{"text": "{\"genre\":\"empirical\",\"subfield\":\"asset_pricing\",\"subfield_secondary\":\"behavioral_finance\",\"primary_method\":\"reduced_form_regression\",\"identification\":{\"estimand\":\"Difference in anomaly (long-minus-short) daily returns on firm-specific information days (earnings announcements, Dow Jones news days) versus noninformation days.\",\"strategy\":\"Panel regressions of daily stock returns on an aggregate anomaly measure interacted with information-day dummies, with time fixed effects. Competing explanations (dynamic risk, biased expectations, data mining) are distinguished through dynamic beta controls, analyst forecast errors, controls for contemporaneous monthly returns, and out-of-sample anomalies.\",\"key_assumptions\":[\"Time fixed effects absorb daily variation in risk premia\",\"The aggregate anomaly factor adequately captures time-varying beta exposure\",\"Out-of-sample anomalies are unlikely to be products of data mining\",\"Analyst forecast errors are not subject to the joint-hypothesis problem\"],\"threats_addressed\":[\"time-varying risk premia\",\"time-varying betas\",\"day-of-the-week effects\",\"macroeconomic news days\",\"reverse causality\",\"high-volatility days\",\"data mining / spurious anomalies\"],\"causal_language_strength\":2,\"uses_causal_claim\":false},\"abstract_moves\":{\"question\":{\"present\":true,\"sentence_index\":1},\"gap\":{\"present\":false,\"sentence_index\":-1},\"method\":{\"present\":true,\"sentence_index\":2},\"results\":{\"present\":true,\"sentence_index\":3},\"magnitude\":{\"present\":true,\"sentence_index\":0},\"contribution\":{\"present\":true,\"sentence_index\":2}},\"introduction\":{\"moves\":[{\"paragraph_index\":0,\"move\":\"motivation\"},{\"paragraph_index\":0,\"move\":\"tension_or_puzzle\"},{\"paragraph_index\":1,\"move\":\"mechanism_or_theory_sketch\"},{\"paragraph_index\":2,\"move\":\"mechanism_or_theory_sketch\"},{\"paragraph_index\":2,\"move\":\"preview_of_approach\"},{\"paragraph_index\":2,\"move\":\"preview_of_results\"},{\"paragraph_index\":3,\"move\":\"mechanism_or_theory_sketch\"},{\"paragraph_index\":4,\"move\":\"preview_of_results\"},{\"paragraph_index\":5,\"move\":\"preview_of_results\"},{\"paragraph_index\":6,\"move\":\"identification_defense\"},{\"paragraph_index\":7,\"move\":\"mechanism_or_theory_sketch\"},{\"paragraph_index\":7,\"move\":\"preview_of_results\"},{\"paragraph_index\":8,\"move\":\"identification_defense\"},{\"paragraph_index\":8,\"move\":\"preview_of_results\"},{\"paragraph_index\":9,\"move\":\"related_literature_positioning\"},{\"paragraph_index\":10,\"move\":\"contribution_statement\"},{\"paragraph_index\":11,\"move\":\"contribution_statement\"},{\"paragraph_index\":12,\"move\":\"contribution_statement\"},{\"paragraph_index\":13,\"move\":\"roadmap\"}],\"contribution_paragraph_index\":10,\"contribution_types\":[\"new_evidence\",\"new_identification\"],\"roadmap_present\":true,\"literature_positioning_location\":\"inside_introduction\"},\"section_roles\":[{\"idx\":0,\"role\":\"introduction\"},{\"idx\":1,\"role\":\"data\"},{\"idx\":2,\"role\":\"results\"},{\"idx\":3,\"role\":\"results\"},{\"idx\":4,\"role\":\"results\"},{\"idx\":5,\"role\":\"results\"},{\"idx\":6,\"role\":\"conclusion\"}],\"results_presentation\":{\"economic_significance_discussed\":true,\"statistical_significance_language\":\"unclear\",\"magnitude_statements_in_intro\":true},\"robustness\":{\"location\":\"within_results\",\"types\":[\"alternative_specification\",\"subsample\",\"falsification\",\"other\"]},\"tables\":{\"convention_notes\":\"Table notes state the regression specification, the unit of observation (firm-day), and that the dependent variable is scaled by 100. They describe sample filters (e.g., a $5 price screen) and how anomaly portfolios are built from quintile sorts. Control-variable coefficients are left unreported. In the visible excerpts, the notes do not describe standard errors.\",\"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\":[\"most consistent with\",\"could be explained by\",\"it is unlikely that\",\"at least partly\",\"likely reflect\"]},\"claim_evidence\":[{\"claim\":\"Anomaly returns are about 50% higher on news days and six times higher on earnings announcement days.