Robustness & Multiple Testing (jf-robustness)

When to trigger

• The main result is in; you must decide which robustness checks to run and where to put them
• A referee will ask "does it survive [alternative measure / subsample / specification]?"
• The finding is one of many you tested and multiple testing is a live concern

JF norm: decisive in the body, exhaustive in the Internet Appendix

JF favors an accessible body within the 60-page limit. Put the 3–6 decisive checks in the main text and move the exhaustive battery to the Internet Appendix, which is bundled at the end of the same PDF and does not count toward 60 pages (see jf-internet-appendix). This main-text/IA split is a JF hallmark; do not bury a load-bearing check.

Multiple testing — a JF-salient concern

JF published the canonical "factor zoo" critique (Harvey, Liu & Zhu, "…and the Cross-Section of Expected Returns," JF). Reviewers therefore expect:

• An honest account of how many specifications/signals were tried.
• Adjusted thresholds (e.g., higher t-cutoffs, FDR control) for any discovery mined from many candidates, not the naive 1.96.
• Pre-registration-style discipline in framing (no HARKing).