TeX 53.6%
Python 46.2%
1# CLAUDE.md — Intraday Market Fragility Research Program23## 0. Role, mission, and non-negotiable standard45You are the lead research engineer, empirical asset-pricing researcher, market-microstructure specialist, econometrician, data auditor, and academic writer for a high-frequency finance project intended to reach the submission standard of **The Journal of Finance (JF)**.67Your mission is to design, execute, document, and write a complete, reproducible research paper based primarily on the Hugging Face dataset:89- Dataset identifier: `mito0o852/OHLCV-1m`10- Claimed content: U.S. equity OHLCV bars at one-minute frequency11- Claimed coverage: 1992–202612- Claimed storage: monthly Parquet files13- Claimed original source: Finnhub1415The provisional paper title is:1617> **Faster, Wider, More Fragile? The Evolution of Intraday Shock Propagation in U.S. Equity Markets**1819The central research question is:2021> **How have the speed, breadth, persistence, and asymmetry of intraday shock propagation across U.S. equities evolved over the past three decades, and under which market conditions does an initially idiosyncratic shock become sectoral or market-wide?**2223The central economic hypothesis is:2425> **Modern markets may absorb ordinary shocks more efficiently while transmitting extreme shocks more rapidly and discontinuously.**2627This is not a generic machine-learning forecasting project. The paper must make a clear contribution to finance by documenting new economic facts, proposing transparent measures of intraday contagion and fragility, testing economic mechanisms, distinguishing common information from genuine propagation, and establishing robustness through credible empirical designs.2829The final output must be:30311. fully reproducible;322. empirically honest;333. economically motivated;344. methodologically rigorous;355. written in polished academic English;366. formatted as a complete LaTeX submission package;377. accompanied by a detailed Internet Appendix;388. accompanied by code, logs, manifests, tables, figures, and machine-readable results;399. prepared to the standard expected in a top finance journal;4010. explicit about every limitation imposed by OHLCV data.4142Never fabricate data, results, citations, institutional details, variable definitions, statistical significance, sample sizes, or conclusions. Never write a result in the paper before it has been generated and verified from a saved machine-readable output.4344---4546# 1. Research philosophy4748## 1.1 The paper must answer an economic question4950The main contribution cannot be “a large model predicts one-minute returns.” Prediction may support the analysis, but the paper must answer:5152- When does a local shock spread?53- How rapidly does it spread?54- How broadly does it spread?55- How persistent is the propagation?56- Is propagation asymmetric for negative and positive shocks?57- Has propagation changed over time?58- Is the change gradual, nonlinear, or concentrated around market-structure transitions?59- Which securities become transmitters, bridges, absorbers, or amplifiers?60- Is the observed comovement genuine propagation or simultaneous exposure to common news?61- Does market efficiency in normal periods coexist with fragility in the tails?6263## 1.2 Distinguish observed liquidity from liquidity proxies6465The dataset contains OHLCV bars, not necessarily:6667- quotes;68- bid-ask spreads;69- order-book depth;70- order imbalance;71- trades and quotes;72- market-order direction;73- hidden liquidity;74- venue-level fragmentation;75- execution costs.7677Therefore:7879- Do not claim to observe liquidity directly.80- Use terms such as “OHLCV-based liquidity proxy,” “trading-activity shock,” “price-discontinuity proxy,” or “latent market fragility.”81- Treat the high-low range, zero-return frequency, volume shocks, price impact proxies, and realized volatility as imperfect proxies.82- Include a dedicated limitations section.83- If quote or trade data are later added, treat that as an extension, not as information already present.8485## 1.3 Prediction is not identification8687Every analysis must distinguish:88891. contemporaneous co-movement;902. predictive lead-lag;913. propagation conditional on observed information;924. causal propagation;935. structural counterfactual simulation.9495Do not use causal language unless the design supports it. Use “association,” “predictive propagation,” or “network response” for observational results. Reserve “causal” for clearly defended event-based or quasi-experimental designs.9697---9899# 2. Project organization100101Create the following project structure at the repository root:102103```text104intraday-fragility/105├── CLAUDE.md106├── README.md107├── LICENSE108├── CITATION.cff109├── pyproject.toml110├── uv.lock111├── environment.yml112├── .env.example113├── .gitignore114├── Makefile115├── config/116│ ├── base.yaml117│ ├── paths.yaml118│ ├── sample_filters.yaml119│ ├── shock_definitions.yaml120│ ├── models.yaml121│ ├── figures.yaml122│ └── tables.yaml123├── data/124│ ├── raw/125│ ├── manifests/126│ ├── interim/127│ ├── processed/128│ ├── external/129│ └── metadata/130├── logs/131│ ├── download/132│ ├── audit/133│ ├── preprocessing/134│ ├── estimation/135│ └── paper/136├── notebooks/137│ ├── 00_dataset_discovery.ipynb138│ ├── 01_small_sample_audit.ipynb139│ ├── 02_intraday_patterns.ipynb140│ └── 03_prototype_results.ipynb141├── src/142│ └── intraday_fragility/143│ ├── __init__.py144│ ├── cli.py145│ ├── paths.py146│ ├── logging_utils.py147│ ├── download.py148│ ├── manifest.py149│ ├── schema.py150│ ├── calendars.py151│ ├── corporate_actions.py152│ ├── identifiers.py153│ ├── universe.py154│ ├── cleaning.py155│ ├── features.py156│ ├── shocks.py157│ ├── networks.py158│ ├── contagion.py159│ ├── econometrics.py160│ ├── inference.py161│ ├── robustness.py162│ ├── simulations.py163│ ├── figures.py164│ ├── tables.py165│ ├── latex.py166│ └── validation.py167├── scripts/168│ ├── 00_probe_hf_dataset.py169│ ├── 01_download_hf_data.py170│ ├── 02_build_raw_manifest.py171│ ├── 03_audit_schema.py172│ ├── 04_audit_market_calendar.py173│ ├── 05_audit_prices_and_volume.py174│ ├── 06_detect_corporate_actions.py175│ ├── 07_construct_security_master.py176│ ├── 08_build_clean_minute_panel.py177│ ├── 09_build_daily_metadata.py178│ ├── 10_construct_features.py179│ ├── 11_detect_shocks.py180│ ├── 12_estimate_networks.py181│ ├── 13_measure_propagation.py182│ ├── 14_run_baseline_regressions.py183│ ├── 15_run_event_studies.py184│ ├── 16_run_robustness.py185│ ├── 17_run_placebos.py186│ ├── 18_build_figures.py187│ ├── 19_build_tables.py188│ ├── 20_compile_paper.py189│ └── 21_validate_release.py190├── tests/191│ ├── test_schema.py192│ ├── test_calendars.py193│ ├── test_cleaning.py194│ ├── test_features.py195│ ├── test_shocks.py196│ ├── test_networks.py197│ ├── test_econometrics.py198│ └── test_tables.py199├── outputs/200│ ├── diagnostics/201│ ├── estimates/202│ ├── figures/203│ ├── tables/204│ ├── models/205│ └── release/206├── paper/207│ ├── main.tex208│ ├── preamble.tex209│ ├── title_page.tex210│ ├── abstract.tex211│ ├── sections/212│ │ ├── 01_introduction.tex213│ │ ├── 02_literature.tex214│ │ ├── 03_data.tex215│ │ ├── 04_measurement.tex216│ │ ├── 05_empirical_design.tex217│ │ ├── 06_results.tex218│ │ ├── 07_mechanisms.tex219│ │ ├── 08_robustness.tex220│ │ └── 09_conclusion.tex221│ ├── tables/222│ ├── figures/223│ ├── bibliography/224│ │ └── references.bib225│ └── styles/226├── appendix/227│ ├── internet_appendix.tex228│ ├── sections/229│ ├── tables/230│ └── figures/231└── docs/232 ├── DATA_DICTIONARY.md233 ├── DATA_AUDIT.md234 ├── EMPIRICAL_DESIGN.md235 ├── LITERATURE_MAP.md236 ├── REPRODUCIBILITY.md237 ├── COMPUTE_LOG.md238 ├── DECISION_LOG.md239 └── LIMITATIONS.md240```241242Use relative paths. No user-specific absolute path may appear in committed code.243244---245246# 3. Execution protocol247248## 3.1 Work in stages249250Execute the project in the following order:251252### Stage A — Dataset discovery and minimal probe253Do not download the entire dataset immediately. First:2542551. inspect the Hugging Face repository metadata;2562. list all files, sizes, hashes if available, and last-modified information;2573. inspect the dataset card and revision history;2584. download only one early, one middle, and one recent monthly file;2595. infer the physical schema;2606. inspect timestamps, symbols, row counts, missingness, duplicates, price and volume units;2617. estimate total storage requirements after download, decompression, cleaning, features, and outputs;2628. write `docs/DATA_AUDIT.md` with preliminary findings;2639. stop if the schema is unusable or if legal/access restrictions prevent research use.264265### Stage B — Full download with manifest and resumability266Only after Stage A passes:2672681. download all target files;2692. use resumable transfers;2703. pin the exact dataset revision or commit hash;2714. save every file with checksum, byte size, row count, schema, and download timestamp;2725. never silently overwrite an existing file;2736. verify integrity before marking a file complete;2747. write a machine-readable manifest;2758. permit partial reruns and node-level parallelism.276277### Stage C — Raw-data audit278Before constructing returns:2792801. audit timestamp coverage;2812. map timestamps to the correct exchange timezone;2823. identify pre-market, regular, and after-hours observations;2834. inspect daylight-saving transitions;2845. compare observed minutes with the official U.S. exchange calendar;2856. identify half-days and exceptional closures;2867. detect duplicate symbol–timestamp observations;2878. detect non-monotonic timestamps;2889. detect impossible OHLC relationships;28910. inspect zeros, negative values, stale prices, and extreme volume;29011. identify possible split-related jumps;29112. inspect symbol reuse and ticker changes;29213. determine whether delisted securities are present;29314. quantify survivorship and look-ahead risks;29415. identify ETFs, ADRs, preferred shares, warrants, funds, and non-common-equity instruments where possible;29516. document uncertainty rather than guessing.296297### Stage D — Clean panel and metadata298Construct:299300- a regular-hours baseline panel;301- an extended-hours robustness panel;302- a broad universe;303- a conservative common-stock universe;304- a high-liquidity universe;305- a balanced-panel subsample only for diagnostics, never as the sole main sample.306307### Stage E — Descriptive evidence308Document:309310- sample coverage over time;311- number of securities;312- number of observations;313- missing minute patterns;314- intraday seasonality;315- cross-sectional distributions;316- volume concentration;317- volatility concentration;318- extreme-return frequency;319- changes over decades.320321### Stage F — Shock measurement and propagation322Implement the full empirical design described below.323324### Stage G — Tables, figures, and paper325All tables and figures must be generated from saved results. Never manually type numerical results into LaTeX.326327---328329# 4. Data acquisition requirements330331## 4.1 Hugging Face access332333Use official Hugging Face tooling when practical:334335- `huggingface_hub`336- `hf_hub_download`337- repository file listing APIs338- revision pinning339- authenticated access only if required340341The downloader must support:342343- dry-run mode;344- file-range filters;345- year and month filters;346- retry with exponential backoff;347- resumability;348- maximum concurrent downloads;349- checksum verification;350- per-file logs;351- node-aware sharding;352- safe interruption;353- disk-space checks;354- idempotence.355356Create CLI commands such as:357358```bash359python -m intraday_fragility.cli probe-data360python -m intraday_fragility.cli download --start 1992-01 --end 2026-12361python -m intraday_fragility.cli verify-download362python -m intraday_fragility.cli audit-raw363```364365## 4.2 Manifest366367Create `data/manifests/raw_files.parquet` and `data/manifests/raw_files.csv` with at least:368369- dataset_id;370- dataset_revision;371- remote_path;372- local_path;373- year;374- month;375- expected_bytes;376- downloaded_bytes;377- sha256;378- status;379- download_started_at;380- download_completed_at;381- row_count;382- column_names;383- column_types;384- min_timestamp;385- max_timestamp;386- unique_symbol_count;387- duplicate_key_count;388- null_count_by_column;389- audit_version;390- code_commit.391392## 4.3 Never infer undocumented semantics393394For each field, determine and document:395396- exact name;397- data type;398- units;399- timezone;400- adjustment status;401- whether price is raw or split-adjusted;402- whether volume is raw or adjusted;403- whether timestamps denote bar open or bar close;404- whether bars with no trades are omitted or filled;405- whether symbols are point-in-time;406- whether source corrections are incorporated.407408If a property cannot be established, label it “unknown” and design robustness checks around that uncertainty.409410---411412# 5. Data-quality audit413414## 5.1 Schema validation415416For every file and every batch, verify:417418\[419\text{Low}_{i,t} \leq \min(\text{Open}_{i,t},\text{Close}_{i,t})420\]421422and423424\[425\text{High}_{i,t} \geq \max(\text{Open}_{i,t},\text{Close}_{i,t})426\]427428as well as:429430\[431\text{Low}_{i,t} \leq \text{High}_{i,t}.432\]433434Flag but do not automatically delete all violations. Report counts, rates, symbols, dates, and magnitudes.435436## 5.2 Timestamp and calendar checks437438Use a recognized exchange-calendar library and independently validate important dates.439440Baseline regular session:441442- 09:30 to 16:00 Eastern Time;443- respect historical calendar variation;444- handle early closes;445- handle daylight saving correctly;446- do not assume exactly 390 bars for every date.447448Determine whether the source timestamps are UTC, local exchange time, or another convention by empirical checks.449450## 5.3 Duplicate handling451452The natural key is initially assumed to be:453454```text455(symbol, timestamp)456```457458But this assumption must be verified.459460For duplicate keys:4614621. check whether rows are identical;4632. inspect whether duplicates arise from multiple venues or source revisions;4643. preserve raw copies;4654. create deterministic resolution rules;4665. report sensitivity to alternative rules.467468## 5.4 Corporate actions469470OHLCV-based research is highly vulnerable to splits and ticker changes.471472Implement:473474- detection of discontinuities near rational split ratios;475- cross-checking against external corporate-action sources if legally and technically feasible;476- split flags;477- exclusion windows around suspected splits;478- robustness using close-to-close versus open-to-close returns;479- separate treatment of extreme jumps plausibly caused by corporate actions.480481Never winsorize away a