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table III\"},{\"claim\":\"Information-day effects appear on both the long and short sides of anomalies.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table IV\"},{\"claim\":\"The effect holds across the four anomaly categories.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table V\"},{\"claim\":\"The effects are not explained by day-of-week effects or macroeconomic news days.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table VI\"},{\"claim\":\"Dynamic betas rise on earnings days but do not explain the earnings-day anomaly effect.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table VII\"},{\"claim\":\"Analyst forecasts are too optimistic for anomaly-shorts and too pessimistic for anomaly-longs.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table VIII\"},{\"claim\":\"Out-of-sample anomalies still show information-day effects, making data mining unlikely as the explanation.\",\"evidence_type\":\"table\",\"evidence_ref\":\"Table IX, Table X\"},{\"claim\":\"Anomaly returns are concentrated within the window around earnings announcements.\",\"evidence_type\":\"figure\",\"evidence_ref\":\"Figure 1\"}],\"structural_summary\":\"The paper opens with a long-standing puzzle and three competing explanations for it: risk, mispricing, and data mining. It previews tests that discriminate among them and situates its contributions inside the introduction. The body first establishes a stylized fact, information-day anomaly returns, and runs falsification checks on it. Separate sections then test each explanation in turn. The conclusion favors biased expectations while keeping the dynamic-risk alternative open.\",\"writing_profile_keywords\":[\"horse-race of explanations\",\"stylized facts first\",\"anomaly portfolios\",\"information days\",\"analyst forecast errors\",\"data mining tests\",\"hedged mispricing conclusion\"],\"confidence\":0.85,\"notes\":\"The Internet Appendix contains a simple representative-agent model. Statistical significance conventions are not visible in the table-note excerpts.\"}", "json": {"genre": "empirical", "subfield": "asset_pricing", "subfield_secondary": "behavioral_finance", "primary_method": "reduced_form_regression", "identification": {"estimand": "Difference in anomaly (long-minus-short) daily returns on firm-specific information days (earnings announcements, Dow Jones news days) versus noninformation days.", "strategy": "Panel regressions of daily stock returns on an aggregate anomaly measure interacted with information-day dummies, with time fixed effects. Competing explanations (dynamic risk, biased expectations, data mining) are distinguished through dynamic beta controls, analyst forecast errors, controls for contemporaneous monthly returns, and out-of-sample anomalies.", "key_assumptions": ["Time fixed effects absorb daily variation in risk premia", "The aggregate anomaly factor adequately captures time-varying beta exposure", "Out-of-sample anomalies are unlikely to be products of data mining", "Analyst forecast errors are not subject to the joint-hypothesis problem"], "threats_addressed": ["time-varying risk premia", "time-varying betas", "day-of-the-week effects", "macroeconomic news days", "reverse causality", "high-volatility days", "data mining / spurious anomalies"], "causal_language_strength": 2, "uses_causal_claim": false}, "abstract_moves": {"question": {"present": true, "sentence_index": 1}, "gap": {"present": false, "sentence_index": -1}, "method": {"present": true, "sentence_index": 2}, "results": {"present": true, "sentence_index": 3}, "magnitude": {"present": true, "sentence_index": 0}, "contribution": {"present": true, "sentence_index": 2}}, "introduction": {"moves": [{"paragraph_index": 0, "move": "motivation"}, {"paragraph_index": 0, "move": "tension_or_puzzle"}, {"paragraph_index": 1, "move": "mechanism_or_theory_sketch"}, {"paragraph_index": 2, "move": "mechanism_or_theory_sketch"}, {"paragraph_index": 2, "move": "preview_of_approach"}, {"paragraph_index": 2, "move": "preview_of_results"}, {"paragraph_index": 3, "move": "mechanism_or_theory_sketch"}, {"paragraph_index": 4, "move": "preview_of_results"}, {"paragraph_index": 5, "move": "preview_of_results"}, {"paragraph_index": 