Research Design

Working title:

From Restatements to Public Review and Correction: Label Observability and the Public Reporting-Risk Cascade

Research setting. Traditional misstatement and restatement benchmarks label firm-years after misconduct is detected, disclosed, or made publicly visible.
Measurement problem. These labels are useful, but they mix the underlying reporting problem with discovery probability, disclosure delay, and selective public observability.
Empirical object. The paper studies a filing-native public reporting-risk cascade built from SEC and PCAOB data.
Primary target. The cascade measures public review-and-correction outcomes from the filing origin, not latent fraud truth.

Research Question and Contribution

Core question. Can reporting-risk prediction be reframed from ex post detected misconduct to filing-origin public review-and-correction risk?
Timing contamination. Static detected-misstatement labels mix occurrence, discovery, disclosure lag, and public visibility.
Main contribution. The intended contribution is a measurement redesign, not a claim that one classifier dominates prior fraud-prediction work.
Filing-origin estimand. The repo defines a filing-origin public reporting-risk estimand based only on information visible at or before origin_date.
Construct claim. The public cascade is expected to be related to, but not identical with, legacy detected-misstatement labels.
Peer comparison boundary. Peer models and metrics are used for compatibility checks; comparisons provide metric-compatible ranking evidence, not same-estimand superiority claims.
Evidence requirement. Credible bridge-based overlap validation is required before any integrated old-benchmark/public-cascade claim.

Design Overview

flowchart LR
    A["Legacy benchmark panel<br/>gvkey x data_year"] --> B["Label-observability diagnostics<br/>naive, proxy-drop, imputed-lag grid"]
    B --> C["Benchmark evidence<br/>timing sensitivity, drift, missingness"]

    D["Public SEC/PCAOB lake<br/>filings, XBRL, Notes, Form AP, AAER"] --> E["Filing-origin gold panels<br/>issuer-year and filing-origin Parquet"]
    E --> F["Public review-and-correction labels<br/>comment thread, amendment, 8-K 4.02, AAER proxy"]
    E --> G["Pre-origin predictors<br/>metadata, XBRL ratios, auditor/oversight, opacity"]
    F --> H["Public cascade models<br/>ranking, calibration, feature-family ablations"]
    G --> H
    G --> I["Public-label opacity DML<br/>adjusted association, not causal effect"]

    C --> J["Bridge gate<br/>gvkey-CIK-year crosswalk<br/>farr candidate, WRDS preferred"]
    H --> J
    I --> J
    J --> K["Integrated construct validation<br/>related but non-identical labels"]
    J -. "if validation remains incomplete" .-> L["Public-cascade measurement result<br/>reportable without fraud-truth claim"]

Prior Literature and Positioning

Literature role. Prior work supplies model families, performance metrics, and construct anchors.
Estimand shift. Prior fraud and restatement studies often predict detected ex post misconduct labels; this paper predicts subsequent public review-and-correction events from a filing-origin information set.
Metric-compatible comparison. Metric-compatible comparison is evidence about ranking behavior under a shared scoring language, not evidence that the tasks share the same estimand.

Stream	Canonical anchors	Typical models and metrics	Role in this paper
Detected misstatement and fraud prediction	Dechow, Ge, Larson, and Sloan (2011); Perols (2011); Bao, Ke, Li, Yu, and Zhang (2020); Bertomeu, Cheynel, Floyd, and Pan (2021), "Using Machine Learning to Detect Misstatements"	Logistic/F-score models, SVM, decision trees, bagging, stacking, neural nets, and tree ensembles; AUC, classification rates, lift, variable importance, and top-fraction ranking metrics	Supplies the benchmark peer suite: Dechow-style scores, a Perols-style legacy model zoo, Bao-style top-fraction balanced accuracy and NDCG, and Bertomeu-style XGBoost feature importance.
Partial observability and hidden misconduct	Barton, Burnett, Gunny, and Miller (2024); Dyck, Morse, and Zingales (2024)	Occurrence/detection separation, hidden misconduct estimation, likelihood and coefficient evidence	Motivates the estimand shift; these models are not PR-AUC comparators for the current pipeline.
SEC comment-letter and disclosure-review research	Cassell, Cunningham, and Myers (2013); Bozanic, Dietrich, and Johnson (2018); Brown, Tian, and Tucker (2018); the SEC filing review process	Regression-style evidence on comment receipt, remediation, and disclosure response	Establishes public comment-letter scrutiny as economically meaningful; this paper embeds it as one public-cascade outcome rather than the sole endpoint. This stream supplies regression-style evidence rather than direct ranking-score comparators.
Public regulatory and structured-data sources	SEC Item 4.02 guidance; SEC AAER pages; PCAOB Form AP; SEC Inline eXtensible Business Reporting Language (XBRL)	Public filing events, audit-participant data, oversight data, and standardized financial facts	Supplies the filing-native public lake and reproducible feature construction. AAER is a severity-tail descriptor, not a complete enforcement universe.

