Does Filter Shape Matter for Momentum Crashes? Pre-registered Evidence from US Industry Portfolios, 1926–2026

Preprint and replication package. Momentum signals are linear filters of past returns, yet whether the shape of that filter affects crash risk has, to the author's knowledge, not been tested. This study holds the effective lookback horizon fixed at a mean lag of 137 trading days and varies only the filter's shape, and therefore its frequency response. Monthly cross-sectional momentum strategies are built on 48 US industry portfolios from daily returns, 1929–2026 (1,172 monthly holding periods). The hypotheses, variables and tests were pre-registered before any returns were analysed (AsPredicted #320713, https://aspredicted.org/ip7x9e.pdf). Neither pre-registered hypothesis is supported: a Gaussian low-pass filter does not reduce standardised tail loss relative to the rectangular 12-1 filter or to an exponential filter (Romano-Wolf p = 0.79). Volatility scaling, by contrast, reduces tail loss by 0.32 more than the change of filter shape (95% CI −0.52 to −0.12). Within the class of linear filters, the shape of the signal does not substitute for managing exposure to recent risk. Files: (1) the preprint (PDF); (2) replication_package.zip with the analysis code, data download script, checksums of the source files, all results and run logs, the literature search log, the list of deviations from the pre-registration, and a copy of the pre-registration. Source data (Kenneth French Data Library) are not redistributed. Use of AI: the author used Claude (Anthropic) for search design, analysis code, drafting and language editing, as detailed in the manuscript's declarations. The author reviewed all content and takes full responsibility for it.