1{2 "validation_errors": [],3 "doc_id": "jof_12890",4 "ontology_version": "1.0",5 "prompt_version": "1.0",6 "model": "claude-opus-5-5",7 "effort": "medium",8 "annotated_at": "2026-11-03T23:13:39+00:00",9 "source_sha256": "33cdec6cb70e1a8fcb35d89148ffd73b76e2cb68fe65fac496744e91af6d82cb",10 "parser": "pymupdf-jof-heuristics-1.3",11 "payload_chars": 26661,12 "usage": {13 "input": 8884,14 "output": 2261,15 "cache_read": 3968,16 "cache_write": 017 },18 "cost_usd": 0.08155,19 "request_id": "req_011CggqpomzBCxHbu6L25jrB",20 "annotation": {21 "genre": "empirical",22 "subfield": "asset_pricing",23 "subfield_secondary": "financial_intermediation_banking",24 "primary_method": "structural_estimation",25 "identification": {26 "estimand": "The fraction of analysts (and brokerages) whose true recommendation-revision abnormal returns are nonzero or high-type, and the distribution of the magnitude of that skill.",27 "strategy": "Model the cross-section of estimated analyst abnormal returns as a finite mixture of skill distributions that separates true skill dispersion from estimation error. Fit constrained and unconstrained variants. Split revisions by whether a RavenPack news event occurs concurrently. Let mixing probabilities depend on analyst characteristics through a logistic function.",28 "key_assumptions": [29 "True skill within each component follows a parametric distribution (normal; lognormal as an alternative).",30 "Estimation error variance is known or estimable per analyst.",31 "Buy-and-hold abnormal returns properly benchmark recommendation revision performance."32 ],33 "threats_addressed": [34 "luck versus skill (estimation noise)",35 "piggybacking on concurrent news versus new information production",36 "negative true ability under the normal assumption (lognormal variant)",37 "selection of the number of components"38 ],39 "causal_language_strength": 2,40 "uses_causal_claim": false41 },42 "abstract_moves": {43 "question": {44 "present": false,45 "sentence_index": -146 },47 "gap": {48 "present": false,49 "sentence_index": -150 },51 "method": {52 "present": true,53 "sentence_index": 054 },55 "results": {56 "present": true,57 "sentence_index": 058 },59 "magnitude": {60 "present": false,61 "sentence_index": -162 },63 "contribution": {64 "present": false,65 "sentence_index": -166 }67 },68 "introduction": {69 "moves": [70 {71 "paragraph_index": 0,72 "move": "motivation"73 },74 {75 "paragraph_index": 0,76 "move": "related_literature_positioning"77 },78 {79 "paragraph_index": 1,80 "move": "literature_gap"81 },82 {83 "paragraph_index": 1,84 "move": "policy_or_broader_implication"85 },86 {87 "paragraph_index": 1,88 "move": "preview_of_approach"89 },90 {91 "paragraph_index": 2,92 "move": "tension_or_puzzle"93 },94 {95 "paragraph_index": 2,96 "move": "preview_of_results"97 },98 {99 "paragraph_index": 3,100 "move": "preview_of_results"101 },102 {103 "paragraph_index": 4,104 "move": "preview_of_results"105 },106 {107 "paragraph_index": 5,108 "move": "preview_of_results"109 },110 {111 "paragraph_index": 6,112 "move": "contribution_statement"113 },114 {115 "paragraph_index": 6,116 "move": "related_literature_positioning"117 },118 {119 "paragraph_index": 7,120 "move": "related_literature_positioning"121 },122 {123 "paragraph_index": 7,124 "move": "mechanism_or_theory_sketch"125 },126 {127 "paragraph_index": 8,128 "move": "roadmap"129 }130 ],131 "contribution_paragraph_index": 6,132 "contribution_types": [133 "new_evidence"134 ],135 "roadmap_present": true,136 "literature_positioning_location": "inside_introduction"137 },138 "section_roles": [139 {140 "idx": 0,141 "role": "introduction"142 },143 {144 "idx": 1,145 "role": "empirical_design"146 },147 {148 "idx": 2,149 "role": "data"150 },151 {152 "idx": 3,153 "role": "results"154 },155 {156 "idx": 4,157 "role": "results"158 },159 {160 "idx": 5,161 "role": "results"162 },163 {164 "idx": 6,165 "role": "conclusion"166 },167 {168 "idx": 7,169 "role": "appendix"170 }171 ],172 "results_presentation": {173 "economic_significance_discussed": true,174 "statistical_significance_language": "unclear",175 "magnitude_statements_in_intro": true176 },177 "robustness": {178 "location": "mixed",179 "types": [180 "alternative_specification",181 "subsample",182 "alternative_sample_period"183 ]184 },185 "tables": {186 "convention_notes": "Notes define the mixture parameters (π, μ, σ, σij) and the sample filter (at least 20 revisions, 1993–2015). They also explain how the long-upgrade/short-downgrade abnormal return is constructed. Tables are organized in panels contrasting constrained and unconstrained models and subsamples.",187 "units_stated_in_notes": false,188 "standard_errors_description": "unclear"189 },190 "figures": {191 "uncertainty_shown": "unclear"192 },193 "conclusion": {194 "bounded_claims": false,195 "policy_implications": false,196 "open_questions": false,197 "restates_contribution": true198 },199 "hedging": {200 "level": "moderate",201 "typical_devices": [202 "'results suggest'",203 "'consistent with'",204 "'appear to be'",205 "'it is likely the case'"206 ]207 },208 "claim_evidence": [209 {210 "claim": "Most analysts (about 94%) belong to the nonzero-skill component in the constrained mixture model.",211 "evidence_type": "table",212 "evidence_ref": "Table II"213 },214 {215 "claim": "Analysts are heterogeneous, with high- and low-type components that both earn positive abnormal returns.",216 "evidence_type": "table",217 "evidence_ref": "Table III"218 },219 {220 "claim": "All analysts can process news, and most (93%) also produce new information.",221 "evidence_type": "table",222 "evidence_ref": "Table IV"223 },224 {225 "claim": "Experience, sell share and industry breadth raise, and recommendation volume lowers, the probability of being high-type.",226 "evidence_type": "table",227 "evidence_ref": "Table V"228 },229 {230 "claim": "A median analyst's pick is worth roughly $40,000 to an institutional client.",231 "evidence_type": "table",232 "evidence_ref": "Table VI"233 },234 {235 "claim": "86% of revisions are informative, versus 12% classified as influential in prior work.",236 "evidence_type": "table",237 "evidence_ref": "Table VII, Table VIII"238 },239 {240 "claim": "A large share of brokerages are skilled, but fewer than at the analyst level; larger brokers are more likely high-type.",241 "evidence_type": "table",242 "evidence_ref": "Table IX, Table X"243 }244 ],245 "structural_summary": "The paper opens with the question of whether professionals can pick mispriced stocks and notes a gap: how widespread analyst skill is remains unknown. It then frames the problem of separating luck from skill. The method is a mixture-of-distributions model, set out before the data section. Results proceed from analyst-level skill fractions to news versus non-news decomposition, characteristic correlates, time variation and dollar value. The paper then reconciles its findings with recommendation-level 'influential' classifications and extends the analysis to brokerages. A short conclusion restates the findings, and an appendix gives a lognormal robustness variant.",246 "writing_profile_keywords": [247 "skill versus luck",248 "mixture model",249 "analyst recommendations",250 "information production versus processing",251 "dollar value",252 "contrast with mutual fund literature"253 ],254 "confidence": 0.8,255 "notes": "Primary method is classified as structural_estimation because the mixture-model distributional estimation is the core approach. The method section is labeled 'theory' by the rule, but its role is empirical design. Standard error reporting is not visible in the excerpts."256 },257 "deterministic_checks": {258 "roadmap_regex": true,259 "contribution_regex_in_intro": true,260 "magnitude_in_abstract": false,261 "abstract_n_sentences": 7,262 "intro_n_paragraphs": 9,263 "intro_n_words": 1525,264 "causal_words_in_abstract": 0,265 "hedge_words_in_abstract": 0,266 "stars_mentioned": false,267 "clustered_se_mentioned": false,268 "internet_appendix_mentioned": false,269 "section_roles_rule": {270 "0": "introduction",271 "1": "theory",272 "2": "data",273 "3": "other",274 "4": "other",275 "5": "other",276 "6": "conclusion",277 "7": "appendix"278 }279 },280 "schema_valid": true,281 "code_version": "87c73cd-dirty"282}