Quant Research

2. Check economic and asset-pricing logic

• Decide whether the idea is grounded in factor exposure, behavioral mispricing, structural friction, or market microstructure.
• Compare the idea against known factor families and asset-pricing intuition before testing.

3. Define the signal precisely

• Specify inputs, transformations, ranking logic, thresholds, lags, and implementation timing.
• Ensure the signal can be reproduced without hidden discretion.

4. Clean the data and define the test design

• Enforce point-in-time correctness.
• Check survivorship bias, look-ahead bias, stale fundamentals, restatement issues, and missing-data distortions.
• Define in-sample, out-of-sample, and validation logic before reviewing results.

5. Run the backtest and validation stack

• Evaluate return, risk, turnover, capacity, cost sensitivity, and benchmark-relative behavior.
• Stress the idea across subperiods, regimes, universes, and parameter ranges.
• Use the validation rules in references/validation-and-overfitting-defense.md.

6. Decompose what is really driving returns

• Determine whether performance comes from intended factor exposure, hidden beta, crowding, leverage, volatility selling, or timing luck.
• Use attribution and risk decomposition before calling the result alpha.

7. Make the research decision

• Conclude with one of: keep researching, conditionally promising, likely overfit, implementation weak, or reject.
• Use the output structure in references/output-contract.md.

Required References

• Read references/integrated-framework.md for the full research stack.
• Read references/validation-and-overfitting-defense.md for robustness and anti-overfitting rules.
• Read references/data-hygiene-risk-and-attribution.md for data controls, risk modeling, and attribution.
• Read references/output-contract.md for required output behavior.

Risk and Uncertainty Rules

• State when the sample is small, regime coverage is thin, or parameter sensitivity is high.
• State when evidence is suggestive rather than conclusive.
• Separate empirical stability from economic plausibility.
• Explicitly note when live implementation friction may erase paper alpha.

Anti-Hallucination Rules

• Do not fabricate performance metrics, factor loadings, transaction costs, or validation results.
• Tag statements as [actual], [inference], or [assumption].
• Use [actual] only for verified data or directly observed test output.
• Use [inference] for reasoned conclusions drawn from the evidence.
• Use [assumption] for scenario inputs, cost assumptions, capacity assumptions, or modeling choices.
• If the data quality or validation setup is weak, lower confidence rather than overstating the result.