Positioning. The paper aligns the prediction target to the observable public process.
Benchmark role. The legacy benchmark remains a disciplined diagnostic for timing sensitivity, label observability, concept drift, and missingness.
Overlap role. The bridge tests where the public cascade agrees or disagrees with detected-misstatement labels.

Measurement Design

Legacy Benchmark Labels

Unit. The legacy benchmark panel uses gvkey, data_year, and misstatement firm-year.
Purpose. It tests whether traditional restatement prediction is sensitive to timing, drift, and missingness.
Label modes.
naive: the observed misstatement firm-year label without detection-timing adjustment.
proxy_drop_observed: a coverage stress test using sparse same-row res_an* timing proxies, excluding positives without usable timing evidence.
proxy_imputed_lag: a timing-assumption grid assigning unknown positives one-, two-, three-, or five-year detection lags.
external_timing: the paper-grade benchmark maturation target, available only if validated public restatement or detection dates are supplied.
Leakage rule. res_an0, res_an1, res_an2, and res_an3 are timing proxies only and never enter predictors.
Required reporting. timing_coverage.csv must report same-row timing coverage, unknown positives, retained-positive share, and class-balance changes.

Public Review-and-Correction Labels

Outcome design. The public cascade is a multi-label outcome system, not a deterministic hierarchy.
Public labels.
label_comment_thread_365: public comment-letter scrutiny, measured from the first public EDGAR date of the comment-thread sequence; source: SEC filing review process and public EDGAR correspondence.
label_amendment_365: broad amendment/friction signal, including administrative amendments, filing friction, and potentially material corrections; source: SEC EDGAR filing access and amended filing form metadata.
label_8k_402_365: Item 4.02 non-reliance and material-correction proxy; source: SEC Form 8-K, Item 4.02.
label_aaer_proxy_730: rare AAER severity-tail descriptor, fit only as robustness when positives are sufficient; source: SEC Accounting and Auditing Enforcement Releases and farr::aaer_* support data.
Co-occurrence rule. A later-stage positive does not mechanically force an earlier-stage label.
Construct meaning. These labels are not alternative names for fraud. They are public observability states: regulatory scrutiny (comment_thread), filing correction or friction (amendment), material non-reliance (8k_402), and rare public enforcement-tail evidence (aaer_proxy).
Target distinction. The target is public review-and-correction risk rather than latent fraud truth.

Timing and Censoring Rules

Origin date. In the current v1 panel, filing_origin_panel.origin_date = filing_date, and issuer_origin_panel.origin_date is the selected annual filing date for the issuer-year.
No post-origin leakage. No event released after origin_date may enter predictors.
Excluded coverage fields. source_available_*, public_date_*, vintage_*, and as_of_date document source availability and public vintages but are excluded from default predictors.
Censoring. 365-day outcomes use censored_365; AAER 730-day robustness uses censored_730.

Claim Boundaries

The public-cascade design supports evidence about a public reporting-risk state. It does not by itself establish latent fraud truth, causal identification, or a stable enforcement-prediction result.

Comment letters are public scrutiny signals, not the full SEC review universe. AAER is a severity-tail descriptor, not a complete enforcement universe.

Bridge validation is mandatory for an integrated claim that the public cascade and the old benchmark measure related but non-identical constructs. Without that validation, the public-cascade result remains a public-data measurement result rather than a validated fraud/restatement overlap paper.

