finance¶
reviewprivate (curator-owned)referee-simulationCurator-private skill — copy text from 100xOS/shared/skills/review/finance.md.
Finance Paper Review Checklist¶
Overview¶
Quality checklist for reviewing empirical finance papers. Covers methodology, data, presentation, and common pitfalls specific to financial economics research.
Data Quality¶
Sample Construction¶
- Sample selection criteria clearly stated and justified.
- Universe defined (CRSP, Compustat, specific exchange, country).
- Filters documented: minimum price, minimum market cap, exclusion of financials/utilities, share type (ordinary common equity only).
- Sample period justified. Long enough for reliable inference (typically 20+ years for asset pricing, 5+ years for corporate finance).
- Number of observations, firms, and time periods reported.
Survivorship Bias¶
- Includes delisted firms (not just survivors).
- Uses delisting returns from CRSP (not just dropping firms at delisting).
- If using a database that only covers survivors (e.g., some international databases), acknowledge the bias.
Look-Ahead Bias¶
- Accounting data available to investors at portfolio formation (use lagged data: e.g., fiscal year-end data available 6 months later — June sorting convention).
- No future information used in variable construction.
- Real-time data vintages used for macro variables if relevant.
Data Cleaning¶
- Outlier treatment documented (winsorization at 1/99% or trimming). Consistent across all variables.
- Penny stocks excluded or results shown with and without (price > $1 or $5).
- Micro-caps addressed (NYSE size breakpoints, or explicit exclusion of bottom quintile by ME).
- Missing data handling explained (how are missing returns, missing accounting data treated?).
Methodology¶
Standard Errors¶
- Appropriate standard errors for the setting:
- Time-series regressions: Newey-West HAC (specify lag length).
- Panel regressions: clustered by firm and/or time (Petersen 2009).
- Fama-MacBeth: Newey-West on the time series of cross-sectional slopes (or Shanken correction).
- Event studies: BMP or Kolari-Pynnonen.
- Justification for clustering choice. Petersen (2009) guidance: cluster on the dimension with fewer clusters.
- NOT using simple OLS standard errors for panel data with firm or time effects.
Return Computation¶
- Log returns vs simple returns: stated and consistent. Log returns for time-series analysis; simple returns for portfolio aggregation.
- Compounding method for multi-period returns: buy-and-hold (BHAR) vs cumulative abnormal (CAR). State which and why.
- Excess returns: clearly defined (in excess of risk-free rate, typically 1-month T-bill).
- Currency: returns in local currency or USD, and how currency effects are handled.
Factor Model Specification¶
- Factor model choice justified (CAPM, FF3, FF5, q-factor, etc.).
- Factor data source cited (Ken French library, AQR, self-constructed).
- Results robust to alternative factor models (show at least two).
- If self-constructed factors: construction methodology detailed and matches standard conventions.
Portfolio Construction¶
- Sorting variable, breakpoints, and rebalancing frequency documented.
- NYSE breakpoints used for CRSP sorts (avoids micro-cap domination).
- Value-weighted returns as primary (equal-weighted as robustness — EW overweights micro-caps).
- Holding period and skip-month convention for momentum documented.
Endogeneity and Identification¶
- Potential endogeneity concerns discussed.
- If causal claims made: identification strategy clearly stated (IV, DiD, RDD, natural experiment).
- Instruments validated: relevance (first-stage F > 10) and exclusion restriction argued.
- Reverse causality addressed.
Statistical Issues¶
Multiple Testing¶
- If testing multiple hypotheses: acknowledge the problem.
- Apply appropriate corrections (Bonferroni, Holm, FDR/Benjamini-Hochberg).
- For new anomalies: use t > 3.0 threshold (Harvey-Liu-Zhu 2016) or bootstrap.
- Pre-registration or holdout sample if feasible.
Economic vs Statistical Significance¶
- Report economic magnitudes (basis points per month, percent per year, dollar amounts).
- Discuss whether the effect is large enough to matter after transaction costs.
