing of observability (t-1 vs t) is spelled out, along with fixed effects and event windows. Descriptive tables report pooled means and quantiles, and an appendix table defines all variables with their data sources.\",\"units_stated_in_notes\":true,\"standard_errors_description\":\"unclear\"},\"figures\":{\"uncertainty_shown\":\"unclear\"},\"conclusion\":{\"bounded_claims\":true,\"policy_implications\":true,\"open_questions\":true,\"restates_contribution\":true},\"hedging\":{\"level\":\"moderate\",\"typical_devices\":[\"consistent with\",\"suggesting\",\"we fail to find evidence\",\"cannot draw any definite conclusion\",\"unlikely to explain\"]},\"claim_evidence\":[{\"claim\":\"No market-wide freeze occurred in the European CD market from 2008 to 2014.\",\"evidence_type\":\"figure\",\"evidence_ref\":\"Figure 1; Section III.A\"},{\"claim\":\"Bank-specific dry-ups predict future declines in ROA, controlling for current performance.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table V\"},{\"claim\":\"Dry-ups predict rising impaired loans, which helps address reverse causality.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table VI\"},{\"claim\":\"Bank size and loan share stay stable after dry-ups, so fire sales are not indicated.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table VII\"},{\"claim\":\"Negative abnormal stock returns precede dry-ups.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table VIII\"},{\"claim\":\"Maturity of new CD issues shortens in the months before dry-ups.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table IX\"},{\"claim\":\"Short-term rating downgrades are followed by drops in CD funding.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table X\"},{\"claim\":\"During stress, funds are reallocated toward banks with high future performance.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table XI\"}],\"structural_summary\":\"The paper opens with the policy-relevant view that wholesale funding is fragile. It sets two classes of asymmetric-information theories against each other, which make opposite predictions about which banks lose funding, and derives testable predictions in a theory section. It then uses transaction-level CD data to document stable aggregate volume alongside bank-specific dry-ups. Dry-ups are shown to predict worse future performance, and reverse causality and run explanations are argued against. Further evidence on maturity shortening, downgrades, lack of reentry and reallocation supports informed-lender models. The conclusion bounds the regulatory and central-banking implications.\",\"writing_profile_keywords\":[\"theory-testing\",\"competing hypotheses\",\"transaction-level data\",\"predictive regressions\",\"event study\",\"reverse causality defense\",\"policy relevance\",\"banking fragility\"],\"confidence\":0.8,\"notes\":\"The empirical design is predictive rather than causal; the abstract explicitly separates predicting from causing. Standard-error treatment is not visible in the excerpts. Section V is labeled robustness by the rule, but its content is additional mechanism results, so it is annotated as results.\"}", "json": {"genre": "empirical", "subfield": "financial_intermediation_banking", "subfield_secondary": "macro_finance", "primary_method": "reduced_form_regression", "identification": {"estimand": "Whether bank-specific CD funding dry-ups predict future bank performance conditional on public information, and whether funds shift toward high-quality banks under stress, as a test of adverse-selection versus informed-lender theories", "strategy": "Predictive panel regressions of future changes in ROA, impaired loans and size on dry-up dummies with current controls, plus event studies of stock returns, maturity shortening and rating downgrades, and probit models of reallocation", "key_assumptions": ["Future performance conditional on current public information proxies for quality unobserved by the market", "Impaired loans reflect loans extended before the dry-up and so are less prone to reverse causality", "CDs are unsecured and lenders are mostly money market funds, so asymmetric information concerns only issuer creditworthiness and liquidity hoarding by lenders is unlikely"], "threats_addressed": ["reverse causality from dry-ups to performance via fire sales or forgone lending", "self-fulfilling runs", "demand-driven switching to central bank funding", "heavy reliance on CD funding driving the results"], "causal_language_strength": 2, "uses_causal_claim": false}, "abstract_moves": {"question": {"present": true, "sentence_index": 0}, "gap": {"present": false, "sentence_index": -1}, "method": {"present": true, "sentence_index": 0}, "results": {"present": true, "sentence_index": 1}, "magnitude": {"present": false, "sentence_index": -1}, "contribution": {"present": true, "sentence_index": 