01 Home02 Research03 Projects04 Notes05 About06 Lab
Home/Notes/Quantitative research

A momentum signal times momentum, not the market

The same crash-exit rule cost SPY and SOXX return but added 1.3 points a year on a momentum ETF, in the order of how much each one holds the factor.

Topic
Quantitative research
Analysis
Sep 2026
Status
Working note
Related
Helix

I built a regime signal on a self-constructed momentum factor. When the factor’s trailing return sits more than two standard deviations above its own five-year history, momentum has been unusually strong and a sharp reversal becomes more likely. On the factor itself it holds up: across 2015–2026 the factor fell in 28 of 39 alert spells against a 47% base rate, and the result survives a correction for all 160 thresholds and holding rules I searched (adjusted p = 0.015).

Then I used it as a plain exit: hold an ETF, move to cash while the alert is on (6% of days), buy back when it clears. On SPY it cost 0.9 points of annual return and did no better than exits on random dates. On SOXX it cost 1.2 points and deepened the worst drawdown from −45.8% to −48.2%. On MTUM, a momentum ETF, it added 1.3 points a year, lifted the Sharpe ratio from 0.77 to 0.91 and beat 92% of randomly placed exits of the same shape.

The order is each ETF’s loading on the factor (R² of 0.26 for MTUM, 0.11 for SOXX, 0.03 for SPY), not its volatility or its return. A factor signal is a statement about the factor, and it reaches an asset only through that loading. Two limits: the MTUM gain rests on 2020 and 2024, and without both it turns negative; and no ETF’s 2022 drawdown, a rates sell-off rather than a momentum crash, was touched, because the alert never fired.