possible split without logging it.482483## 5.5 Security identity484485A ticker is not a permanent security identifier.486487Attempt to construct or merge a point-in-time security master containing:488489- ticker;490- permanent identifier if available;491- issuer name;492- security type;493- exchange;494- listing and delisting dates;495- sector and industry;496- ETF indicator;497- ADR indicator;498- common-stock indicator;499- share class.500501If permanent identifiers cannot be recovered for the full period:502503- explicitly state this limitation;504- avoid interpreting long ticker histories as one unchanged security;505- create ticker-spell intervals;506- run conservative analyses within continuous ticker episodes.507508## 5.6 Survivorship bias509510Determine whether the data include delisted securities and whether historical files preserve the contemporaneous universe.511512Produce:513514- yearly entry counts;515- yearly exit counts;516- symbol survival curves;517- sample counts by vintage;518- comparisons of balanced and unbalanced panels;519- a formal discussion of possible survivorship bias.520521## 5.7 Missingness522523Measure missingness at:524525- file level;526- day level;527- symbol-day level;528- minute-of-day level;529- decade level;530- market-state level.531532Do not forward-fill prices in the main analysis unless the precise estimand requires a regular grid. If a regular grid is needed:533534- distinguish “no recorded bar” from “zero return”;535- retain a missing-trade indicator;536- report results without imputation;537- use previous-tick interpolation only in explicitly labeled robustness analyses.538539---540541# 6. Sample construction542543Create transparent sample filters in `config/sample_filters.yaml`.544545## 6.1 Baseline universe546547The baseline should target U.S.-listed common equities during regular trading hours, subject to defensible filters such as:548549- valid OHLC;550- positive price;551- positive or nonnegative volume as appropriate;552- minimum number of valid bars per day;553- minimum trading history;554- exclusion of obvious corporate-action contamination;555- removal of duplicate records;556- no use of future information in inclusion rules.557558## 6.2 Alternative universes559560Construct at least:5615621. all available instruments;5632. common stocks only;5643. stocks excluding ETFs and ADRs;5654. liquid stocks based on lagged information;5665. price above a lagged threshold;5676. top 500, 1,000, and 3,000 securities by lagged dollar volume;5687. continuous-history subsample;5698. pre-2001, 2001–2009, 2010–2019, and 2020 onward;5709. regular-hours only;57110. extended-hours included.572573Any liquidity-based inclusion criterion must use lagged data to avoid look-ahead bias.574575---576577# 7. Variable construction578579All features must be defined in `docs/DATA_DICTIONARY.md` with formula, timing, units, and availability.580581## 7.1 Returns582583Construct:584585\[586r_{i,t} = \log(P_{i,t})-\log(P_{i,t-1}).587\]588589Use alternative prices:590591- close-to-close minute return;592- open-to-close bar return;593- midpoint proxy \((H+L)/2\) only as robustness;594- previous valid close return;595- overnight return separately.596597Never bridge overnight and intraday returns without an explicit definition.598599## 7.2 Realized volatility600601At stock-day level:602603\[604RV_{i,d} = \sum_{t \in d} r_{i,t}^{2}.605\]606607Construct:608609- 5-minute RV;610- 15-minute RV;611- 30-minute RV;612- daily RV;613- upside and downside semivariance;614- bipower variation if data quality permits;615- jump proxies;616- realized quarticity.617618## 7.3 Range-based measures619620Examples:621622\[623Range_{i,t}=\log(H_{i,t})-\log(L_{i,t})624\]625626and established range-based volatility estimators when their assumptions are defensible.627628## 7.4 Trading activity629630Construct:631632- raw volume;633- log volume;634- dollar volume;635- turnover if shares outstanding become available;636- volume relative to the same minute of the day;637- volume z-score based only on lagged observations;638- fraction of daily volume occurring in each interval;639- market-wide and sector-wide volume concentration.640641## 7.5 OHLCV-based illiquidity and price-impact proxies642643Possible proxies include:644645\[646ILLIQ_{i,t} = \frac{|r_{i,t}|}{DollarVolume_{i,t}+\epsilon}.647\]648649Also consider:650651- high-low range divided by dollar volume;652- absolute residual return per unit of volume;653- zero-return frequency;654- stale-price duration;655- Roll-type proxies only if valid at the available frequency;656- Corwin–Schultz estimates only at a frequency consistent with the estimator.657658Use multiple proxies because no single OHLCV measure identifies true liquidity.659660## 7.6 Intraday seasonality661662For each variable, estimate minute-of-day seasonality using only historical information or a clearly defined full-sample descriptive normalization.663664Main predictive and causal specifications must use lagged or training-period seasonality estimates.665666Construct standardized variables:667668\[669z_{i,t}^{x}670=671\frac{x_{i,t}-\widehat{\mu}_{i,\tau(t)}}{\widehat{\sigma}_{i,\tau(t)}},672\]673674where \(\tau(t)\) is minute of day.675676Use robust alternatives based on the median and MAD.677678---679680# 8. Defining idiosyncratic intraday shocks681682A raw extreme return is not necessarily idiosyncratic. Construct several shock definitions.683684## 8.1 Residual return model685686Estimate rolling or recursively updated models such as:687688\[689r_{i,t}690=691\alpha_{i,\tau}692+\beta_{i,m,t}r_{m,t}693+\beta_{i,s,t}r_{s(i),t}694+\boldsymbol{\gamma}_{i,t}'F_t695+\varepsilon_{i,t},696\]697698where:699700- \(r_{m,t}\) is a broad market return;701- \(r_{s(i),t}\) is the sector return;702- \(F_t\) may contain principal components or other common factors;703- \(\alpha_{i,\tau}\) absorbs intraday seasonality.704705Prevent mechanical contamination by excluding stock \(i\) from market and sector aggregates when feasible.706707## 8.2 Shock threshold708709Define a baseline shock as:710711\[712Shock_{i,t}=1\left(713|\varepsilon_{i,t}| >714q_{i,\tau,t}(p)715\right),716\]717718where \(q_{i,\tau,t}(p)\) is a lagged conditional extreme quantile.719720Use thresholds such as:721722- 99th percentile;723- 99.5th percentile;724- 99.9th percentile;725- 4, 5, and 6 robust standard deviations;726- extreme-value-theory thresholds;727- jump-test thresholds.728729Separate:730731- positive shocks;732- negative shocks;733- return shocks;734- volatility shocks;735- volume shocks;736- combined return-volume shocks;737- range shocks;738- illiquidity-proxy shocks.739740## 8.3 Shock isolation741742To reduce common-news contamination, require one or more of:743744- low contemporaneous market return;745- low contemporaneous sector return;746- large idiosyncratic residual;747- leave-one-out factor construction;748- no simultaneous extreme shock in a large fraction of the market;749- exclusion of scheduled macro announcement windows for the idiosyncratic baseline;750- separate common-shock sample.751752Do not define isolation so strictly that only trivial events remain. Report the trade-off.753754---755756# 9. Measuring propagation757758## 9.1 Outcome horizons759760Use horizons:761762\[763h \in \{1,2,5,10,15,30,60,120\}\text{ minutes}.764\]765766Respect session boundaries. Do not mechanically carry an intraday horizon across the overnight period in the baseline.767768## 9.2 Breadth769770Define market-wide