Data and Feature Construction

Legacy Benchmark Panel

File. data/raw_dataset_misstatement.parquet.
Grain. gvkey x data_year.
Coverage. 2001-2019.
Required fields. gvkey, data_year, misstatement firm-year, res_an0 to res_an3, missing_* flags, and accounting/audit/governance/market/industry predictors.
Limitation. No CIK, ticker, PERMNO, restatement filing date, detector identity, or complete public filing history.

Public SEC/PCAOB Lake

Storage design. data/public_lake/ is organized as bronze, silver, and gold.
Bronze. Downloaded public files with source URL, timestamp, SHA256 hash, parser version, schema version, and as-of date.
Silver. Normalized issuer, filing, XBRL, Notes, comment-thread, correction, Form AP, PCAOB inspection, and AAER proxy tables; large Silver tables are Parquet-first.
Gold. issuer_origin_panel.parquet and filing_origin_panel.parquet.
DuckDB path. The default DuckDB path uses SQL for XBRL core-tag pivoting, label-horizon joins, and Parquet output on the annual issuer-year modeling panel.
Filing-origin provenance. The full filing-origin panel is retained as a lightweight, year-sharded provenance panel rather than a fully labeled 20M-row modeling table.
Required v1 sources. SEC submissions, SEC Financial Statement Data Sets (FSDS), SEC UPLOAD and CORRESP, 10-K/A and 10-Q/A amendments, 8-K Item 4.02, PCAOB Form AP, PCAOB inspection datasets, and SEC AAER pages.
Main public sample. Domestic U.S. GAAP issuer-years from 2011-2023, with 2026-04-23 as the current reproducibility as-of date.

Bridge and External Validation Inputs

Bridge file. data/external/gvkey_cik_year.csv.
Required fields. gvkey, issuer_cik, a single year or start/end years, and provenance fields such as source, version, extraction date, match method, and match score.
WRDS-preferred route.

set -a; source .env; set +a
uv run python scripts/prepare_gvkey_cik_crosswalk.py \
  --input path/to/wrds_cik_gvkey_link.csv \
  --out data/external/gvkey_cik_year.csv

Public candidate route while WRDS access is pending.

bash scripts/prepare_farr_gvkey_cik_bridge.sh --install-missing
bash scripts/prepare_farr_support_data.sh --install-missing

Current candidate bridge. farr::gvkey_ciks is the working high-coverage candidate bridge; it must be reported with coverage and multiplicity tables and should not be described as WRDS-verified.
AAER support. farr::aaer_firm_year and farr::aaer_dates are external AAER validation anchors.
Metadata support. farr::state_hq is a date-bounded headquarters-state metadata control.
Missing bridge behavior. If no usable external crosswalk exists, the bridge probe must report raw_identifier_blocker rather than infer links from benchmark identifiers alone.

Feature Families

Common sample rule. Feature families use the same filtered issuer-year sample for fair ablations.
Missing-value rule. Tree models with native missing-value handling retain numeric np.nan; non-tree adapters use fold-internal imputation only when required.
Excluded columns. Label, censoring, identifier, source-availability, public-date, and vintage columns are excluded by default.
Metadata. SIC, form, entity type, filing size, XBRL flags, prior filing count, days since prior filing, and headquarters-state controls when available.
Filing friction and public history. Current-cycle NT status and amendment friction, plus strictly pre-origin rolling counts and recency for prior NT filings, comment threads, amendments, and 8-K instability items.
XBRL. xbrl_ratio_* and xbrl_coverage_* features from controlled core tags, including size, leverage, profitability, working capital, receivables, inventory, cash, debt, operating cash flow, and year-over-year revenue/assets changes.
Auditor and oversight. PCAOB Form AP fields, engagement-partner exposure, and PCAOB inspection features in their public source windows.
Note opacity. Note count, note character count, note-tag coverage, and tag entropy as a disclosure breadth measure.
Deferred extensions. Proxy-governance content, SEC insider-pressure features, macro-vintage controls, auditor-firm public-status fields, and broader security/attention layers are useful extensions, not required for the current v1 paper claim.