- For trading strategies: account for bid-ask spreads, market impact, short-selling costs, and capacity constraints.
- Sharpe ratio or information ratio for strategy performance.
Small Sample / Power¶
- Is the sample large enough for the tests used?
- For GRS tests: N (test assets) should be much less than T (time periods).
- For Fama-MacBeth: at least 200+ months for reliable inference.
- For event studies: sufficient events for cross-sectional tests (N > 30 minimum, ideally 100+).
Non-Stationarity¶
- Unit root tests performed if using price levels, valuation ratios, or macro variables.
- Stambaugh (1999) bias addressed if using persistent regressors in predictive regressions.
- Regime changes or structural breaks considered (pre/post crisis, regulatory changes).
Presentation¶
Tables¶
- Summary statistics table: N, mean, median, SD, min, max, percentiles for all key variables.
- Correlation matrix for main variables.
- Regression tables: coefficients, t-statistics (or standard errors), R^2, N, fixed effects indicators.
- Portfolio sort tables: mean return, alpha, t-statistics, monotonicity pattern, high-minus-low spread.
- Standard errors in parentheses, t-statistics in brackets — be consistent.
- Significance stars: * p<0.10, ** p<0.05, *** p<0.01 (state convention).
Figures¶
- Time series of key variables or cumulative returns.
- Event study: cumulative average abnormal returns with confidence bands.
- Portfolio performance: cumulative wealth plot.
- Coefficient stability across subperiods or specifications (coefficient plot).
Reporting Conventions¶
- Returns in percent (not decimals). State frequency (daily, monthly, annual).
- Report annualized returns and volatilities where appropriate (multiply monthly by 12 and sqrt(12)).
- Market cap in millions or billions. State currency and date for nominal values.
- Book-to-market, not market-to-book (Fama-French convention).
- Basis points (bps) for small return differentials (1 bp = 0.01%).
Common Pitfalls in Finance Papers¶
-
Cross-sectional correlation: Fama-MacBeth addresses this, but pooled OLS with firm-clustered SEs does not account for time effects. Use double-clustering (Petersen 2009) or Fama-MacBeth.
-
Microstructure noise: Using daily closing prices for high-frequency analysis introduces bid-ask bounce. Use midpoint prices or apply appropriate filters.
-
Rebalancing assumptions: Monthly rebalanced portfolios assume zero transaction costs. Discuss turnover and implementation costs.
-
Short-selling constraints: Long-short strategies assume costless shorting. Discuss short interest, lending fees, and institutional constraints.
-
Data snooping: The finance "factor zoo" problem. New anomalies must clear a high bar. Pre-specify hypotheses.
-
Stale prices: Illiquid stocks have stale prices that induce spurious predictability and autocorrelation. Use the Dimson (1979) beta correction.
-
Penny stock bias: Many anomalies concentrate in micro-cap and penny stocks. Results should survive exclusion of stocks below $5 or below the 20th NYSE size percentile.
-
Backfill bias: Databases may backfill historical data when a firm is added, creating upward bias in performance.
-
Index inclusion effects: S&P 500 additions/deletions have well-documented price effects. Control for or exclude these events if they contaminate the sample.
-
Seasonality: January effect, turn-of-month, day-of-week effects. Check if results are driven by calendar regularities.
Key References¶
- Petersen, M.A. (2009). Estimating standard errors in finance panel data sets: Comparing approaches. Review of Financial Studies.
- Harvey, C.R., Liu, Y., and Zhu, H. (2016). ... and the cross-section of expected returns. Review of Financial Studies.
- Stambaugh, R.F. (1999). Predictive regressions. Journal of Financial Economics.
- Hou, K., Xue, C., and Zhang, L. (2020). Replicating anomalies. Review of Financial Studies.
- Fama, E.F. and French, K.R. (2008). Dissecting anomalies. Journal of Finance.
- Cochrane, J.H. (2011). Presidential address: Discount rates. Journal of Finance.
- McLean, R.D. and Pontiff, J. (2016). Does academic research destroy stock return predictability? Journal of Finance.