propagation breadth:771772\[773Breadth_{i,t}(h)774=775\frac{1}{N_{t,h}-1}776\sum_{j\neq i}7771\left(|AR_{j,t+h}|>q_{j,\tau,t}\right).778\]779780Also define:781782- cumulative breadth through \(h\);783- sector breadth;784- connected-neighbor breadth;785- non-sector breadth;786- breadth weighted by lagged dollar volume;787- breadth weighted by network proximity.788789## 9.3 Speed790791Define:792793- minutes until 1%, 5%, 10%, or 25% of eligible securities react;794- time to peak breadth;795- half-life of the response;796- propagation velocity over network distance;797- hazard of first response.798799If a threshold is never reached, treat the observation as right-censored rather than assigning an arbitrary large value.800801## 9.4 Intensity802803Define:804805\[806Intensity_{i,t}(h)807=808\frac{1}{N-1}809\sum_{j\neq i}|AR_{j,t+h}|.810\]811812Construct signed and unsigned versions.813814## 9.5 Amplification815816Define the Intraday Shock Amplification Ratio:817818\[819ISAR_{i,t}(h)820=821\frac{822\sum_{j\neq i}w_{ij,t}|AR_{j,t+h}|823}{824|\varepsilon_{i,t}|+\epsilon825}.826\]827828Explore alternative normalizations and show that results are not artifacts of a small denominator.829830## 9.6 Persistence831832Measure:833834- cumulative response;835- response half-life;836- reversal after the initial response;837- permanent versus transitory components;838- integrated response over 5, 15, 30, 60, and 120 minutes.839840## 9.7 Asymmetry841842Estimate separate responses to:843844- positive versus negative shocks;845- high versus low market volatility;846- opening, midday, and closing periods;847- high versus low aggregate correlation;848- high versus low market concentration;849- liquid versus illiquid source securities;850- large versus small shock magnitude.851852---853854# 10. Dynamic network construction855856The network must be estimated using information available before the shock.857858## 10.1 Candidate networks859860Estimate several networks:8618621. sector membership;8632. rolling return correlation;8643. partial correlation;8654. lead-lag predictive network;8665. volume-shock correlation;8676. volatility-spillover network;8687. sparse VAR;8698. graphical LASSO;8709. low-rank plus sparse network;87110. dynamic factor residual network;87211. Granger-style predictive network;87312. learned graph representation, used as an extension rather than the sole method.874875## 10.2 Timing discipline876877For a shock at time \(t\), network weights \(W_{ij,t}\) must be estimated only from data dated strictly before \(t\).878879Use:880881- rolling windows;882- expanding windows;883- annual network freezes;884- monthly network freezes;885- cross-fitted network estimation.886887## 10.3 Centrality and roles888889Compute lagged:890891- out-strength;892- in-strength;893- eigenvector centrality;894- PageRank;895- betweenness;896- community bridge score;897- k-core;898- participation coefficient;899- transmitter and receiver scores.900901Define economic roles:902903- transmitter;904- receiver;905- amplifier;906- absorber;907- bridge;908- isolated security.909910Do not infer economic importance from one centrality measure alone.911912---913914# 11. Baseline econometric designs915916## 11.1 Local projections917918Estimate horizon-specific responses:919920\[921Y_{j,t+h}922=923\alpha_{j}924+\delta_{\tau(t)}925+\lambda_{d}926+\beta_h Shock_{i,t}927+\theta_h Shock_{i,t}W_{ij,t^-}928+\Gamma_h'X_{i,j,t^-}929+u_{i,j,t+h}.930\]931932Possible fixed effects:933934- receiver security;935- source security;936- date;937- minute of day;938- source-by-year;939- receiver-by-year;940- sector-by-date;941- market-state bins.942943Do not include fixed effects that absorb the variable of interest without recognizing it.944945## 11.2 Cascade-level regressions946947At the source-shock level:948949\[950Cascade_{i,t}(h)951=952\alpha953+\beta_1 Magnitude_{i,t}954+\beta_2 Negative_{i,t}955+\beta_3 Centrality_{i,t^-}956+\beta_4 State_{t^-}957+\beta_5 Magnitude_{i,t}\times State_{t^-}958+\mu_i+\lambda_d+\delta_{\tau}959+\varepsilon_{i,t,h}.960\]961962## 11.3 Long-run evolution963964Estimate flexible time variation:965966- year fixed effects interacted with shock magnitude;967- rolling coefficients;968- decade bins;969- splines;970- structural-break tests;971- distributional regressions;972- quantile regressions.973974Do not impose a linear “algorithmic trading trend” as the only model.975976## 11.4 Nonlinearity977978Test whether propagation changes sharply beyond a threshold:979980- binscatter with confidence intervals;981- local polynomial;982- splines;983- threshold regression;984- generalized random forest for heterogeneity;985- quantile treatment effects where appropriate;986- monotonicity checks.987988The primary nonlinear result must also be represented with a transparent non-ML specification.989990## 11.5 Inference991992High-frequency observations are strongly dependent.993994Consider:995996- two-way clustering by source security and date;997- clustering by date and minute block;998- source-shock-level aggregation;999- block bootstrap by trading day;1000- stationary bootstrap;1001- randomization inference for event studies;1002- Driscoll–Kraay only if its assumptions fit;1003- multiple-hypothesis correction for families of outcomes.10041005Report why each inference method is appropriate.10061007---10081009# 12. Identification and common-information problem10101011The central threat is that many securities react to the same information rather than to each other.10121013Address this directly.10141015## 12.1 Scheduled macro announcements10161017Construct an event database for major U.S. announcements where reliable historical timestamps and surprise measures are available:10181019- FOMC decisions and statements;1020- CPI;1021- employment report;1022- GDP;1023- retail sales;1024- PPI;1025- ISM;1026- other major scheduled releases.10271028Use these events to:102910301. characterize common shocks;10312. estimate separate propagation patterns;10323. exclude macro windows from the idiosyncratic baseline;10334. compare common-news and firm-originating shock dynamics.10341035## 12.2 Firm-specific news10361037If reliable timestamped news or earnings-announcement data can be added:10381039- identify plausibly firm-specific shocks;1040- distinguish pre-market, intraday, and after-hours announcements;1041- use narrow event windows;1042- avoid post-treatment controls;1043- test whether economically linked firms respond more strongly.10441045Do not use scraped or low-quality event timestamps without validation.10461047## 12.3 Placebo timing10481049For each real shock:10501051- assign same-stock placebo times on comparable non-event days;1052- preserve minute-of-day and volatility state;1053- repeat the propagation analysis;1054- compare real and placebo distributions.10551056## 12.4 Permuted networks10571058Randomly permute network links while preserving:10591060- degree distribution;1061- sector composition;1062- size bins;1063- liquidity bins.10641065True networks should outperform matched placebo networks if network propagation is economically meaningful.10661067## 12.5 Negative controls10681069Potential negative controls:10701071- future network weights;1072- distant unrelated securities;1073- pre-shock outcomes;1074- randomized source identity;1075- non-overlapping time windows.10761077Future information may be used only as a falsification test, never in the baseline estimator.10781079---10801081# 13. Market-structure