5}}, "introduction": {"moves": [{"paragraph_index": 0, "move": "motivation"}, {"paragraph_index": 0, "move": "policy_or_broader_implication"}, {"paragraph_index": 1, "move": "preview_of_approach"}, {"paragraph_index": 1, "move": "mechanism_or_theory_sketch"}, {"paragraph_index": 1, "move": "tension_or_puzzle"}, {"paragraph_index": 2, "move": "preview_of_results"}, {"paragraph_index": 3, "move": "identification_defense"}, {"paragraph_index": 4, "move": "preview_of_results"}, {"paragraph_index": 4, "move": "mechanism_or_theory_sketch"}, {"paragraph_index": 5, "move": "contribution_statement"}, {"paragraph_index": 6, "move": "contribution_statement"}, {"paragraph_index": 6, "move": "related_literature_positioning"}, {"paragraph_index": 7, "move": "related_literature_positioning"}, {"paragraph_index": 8, "move": "roadmap"}], "contribution_paragraph_index": 6, "contribution_types": ["new_evidence", "new_data_or_measure"], "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": "results"}, {"idx": 5, "role": "results"}, {"idx": 6, "role": "results"}, {"idx": 7, "role": "conclusion"}, {"idx": 8, "role": "appendix"}], "results_presentation": {"economic_significance_discussed": true, "statistical_significance_language": "unclear", "magnitude_statements_in_intro": false}, "robustness": {"location": "mixed", "types": ["alternative_measure", "subsample", "falsification"]}, "tables": {"convention_notes": "Notes open by naming the equation estimated and the dependent variable. Panels split partial and full dry-ups. Timing of observability (t-1 vs t) is spelled out, along with fixed effects and event windows. Descriptive tables report pooled means and quantiles, and an appendix table defines all variables with their data sources.", "units_stated_in_notes": true, "standard_errors_description": "unclear"}, "figures": {"uncertainty_shown": "unclear"}, "conclusion": {"bounded_claims": true, "policy_implications": true, "open_questions": true, "restates_contribution": true}, "hedging": {"level": "moderate", "typical_devices": ["consistent with", "suggesting", "we fail to find evidence", "cannot draw any definite conclusion", "unlikely to explain"]}, "claim_evidence": [{"claim": "No market-wide freeze occurred in the European CD market from 2008 to 2014.", "evidence_type": "figure", "evidence_ref": "Figure 1; Section III.A"}, {"claim": "Bank-specific dry-ups predict future declines in ROA, controlling for current performance.", "evidence_type": "table", "evidence_ref": "Table V"}, {"claim": "Dry-ups predict rising impaired loans, which helps address reverse causality.", "evidence_type": "table", "evidence_ref": "Table VI"}, {"claim": "Bank size and loan share stay stable after dry-ups, so fire sales are not indicated.", "evidence_type": "table", "evidence_ref": "Table VII"}, {"claim": "Negative abnormal stock returns precede dry-ups.", "evidence_type": "table", "evidence_ref": "Table VIII"}, {"claim": "Maturity of new CD issues shortens in the months before dry-ups.", "evidence_type": "table", "evidence_ref": "Table IX"}, {"claim": "Short-term rating downgrades are followed by drops in CD funding.", "evidence_type": "table", "evidence_ref": "Table X"}, {"claim": "During stress, funds are reallocated toward banks with high future performance.", "evidence_type": "table", "evidence_ref": "Table XI"}], "structural_summary": "The paper opens with the policy-relevant view that wholesale funding is fragile. It sets two classes of asymmetric-information theories against each other, which make opposite predictions about which banks lose funding, and derives testable predictions in a theory section. It then uses transaction-level CD data to document stable aggregate volume alongside bank-specific dry-ups. Dry-ups are shown to predict worse future performance, and reverse causality and run explanations are argued against. Further evidence on maturity shortening, downgrades, lack of reentry and reallocation supports informed-lender models. The conclusion bounds the regulatory and central-banking implications.", "writing_profile_keywords": ["theory-testing", "competing hypotheses", "transaction-level data", "predictive regressions", "event study", "reverse causality defense", "policy relevance", "banking fragility"], "confidence": 0.8, "notes": "The empirical design is predictive rather than causal; the abstract explicitly separates predicting from causing. Standard-error treatment is not visible in the excerpts. Section V is labeled robustness by the rule, but its content is additional mechanism results, so it is annotated as results."}, "model": "claude-opus-5-5", "stop_reason": "end_turn", "latency_s": 20.57, "usage": {"input": 8752, "output": 2321, "cache_read": 3968, "cache_write": 0}, "cost_usd": 0.082222, "request_id": "req_011CggpqwzFcFqqL9gY1AXYW", "cached": false, "ts": "2026-11-03T23:00:48+00:00", "effort": "medium"}