Jfm Internet Appendix

Internet Appendix & Replication (jfm-internet-appendix)

When to trigger

• The main text is overflowing with secondary robustness tables that should move to an appendix
• A referee asked for a check that does not deserve main-text space but must be available
• The liquidity-measure construction and sample filters are described too thinly to be reproduced
• You are preparing the replication package and must document a proprietary or restricted data feed
• It is unclear which results are load-bearing (main text) vs. supporting (appendix)

What belongs where

The discipline is the main paper must be self-contained: every claim a referee must accept to believe the headline lives in the main text. The Internet Appendix carries depth, not load-bearing claims.

Building the replication package (Elsevier / JFM context)

JFM is an Elsevier journal; confirm the current data-and-code policy and whether a deposit is required at submission vs. acceptance (检索于 2026-06；以官网为准 — 待核实). Regardless of the minimum, prepare:

1. Data lineage: raw feed (TAQ / LOBSTER / proprietary) → cleaning scripts → analysis files, with a README mapping each table/figure to the script that builds it.
2. Codebook: every variable, its construction (effective spread formula, impact horizon, sign-classification rule), and every filter with its rationale.
3. Proprietary-data path: if the feed cannot be redistributed (e.g., a vendor or exchange feed), document the access path — vendor, license, query/sample definition — so a replicator can obtain equivalent data. Never imply data can be shared that cannot.
4. Environment: software versions and a lockfile so estimates reproduce (PIN/lambda estimation is sensitive to solver settings).
5. Map from appendix to claim: each appendix table states which main-text sentence it supports.

Worked self-containment test (illustrative)

Read only the main text and ask: could a referee accept the headline without ever opening the appendix? If the only evidence that the result is not a measurement artifact sits in Internet Appendix Table IA.7, the paper fails — move at least one alternative-measure column into the main text and leave the full battery in the appendix. Conversely, if the main text reproduces every sub-period split and influential-name drop, it is bloated — push those down and keep a one-line summary ("results are stable across sub-periods and to dropping the most-traded names; see IA.2-IA.4"). The main text carries the claims; the appendix carries the exhaustiveness.

Proprietary-data documentation patterns

Microstructure papers routinely use feeds that cannot be redistributed (exchange order-book feeds, a broker's order data, Nasdaq ITCH, a vendor's TAQ license). The honest pattern is to document the access path precisely: the vendor/exchange, the product name, the license type, the exact query or universe/period filter used, and the cost tier if relevant. A replicator with their own license can then reconstruct the sample. State plainly what is shareable (code, derived non-proprietary aggregates) and what is not. Do not write "data available on request" for data you have no right to share — that is a common and easily caught evasion.

Checklist

• Main text is self-contained: no headline-supporting claim lives only in the appendix
• Each appendix exhibit names the main-text claim it backs
• Full liquidity-measure construction and filter rules are in the codebook
• README maps every table/figure to its build script
• Proprietary/restricted data has a documented access path, not a false promise of sharing
• Software versions / lockfile recorded so numerically sensitive estimates reproduce
• Data-and-code policy specifics verified or marked 待核实