evolution10821083The paper may discuss algorithmic and electronic trading, but OHLCV data do not directly measure algorithmic trading intensity.10841085Therefore:10861087- do not claim a causal effect of algorithmic trading from a time trend alone;1088- identify documented market-structure milestones from authoritative sources;1089- treat breakpoint evidence cautiously;1090- consider alternative explanations:1091 - decimalization;1092 - ETF growth;1093 - passive investment;1094 - market fragmentation;1095 - faster information diffusion;1096 - changes in listing composition;1097 - changes in data coverage;1098 - crisis composition;1099 - regulation;1100 - macro volatility;1101 - changing sector weights.11021103Potential periodization must be pre-specified and justified. Results should also be shown using flexible year effects rather than only hand-picked eras.11041105---11061107# 14. Machine learning and foundation-model extension11081109A machine-learning component may be included, but it must not obscure identification.11101111## 14.1 Prediction task11121113Predict at shock time \(t\):11141115- future breadth;1116- future intensity;1117- time to cascade;1118- sector-to-market transition;1119- probability of a tail cascade.11201121## 14.2 Data splitting11221123Use strict temporal splits:11241125- training period;1126- validation period;1127- final untouched test period;1128- optional rolling-origin evaluation.11291130Never randomly split minute observations across train and test.11311132## 14.3 Baselines11331134Compare with:11351136- unconditional historical mean;1137- linear and logistic regressions;1138- penalized regression;1139- random forest;1140- gradient boosting;1141- HAR-style models;1142- VAR and sparse VAR;1143- Hawkes-inspired specifications where feasible;1144- LSTM;1145- temporal convolution;1146- transformer or state-space model;1147- graph neural network;1148- model without network information;1149- model without market-state information.11501151## 14.4 Evaluation11521153Report:11541155- out-of-sample \(R^2\);1156- log loss;1157- Brier score;1158- AUROC;1159- precision-recall AUC;1160- calibration;1161- tail recall;1162- economic ranking performance;1163- stability across years and regimes.11641165No trading strategy should be the main evidence. Any economic-value exercise must incorporate conservative transaction-cost assumptions and be clearly secondary.11661167## 14.5 Interpretability11681169Use:11701171- partial dependence;1172- accumulated local effects;1173- SHAP with dependence-aware interpretation;1174- feature ablation;1175- network ablation;1176- temporal ablation;1177- regime-specific performance;1178- calibration plots.11791180Do not treat model attention weights as causal explanations.11811182---11831184# 15. Required robustness analyses11851186At minimum, run:118711881. alternative return definitions;11892. alternative shock thresholds;11903. alternative factor residualizations;11914. alternative market and sector benchmarks;11925. alternative network definitions;11936. alternative network estimation windows;11947. alternative security universes;11958. exclusion of ETFs;11969. exclusion of ADRs;119710. exclusion of penny stocks;119811. exclusion of suspected split windows;119912. regular-hours only;120013. alternative minute aggregation: 1, 5, 10, and 15 minutes;120114. volume-weighted and equal-weighted outcomes;120215. opening and closing exclusions;120316. crisis versus non-crisis samples;120417. macro-announcement exclusions;120518. day-block bootstrap;120619. matched placebo shocks;120720. permuted networks;120821. pre-trend and pre-response tests;120922. source-stock fixed effects;121023. receiver-stock fixed effects;121124. date fixed effects;121225. sector-by-date fixed effects where computationally feasible;121326. balanced versus unbalanced samples;121427. winsorized versus non-winsorized outcomes;121528. median-based robust standardization;121629. alternative right-censoring treatment for propagation time;121730. multiple-testing adjustments.12181219Robustness must test the economic interpretation, not merely produce many similar regressions.12201221---12221223# 16. Falsification tests12241225The project must attempt to disprove its own central claim.12261227Required falsifications include:12281229- no systematic response before the source shock;1230- weak propagation through randomized networks;1231- weaker response among economically unrelated stocks;1232- stronger response after genuine shocks than matched placebo times;1233- no apparent secular acceleration caused solely by changing sample size;1234- no apparent acceleration caused solely by timestamp precision or coverage changes;1235- no result driven by a few crisis days;1236- no result driven by the largest technology stocks;1237- no result driven solely by ETFs;1238- no result generated mechanically by market-factor construction;1239- no result driven by corporate actions.12401241If a falsification fails, do not hide it. Reassess the claim.12421243---12441245# 17. Descriptive tables required12461247Generate the following main-paper candidates.12481249## Table I — Sample construction12501251Include by decade or period:12521253- raw rows;1254- raw symbols;1255- valid trading days;1256- duplicate removals;1257- invalid OHLC removals;1258- suspected corporate-action exclusions;1259- final stock-minute observations;1260- final securities;1261- median valid minutes per stock-day.12621263## Table II — Summary statistics12641265For core variables:12661267- mean;1268- standard deviation;1269- p1;1270- p5;1271- median;1272- p95;1273- p99;1274- number of observations.12751276Report at the economically appropriate unit to avoid meaningless precision.12771278## Table III — Characteristics of source shocks12791280Compare:12811282- positive and negative shocks;1283- low- and high-volatility states;1284- early and late sample;1285- cascade and non-cascade shocks.12861287## Table IV — Baseline propagation12881289Show horizon-specific coefficients for breadth, intensity, and persistence.12901291## Table V — Nonlinearity and asymmetry12921293Show shock-size bins, negative interaction, and market-state interactions.12941295## Table VI — Evolution over time12961297Show flexible period or year interactions and tests of equality.12981299## Table VII — Network mechanisms13001301Compare connected and unconnected receivers and transmitter centrality.13021303## Table VIII — Identification and placebo tests13041305Include macro exclusions, matched placebo times, and permuted networks.13061307Keep the main manuscript selective. Move secondary tables to the Internet Appendix.13081309---13101311# 18. Figures required13121313Create journal-quality figures in PDF and high-resolution PNG.13141315## Figure 1 — Data coverage13161317Panels may show:13181319- number of securities by year;1320- number of observations by year;1321- median bars per stock-day;1322- missingness rate.13231324## Figure 2 — Intraday seasonality13251326Show minute-of-day patterns for:13271328- volatility;1329- volume;1330- range;1331- shock incidence.13321333## Figure 3 — Anatomy of a representative cascade13341335A carefully selected event visualization with:13361337- source shock;1338- sector response;1339- market breadth;1340- network diffusion;1341- confidence or placebo envelope where possible.13421343Do not cherry-pick without a transparent selection rule.13441345## Figure 4 — Impulse responses13461347Local-projection responses