Empirical Design

Evaluation Metrics

Predictive metrics. PR-AUC, ROC-AUC, Brier score, Brier Skill Score, expected calibration error, top-50/100/200 precision, and Bao-style top-fraction metrics.
Bao-style metrics. Top-fraction precision, sensitivity, specificity, balanced accuracy, and binary-relevance NDCG@k.
Calibration. Calibration metrics are diagnostic under class imbalance and resampling.
Prevalence. Prevalence is the positive-class rate in the evaluated sample and the natural random-ranking baseline for PR-AUC.
PR-AUC interpretation. When positives are rare, a numerically small PR-AUC can still represent meaningful lift over the base rate.
ROC-AUC contrast. ROC-AUC has a fixed random baseline near 0.5 and can look much larger than PR-AUC in rare-event settings.
Split design. Prediction experiments use annual out-of-time evaluation, not random cross-validation.
Training windows. For a given test year, training uses earlier years only, with expanding or rolling 5-, 7-, and 10-year windows.
DML separation. Cross-fitting appears separately in Double / Debiased Machine Learning (DML) opacity diagnostics; it is not the train/test split used for headline prediction tables.

Experiment 1: Label Observability and Detection Timing

Purpose. Quantify how sensitive traditional restatement evaluation is to timing coverage and unknown-positive assumptions.
Design. Run annual out-of-time benchmark backtests across expanding, rolling 5-year, rolling 7-year, and rolling 10-year windows; compare naive, proxy_drop_observed, and proxy_imputed_lag labels.
Outputs. rolling_metrics.csv, rolling_predictions.parquet, timing_coverage.csv, timing_summary.json, timing_claim_status, and window summaries.
Interpretation. This is a benchmark-validity diagnostic; a decline under proxy_drop_observed is timing-observability sensitivity, not proof of look-ahead bias by itself.

Experiment 2: Concept Drift and Model Shelf-Life

Purpose. Estimate whether reporting-risk models trained in one regime remain useful in later regimes.
Design. Compare rolling and expanding windows over test years; track feature-family importance; report pre/post diagnostics around major regulatory and data-regime breakpoints.
Outputs. Annual metrics, window summaries, structural-break diagnostics, and feature-family importance.
Interpretation. The experiment supports model shelf-life and retraining-window evidence; it does not establish structural causality from predictive drift alone.

Experiment 3: Opacity and Public Review/Correction Risk

Purpose. Test whether pre-origin opacity and missingness profiles predict later public scrutiny or correction.
Design. Construct missingness-density and missing-profile indicators; estimate Double / Debiased Machine Learning (DML) partially linear regressions on public labels.
Primary outcomes. label_comment_thread_365, label_amendment_365, and label_8k_402_365.
Treatment-like variable. D = missingness_density_score.
Controls. X = pre-origin metadata, XBRL, filing-friction, public-history, auditor, oversight, note-opacity, and calendar controls.
Outputs. Missing-profile clusters, public-label PLR spec results, nuisance-model metadata, and diagnostic benchmark-side DML outputs.
Interpretation. Coefficients are adjusted associations, not causal effects; the old misstatement firm-year outcome remains a legacy diagnostic only.

Experiment 4: Public Cascade Construction

Purpose. Demonstrate that public data can support a defensible review-and-correction cascade.
Design. Build the public lake from SEC/PCAOB sources; construct labels from first public dates; report source coverage, event rates, censoring, and task readiness.
Outputs. Source coverage tables, event-rate tables, censoring summaries, public-lake metadata, and task-positive counts.
Interpretation. This experiment validates the measurement surface; AAER remains descriptive severity-tail evidence unless robust positive counts are available.

Experiment 5: Public Cascade Prediction

Purpose. Estimate the pre-disclosure public reporting-risk state from public features.
Design. Use issuer_origin_panel to predict comment-thread scrutiny, broad amendment/friction, and 8-K Item 4.02 outcomes; run feature-family ablations over metadata, XBRL, auditor, oversight, and all-feature sets.
Skip rule. Skip task/family/window fits with one-class train or test labels.
Outputs. public_cascade_metrics.csv, public_cascade_predictions.parquet, public_cascade_task_status.csv, public_cascade_summary.md, and public_opacity_dml.csv.
Interpretation. Full public-cascade claims require non-metadata features; metadata_baseline is a readiness state, xbrl_ratio_baseline is the first non-metadata empirical baseline, and sparse AAER folds are blockers rather than failed headline models.