6, "move": "identification_defense"}, {"paragraph_index": 7, "move": "mechanism_or_theory_sketch"}, {"paragraph_index": 7, "move": "preview_of_results"}, {"paragraph_index": 8, "move": "identification_defense"}, {"paragraph_index": 8, "move": "preview_of_results"}, {"paragraph_index": 9, "move": "related_literature_positioning"}, {"paragraph_index": 10, "move": "contribution_statement"}, {"paragraph_index": 11, "move": "contribution_statement"}, {"paragraph_index": 12, "move": "contribution_statement"}, {"paragraph_index": 13, "move": "roadmap"}], "contribution_paragraph_index": 10, "contribution_types": ["new_evidence", "new_identification"], "roadmap_present": true, "literature_positioning_location": "inside_introduction"}, "section_roles": [{"idx": 0, "role": "introduction"}, {"idx": 1, "role": "data"}, {"idx": 2, "role": "results"}, {"idx": 3, "role": "results"}, {"idx": 4, "role": "results"}, {"idx": 5, "role": "results"}, {"idx": 6, "role": "conclusion"}], "results_presentation": {"economic_significance_discussed": true, "statistical_significance_language": "unclear", "magnitude_statements_in_intro": true}, "robustness": {"location": "within_results", "types": ["alternative_specification", "subsample", "falsification", "other"]}, "tables": {"convention_notes": "Table notes state the regression specification, the unit of observation (firm-day), and that the dependent variable is scaled by 100. They describe sample filters (e.g., a $5 price screen) and how anomaly portfolios are built from quintile sorts. Control-variable coefficients are left unreported. In the visible excerpts, the notes do not describe standard errors.", "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": ["most consistent with", "could be explained by", "it is unlikely that", "at least partly", "likely reflect"]}, "claim_evidence": [{"claim": "Anomaly returns are about 50% higher on news days and six times higher on earnings announcement days.", "evidence_type": "table", "evidence_ref": "Table III"}, {"claim": "Information-day effects appear on both the long and short sides of anomalies.", "evidence_type": "table", "evidence_ref": "Table IV"}, {"claim": "The effect holds across the four anomaly categories.", "evidence_type": "table", "evidence_ref": "Table V"}, {"claim": "The effects are not explained by day-of-week effects or macroeconomic news days.", "evidence_type": "table", "evidence_ref": "Table VI"}, {"claim": "Dynamic betas rise on earnings days but do not explain the earnings-day anomaly effect.", "evidence_type": "table", "evidence_ref": "Table VII"}, {"claim": "Analyst forecasts are too optimistic for anomaly-shorts and too pessimistic for anomaly-longs.", "evidence_type": "table", "evidence_ref": "Table VIII"}, {"claim": "Out-of-sample anomalies still show information-day effects, making data mining unlikely as the explanation.", "evidence_type": "table", "evidence_ref": "Table IX, Table X"}, {"claim": "Anomaly returns are concentrated within the window around earnings announcements.", "evidence_type": "figure", "evidence_ref": "Figure 1"}], "structural_summary": "The paper opens with a long-standing puzzle and three competing explanations for it: risk, mispricing, and data mining. It previews tests that discriminate among them and situates its contributions inside the introduction. The body first establishes a stylized fact, information-day anomaly returns, and runs falsification checks on it. Separate sections then test each explanation in turn. The conclusion favors biased expectations while keeping the dynamic-risk alternative open.", "writing_profile_keywords": ["horse-race of explanations", "stylized facts first", "anomaly portfolios", "information days", "analyst forecast errors", "data mining tests", "hedged mispricing conclusion"], "confidence": 0.85, "notes": "The Internet Appendix contains a simple representative-agent model. Statistical significance conventions are not visible in the table-note excerpts."}, "model": "claude-opus-5-5", "stop_reason": "end_turn", "latency_s": 17.53, "usage": {"input": 8184, "output": 2331, "cache_read": 3968, "cache_write": 0}, "cost_usd": 0.08015, "request_id": "req_011Cggq2wa1WmUbXuGjYDRrA", "cached": false, "ts": "2026-11-03T23:03:14+00:00", "effort": "medium"}