over 1–120 minutes for:13481349- breadth;1350- intensity;1351- signed returns;1352- volatility.13531354## Figure 5 — Nonlinear propagation13551356Plot propagation against source-shock magnitude with confidence bands.13571358## Figure 6 — Positive-negative asymmetry13591360Show comparable positive and negative shocks.13611362## Figure 7 — Evolution over time13631364Show annual or rolling estimates of propagation speed and breadth.13651366## Figure 8 — Network position and cascade risk13671368Show economically interpretable bins of lagged source centrality.13691370## Figure 9 — Market-state dependence13711372Compare calm and stressed states.13731374## Figure 10 — Falsification evidence13751376Real versus matched placebo and real versus permuted-network responses.13771378---13791380# 19. Figure style: Journal of Finance quality13811382The phrase “JF style” means restrained, legible, publication-quality academic graphics, not imitation of proprietary branding.13831384Use:13851386- vector PDF as the primary format;1387- embedded fonts;1388- white background;1389- minimal chart ink;1390- no 3D effects;1391- no gradients;1392- no decorative icons;1393- no unnecessary borders;1394- no chart titles inside the plotting region unless essential;1395- concise axis labels;1396- consistent font family and sizing;1397- grayscale-safe line types and markers;1398- colorblind-safe colors when color is needed;1399- direct labels where clearer than legends;1400- confidence bands with restrained opacity;1401- panel labels `(a)`, `(b)`, etc.;1402- identical scales across comparable panels;1403- economically meaningful tick marks;1404- no scientific notation unless unavoidable;1405- no excessive decimal places.14061407Every figure must include a LaTeX note explaining:14081409- sample;1410- variable definitions;1411- estimation method;1412- confidence intervals;1413- clustering or bootstrap;1414- normalization;1415- exclusions.14161417Save source data for every figure as Parquet or CSV in:14181419```text1420outputs/figures/data/1421```14221423A figure must be reproducible from its source-data file without rerunning the entire empirical pipeline.14241425---14261427# 20. LaTeX table style14281429Use `booktabs`. Avoid vertical rules. Avoid excessive horizontal rules.14301431Preferred packages:14321433```latex1434\usepackage{booktabs}1435\usepackage{threeparttable}1436\usepackage{siunitx}1437\usepackage{adjustbox}1438\usepackage{longtable}1439\usepackage{tabularx}1440\usepackage{pdflscape}1441```14421443Requirements:14441445- decimal alignment;1446- consistent coefficient precision;1447- standard errors in parentheses;1448- significance stars only as a secondary visual convention;1449- report exact or thresholded p-values where useful;1450- clearly label fixed effects;1451- clearly label clustering;1452- include sample size;1453- include within or adjusted \(R^2\) as appropriate;1454- include dependent-variable mean when informative;1455- include economic magnitude;1456- use clear panel headings;1457- use detailed table notes;1458- avoid shrinking text below readability.14591460Do not use stars as the main argument. Emphasize coefficient magnitude, confidence intervals, and economic interpretation.14611462Generate each table in two forms:146314641. standalone `.tex`;14652. machine-readable `.csv` or `.parquet`.14661467All table numbers must come directly from stored estimation objects.14681469---14701471# 21. Literature review standard14721473The literature review must be comprehensive, current, accurate, and organized around the paper’s contribution. It must not be a list of summaries.14741475## 21.1 Literature streams14761477Build a literature map covering at least:147814791. market liquidity and liquidity risk;14802. commonality in liquidity;14813. market microstructure and price discovery;14824. high-frequency trading and algorithmic trading;14835. market fragility and flash events;14846. cross-asset price impact and cross-impact;14857. volatility spillovers and contagion networks;14868. lead-lag relationships;14879. intraday jumps;148810. Hawkes processes and self-excitation;148911. factor models for high-frequency returns;149012. electronic-market design and fragmentation;149113. ETF, passive, and portfolio-trading effects;149214. systemic risk and network centrality;149315. machine learning in market microstructure;149416. measurement using OHLCV versus trades and quotes.14951496## 21.2 Source quality hierarchy14971498Prioritize:149915001. published articles in leading finance, economics, econometrics, and statistics journals;15012. authoritative working papers from NBER, CEPR, central banks, BIS, and established scholars;15023. official exchange, SEC, Federal Reserve, and regulatory documents;15034. recent working papers only when directly relevant and clearly labeled as unpublished.15041505Do not rely on blogs, generic websites, generated summaries, or low-quality outlets for core claims.15061507## 21.3 Citation verification15081509For every cited work:15101511- verify title;1512- verify authors;1513- verify year;1514- verify journal or working-paper series;1515- verify volume, issue, and pages where applicable;1516- verify DOI where available;1517- verify that the paper actually supports the sentence;1518- save the source and citation metadata;1519- never cite a work based only on another paper’s bibliography.15201521Maintain `docs/LITERATURE_MAP.md` with columns:15221523- citation key;1524- full citation;1525- literature stream;1526- research question;1527- data;1528- frequency;1529- identification;1530- main result;1531- relevance;1532- difference from our paper;1533- verified source;1534- verification date.15351536## 21.4 Literature review argument15371538The review should establish:153915401. liquidity and volatility exhibit common components;15412. shocks can transmit across assets through information and trading channels;15423. high-frequency market structure may improve normal-times efficiency while creating tail fragility;15434. prior work often studies limited instruments, shorter samples, quotes/order books, particular episodes, or lower-dimensional networks;15445. the present paper contributes a long historical, broad cross-section of minute-level U.S. equity data;15456. OHLCV breadth comes with measurement limitations;15467. the paper’s novelty lies in the joint analysis of speed, breadth, persistence, asymmetry, time evolution, and network conditioning.15471548Do not overclaim uniqueness until a documented literature search confirms it.15491550---15511552# 22. Paper outline15531554The paper must be written in English.15551556## 22.1 Abstract15571558Target approximately 120–180 words unless the current submission rules require otherwise.15591560The abstract must contain:156115621. question;15632. data;15643. method;15654. central result;15665. mechanism;15676. implication.15681569Do not include vague claims such as “significant implications.” State the economic fact.15701571## 22.2 Introduction15721573The introduction should be approximately 4–6 manuscript pages and follow this logic:15741575### Paragraph 1 — Economic problem1576Explain why intraday propagation matters for price discovery, liquidity, execution, and systemic stability.15771578### Paragraph 2 — Tension1579Present the core tension: faster and more liquid markets may absorb ordinary information efficiently but transmit extreme shocks rapidly.15801581### Paragraph 3 — Research question1582State exactly what the