Experiment 6: Old Benchmark and Public Cascade Overlap

Purpose. Test whether legacy detected-misstatement labels and public review-and-correction labels measure related but non-identical constructs.
Design. Run the bridge probe, report coverage and multiplicity, then test event-time concentration and reciprocal risk-score alignment in the mapped sample.
Current bridge. The current implementation uses farr gvkey_ciks as a high-coverage candidate bridge; WRDS remains the preferred final validation source.
Outputs. bridge_probe_summary.json, coverage_report.csv, multiplicity_report.csv, unmatched_raw_characteristics.csv, construct_overlap/label_contingency_lift.csv, construct_overlap/public_score_legacy_ranking.csv, construct_overlap/reciprocal_alignment.csv, construct_overlap/event_time_concentration.csv, and AAER severity-tail support tables.
Interpretation. This is the integrated-paper gate; the farr bridge supports candidate overlap validation, and WRDS remains preferred before final manuscript claims.

Evidence Gates

Component	Current status	Gate before paper claim
Benchmark timing	implemented as observability sensitivity	report `timing_coverage.csv`, retained positives, and imputed-lag scenarios; external timing required for paper-grade maturation
Concept drift	implemented as rolling-window diagnostics	validate annual PR-AUC, Brier Skill Score, feature-importance drift, and breakpoint summaries
Opacity	public-label DML implemented; refresh summary is separate from construct overlap	public-label PLR results must use `label_comment_thread_365`, `label_amendment_365`, and `label_8k_402_365` as primary outcomes
Public lake	full public lake path implemented	refreshed source coverage, row counts, censoring, and reproducibility metadata
Public cascade	current full-run state is `xbrl_ratio_baseline`	non-degenerate comment-thread, amendment, and 8-K Item 4.02 tasks; AAER framed as severity-tail evidence
Bridge overlap	farr candidate overlap implemented	coverage, multiplicity, reciprocal alignment, no silent many-to-many joins, and WRDS-preferred validation before final integrated claims

Data integrity gates.
No post-origin_date event enters predictors.
No res_an* column enters benchmark predictors.
source_available_*, public_date_*, vintage_*, and as_of_date stay outside default predictors.
Censoring masks are horizon-specific.
Crosswalk coverage and multiplicity are reported before overlap validation.
Construct-overlap outputs carry validation_tier = candidate_farr until a verified WRDS bridge is supplied.
Empirical sufficiency gates.
Benchmark outputs non-empty rolling metrics, timing coverage, and missingness diagnostics.
Public cascade covers fiscal years 2011-2023 in the full panel.
Comment-thread, amendment, and 8-K Item 4.02 tasks have nonzero positives.
xbrl_ratio_* and xbrl_coverage_* features are present for non-metadata public-cascade evidence.
Zero-positive or sparse AAER robustness tasks are skipped and reported as severity-tail blockers.
Paper-readiness gates.
Claims remain measurement and decision-useful prediction claims, not causal proof of fraud occurrence.
AAER is described as a severity-tail descriptive proxy.
Comment letters are described as public scrutiny, not complete SEC review.
Bridge validation is mandatory for the integrated old-benchmark/public-cascade paper claim.
Candidate farr overlap can support a related-but-non-identical construct argument, but not final WRDS-quality validation.

Execution Contract

Operational source of truth. The operational command surface lives in the repository home page and README so there is a single maintained entrypoint for users and coauthors.
Quality gate.

just check

Paper-facing core run.

just full full raw artifacts/full

Peer-compatible model-family transfer.

just task study raw artifacts/full_with_peer \
  extra="--peer-comparison-mode full --peer-target both --parallel-jobs 4 --model-threads 2 --seed-policy task-isolated"

Command boundary. just check is the local quality gate; just full full raw artifacts/full is the paper-facing core run for data engineering and core experiments; full_with_peer adds the legacy and public-label peer model-family transfer suites. Use --peer-target public when only the public-label peer transfer needs to be refreshed.
Detailed operations. Component-level reruns and public-lake operational details are documented in the repository home page, which includes the root README.md.