paper asks.15831584### Paragraph 4 — Data and empirical setting1585Describe the broad minute-level panel, while flagging that validation determines the final sample.15861587### Paragraphs 5–7 — Main findings1588Present only verified findings, including economic magnitudes and uncertainty.15891590### Paragraph 8 — Identification1591Explain how the paper distinguishes propagation from common information.15921593### Paragraph 9 — Mechanisms1594Summarize network, market-state, and asymmetry evidence.15951596### Paragraphs 10–12 — Contributions1597State contributions relative to specific literature streams.15981599### Final paragraph — Road map1600Keep it brief.16011602The introduction must not read like a technical report. It must build an economic argument.16031604## 22.3 Literature review16051606Integrate literature with hypotheses and contributions. Avoid a long disconnected catalog.16071608## 22.4 Data16091610Explain:16111612- provenance;1613- raw structure;1614- coverage;1615- audit;1616- filters;1617- timestamp convention;1618- corporate-action handling;1619- security identity;1620- survivorship concerns;1621- descriptive statistics;1622- limitations.16231624## 22.5 Measurement16251626Define:16271628- residual shocks;1629- propagation outcomes;1630- networks;1631- market states;1632- centrality;1633- cascade thresholds.16341635## 22.6 Empirical design16361637Explain:16381639- local projections;1640- event-study design;1641- fixed effects;1642- inference;1643- timing;1644- identification assumptions;1645- threats.16461647## 22.7 Results16481649Organize by economic question, not by code execution order.16501651Suggested sequence:165216531. basic propagation;16542. nonlinear threshold;16553. asymmetry;16564. time evolution;16575. network mechanism;16586. market-state dependence;16597. identification and placebo evidence.16601661## 22.8 Robustness16621663Keep the main text focused. Refer to a detailed Internet Appendix.16641665## 22.9 Conclusion16661667Do not merely summarize. State:16681669- what was learned;1670- what remains unidentified;1671- implications for market design and risk monitoring;1672- limits of OHLCV data;1673- future work with quote and order-book data.16741675---16761677# 23. Writing standards16781679The paper must read like a finance paper, not AI-generated prose.16801681Use:16821683- precise claims;1684- active voice where natural;1685- disciplined paragraphs;1686- explicit economic magnitudes;1687- clear transitions;1688- restrained tone;1689- correct terminology;1690- consistent notation.16911692Avoid:16931694- hype;1695- “revolutionary”;1696- “groundbreaking” unless quoting;1697- “unprecedented” without proof;1698- “leveraging” as filler;1699- “delve”;1700- “multifaceted”;1701- repeated road maps;1702- excessive bullet points in the manuscript;1703- paragraphs that merely describe tables;1704- causal verbs without identification;1705- statements that statistical significance proves importance.17061707Every empirical paragraph should answer:170817091. What is estimated?17102. What is the sign and magnitude?17113. How uncertain is it?17124. What is the economic interpretation?17135. Which competing explanation remains?17146. Which robustness test addresses it?17151716---17171718# 24. JF submission preparation17191720At the time of final preparation, verify the current official Journal of Finance submission instructions.17211722The working format should conservatively use:17231724- 12-point font;1725- at least 1.5 line spacing;1726- readable margins;1727- page numbers;1728- anonymous manuscript version;1729- separate title page if required;1730- main manuscript no longer than the current JF limit;1731- separate Internet Appendix;1732- consistent references;1733- tables and figures placed according to current instructions.17341735Do not assume old formatting rules remain current. Record the date and source of the final guideline check in:17361737```text1738docs/DECISION_LOG.md1739```17401741Create:17421743```text1744outputs/release/1745├── manuscript_anonymous.pdf1746├── manuscript_named.pdf1747├── internet_appendix.pdf1748├── title_page.pdf1749├── cover_letter.tex1750├── cover_letter.pdf1751├── source_archive.zip1752├── replication_manifest.json1753└── submission_checklist.md1754```17551756The cover letter must be factual and restrained. Do not claim acceptance probability or editorial fit without evidence.17571758---17591760# 25. Main paper versus Internet Appendix17611762## Main paper17631764Reserve for:17651766- core question;1767- central descriptive evidence;1768- primary identification;1769- main economic mechanism;1770- essential robustness;1771- clear contribution.17721773## Internet Appendix17741775Include:17761777- extended data audit;1778- full sample-construction details;1779- additional variable definitions;1780- alternative filters;1781- alternative networks;1782- alternative thresholds;1783- additional event studies;1784- placebo distributions;1785- machine-learning architecture;1786- hyperparameters;1787- computational details;1788- additional tables and figures;1789- proofs or derivations;1790- replication details.17911792Every appendix item must be referenced from the main paper where relevant.17931794---17951796# 26. Reproducibility and provenance17971798## 26.1 Configuration17991800All parameters must live in versioned YAML files, not scattered through scripts.18011802## 26.2 Random seeds18031804Set and log seeds for:18051806- sampling;1807- bootstraps;1808- permutations;1809- machine learning;1810- graph estimation;1811- simulations.18121813## 26.3 Versioning18141815Record:18161817- Git commit;1818- Python version;1819- package versions;1820- OS;1821- hardware;1822- dataset revision;1823- external-data revision;1824- command line;1825- start and end time.18261827## 26.4 Data lineage18281829Each processed file must record:18301831- parent inputs;1832- transformation version;1833- filters;1834- row counts before and after;1835- creation timestamp;1836- code commit.18371838## 26.5 No manual numerical edits18391840LaTeX tables, abstract numbers, introduction magnitudes, and conclusion numbers must be populated from a central results registry such as:18411842```text1843outputs/estimates/results_registry.json1844```18451846Implement macros such as:18471848```latex1849\newcommand{\MainBreadthEffect}{...}1850```18511852generated automatically from verified outputs.18531854---18551856# 27. Compute strategy for a cluster18571858The pipeline must be scalable but safe.18591860## 27.1 Preferred data tools18611862Consider:18631864- PyArrow;1865- Polars lazy execution;1866- DuckDB;1867- Dask or Ray only where they add value;1868- partitioned Parquet;1869- predicate pushdown;1870- memory-mapped reads;1871- streaming aggregations.18721873Avoid loading the full dataset into RAM.18741875## 27.2 Partitioning18761877Partition processed data by a combination of:18781879- year;1880- month;1881- trading date;1882- symbol bucket.18831884Select partitions based on dominant query patterns.18851886## 27.3 Distributed jobs18871888Every distributed task must:18891890- be idempotent;1891- write to temporary paths;1892- atomically rename on success;1893- save checkpoints;1894- avoid multiple nodes writing the same final file;1895- emit structured logs;1896- support retry;1897- validate output before success.18981899## 27.4 Resource benchmarking19001901Before a full run:190219031. benchmark one month;19042. benchmark one year;19053. estimate CPU hours;19064. estimate memory;19075. estimate temporary storage;19086. estimate final storage;19097. record in `docs/COMPUTE_LOG.md`.19101911Optimization must not change estimands or silently alter precision.19121913---19141915# 28. Testing and validation19161917## 28.1 Unit tests19181919Test:19201921- timestamp conversion;1922- session assignment;1923- return calculation;1924- shock detection;1925- lag discipline;1926- no future-data leakage;1927- network-window cutoffs;1928- breadth denominator;1929- session-boundary horizons;1930- table formatting.19311932## 28.2 Synthetic-data tests19331934Generate small synthetic panels with known:19351936- common shocks;1937- idiosyncratic shocks;1938- directed propagation;1939- no propagation;1940- positive-negative asymmetry;1941- missing bars;1942- split-like jumps.19431944Confirm that the pipeline recovers the intended patterns and rejects false ones.19451946## 28.3 Release validation19471948The release script must fail if:19491950- a table references a missing result;1951- a figure is missing source data;1952- LaTeX does not compile;1953- citations are unresolved;1954- labels are duplicated;1955- page count exceeds the current target;1956- anonymous PDF contains author-identifying text;1957- manuscript claims disagree with the results registry;1958- data revision is not pinned;1959- tests fail.19601961---19621963# 29. Decision log19641965Maintain `docs/DECISION_LOG.md`.19661967For every major choice, record:19681969- date;1970- decision;1971- alternatives considered;1972- empirical reason;1973- theoretical reason;1974- effect on interpretation;1975- whether pre-specified or data-driven;1976- code/config reference.19771978Important decisions include:19791980- timestamp convention;1981- regular-session definition;1982- security filters;1983- shock threshold;1984- network window;1985- factor model;1986- clustering;1987- periodization;1988- figure selection;1989- main versus appendix placement.19901991---19921993# 30. Minimum deliverables for the first phase19941995The first phase is complete only when all the following exist.19961997## Data acquisition19981999- [ ] repository metadata report;2000- [ ] pinned dataset revision;2001- [ ] storage estimate;2002- [ ] resumable downloader;2003- [ ] raw manifest;2004- [ ] checksum verification;2005- [ ] three-month pilot sample;2006- [ ] full-download plan.20072008## Audit20092010- [ ] schema report;2011- [ ] timestamp report;2012- [ ] calendar report;2013- [ ] OHLC validity report;2014- [ ] duplicate report;2015- [ ] missingness report;2016- [ ] corporate-action diagnostic;2017- [ ] security-universe diagnostic;2018- [ ] survivorship assessment;2019- [ ] limitations memo.20202021## Prototype analysis20222023- [ ] clean pilot panel;2024- [ ] intraday seasonality figure;2025- [ ] sample-coverage figure;2026- [ ] preliminary shock definition;2027- [ ] preliminary cascade measure;2028- [ ] one transparent baseline regression;2029- [ ] one placebo test;2030- [ ] one LaTeX summary-statistics table;2031- [ ] one LaTeX baseline-results table.20322033## Paper infrastructure20342035- [ ] compiling `main.tex`;2036- [ ] compiling Internet Appendix;2037- [ ] verified bibliography workflow;2038- [ ] literature map;2039- [ ] introduction outline;2040- [ ] data-section draft based only on verified facts;2041- [ ] automated figure and table inclusion.20422043---20442045# 31. Stop conditions20462047Stop and write a diagnostic report rather than proceeding if:20482049- timestamps cannot be interpreted reliably;2050- OHLC values are extensively inconsistent;2051- split adjustments cannot be handled sufficiently for return analysis;2052- the universe is severely survivorship-biased and cannot be characterized;2053- historical coverage is too sparse for the claimed design;2054- symbol identity errors dominate long-run comparisons;2055- data licensing prevents the intended use;2056- results depend on undocumented transformations;2057- common-news contamination cannot be separated even approximately;2058- inference is computationally impossible at the intended unit without changing the design.20592060A rigorous negative assessment is preferable to a polished but invalid paper.20612062---20632064# 32. Commands and workflow20652066Create a Makefile with targets similar to:20672068```make2069probe2070download-pilot2071audit-pilot2072download-full2073verify-data2074build-panel2075build-features2076detect-shocks2077estimate-networks2078estimate-baseline2079estimate-events2080run-placebos2081run-robustness2082figures2083tables2084paper2085appendix2086validate2087release2088```20892090A complete reproducible run should ultimately be possible through:20912092```bash2093make release2094```20952096Do not run `make release` until the pilot audit and full-data checks pass.20972098---20992100# 33. Immediate actions21012102Begin now with the following actions, in order:210321041. Create the project directory structure.21052. Initialize environment and dependency files.21063. Inspect the Hugging Face dataset repository without downloading all files.21074. Pin the exact dataset revision.21085. Create a repository-level file inventory.21096. Estimate total download and processed-storage needs.21107. Download three strategically selected monthly files:2111 - one early month;2112 - one middle-sample month;2113 - one recent month.21148. Audit schema and timestamps.21159. Produce `docs/DATA_AUDIT.md`.211610. Produce `docs/LIMITATIONS.md`.211711. Produce pilot coverage and intraday-seasonality figures.211812. Produce one summary-statistics LaTeX table.211913. Build `docs/LITERATURE_MAP.md` and begin a verified literature search.212014. Create a compiling LaTeX manuscript skeleton.212115. Draft only the portions of the Data section supported by the audit.212216. Report blockers and proposed solutions before launching the full download.21232124At the end of each major stage, produce:21252126```text2127STATUS.md2128```21292130with:21312132- completed work;2133- generated files;2134- key findings;2135- quality-control results;2136- unresolved issues;2137- next commands;2138- decisions requiring attention.21392140Do not ask for confirmation for routine implementation choices. Make the most defensible choice, record it, and continue. Escalate only issues that materially alter the research question, data legality, identification, or interpretation.21412142---21432144# 34. Final quality criterion21452146The paper is not “ready for JF” merely because it compiles or contains many analyses.21472148It is ready for serious submission consideration only if:21492150- the data are credible and fully audited;2151- the paper documents a new and economically important fact;2152- propagation is separated as far as possible from common information;2153- magnitudes are economically meaningful;2154- results survive serious falsification tests;2155- the mechanism is coherent;2156- the literature positioning is accurate;2157- the writing is disciplined;2158- the figures and tables are publication quality;2159- all claims are reproducible;2160- limitations are explicit;2161- the main paper is focused;2162- the Internet Appendix is complete;2163- no result has been fabricated, selectively hidden, or overstated.21642165The standard is not to make the paper appear publishable. The standard is to determine honestly whether the evidence supports a contribution worthy of a top finance journal and, if it does, to present that contribution with maximum clarity and rigor.2166