Too Big to Co-Move
Concentrated sectors co-move less with the rest of the market: a one-point rise in top-three share lowers monthly R² by 0.75 points (permutation p = 0.04), though concentration does not explain why markets disconnected.
Research / Calm Months, Crash Months
Calm Months, Crash Months: Why Volatility Management Pays for Momentum and Not the Market
Scaling a factor's exposure by the inverse of its recent variance raises many Sharpe ratios in Moreira and Muir (2017), but in real time most of the gains disappear and momentum survives. Why momentum, and why not the market, for which the rule was first proposed?
Key results
0.43 to 0.94
momentum's Sharpe ratio in real time, net of costs, July 1973 to August 2026 (Holm-adjusted p = 0.023)
0 of 5
other factors whose Sharpe ratio improves significantly; the market's falls from 0.51 to 0.48
76 to 23
momentum's largest loss at equal volatility, in percentage points
0.92
slope of realized on predicted momentum beta, CRSP 1932 to 2025, no fitted parameter (standard error 0.19)
29% / 71%
share of months that follow market declines, and their share of the covariance through which timing adds to momentum's return
0.78 to 0.97
managed momentum's Sharpe ratio across variance signals, block lengths, caps, scaling rules and start dates
Abstract
Built with only past data, capped leverage and trading costs, volatility management improves momentum, from a Sharpe ratio of 0.43 to 0.94 over July 1973 to August 2026, but none of the other Fama-French factors. We explain why. Sorting stocks on past returns loads momentum on the market in proportion to the market's own return over the formation window. Using Tweedie's formula, we derive this beta in closed form from two cross-sectional quantities, the dispersion of stock betas and the density of past returns at the portfolio breakpoints, and test it on CRSP with no fitted parameter. Against realized betas from 1932 to 2025, the prediction has a slope of 0.92 (standard error 0.19) and 79% of the explanatory power of an in-sample regression: it captures how momentum's beta moves, while the realized level sits 0.20 below the prediction. Months after market declines are 29% of the sample and supply 71% of the covariance through which timing adds to momentum's return. If the market's premium rises with its variance, the same logic implies that the market should not be volatility-timed; in our sample it is not improved by it.
Methods
- Every position uses only data available at the end of the prior month: an expanding-window scale constant, exposure capped at 1.5, and 14 basis points per unit of exposure change
- Variance from daily returns with a one-month half-life, chosen for momentum on 1973 to 2016 data among 24 variants fixed in advance
- Sharpe ratios compared with the Ledoit-Wolf studentized block bootstrap, Holm-adjusted across the six factors
- A closed-form formula for momentum's market beta via Tweedie's formula, tested on CRSP with no fitted parameter; the test was registered before the data were downloaded
What it does not show
- The formula captures how momentum's beta moves, not its level: realized betas sit 0.20 below the prediction, so a joint test of an exact prediction rejects.
- After the published samples end the gain is smaller, 0.21 to 0.44 over 116 months, and the change between the periods is not significant (p = 0.24).
- Net figures charge costs on changes in exposure, not on the trading inside each factor portfolio.
JEL: G11, G12, G14
Cite
BibTeX
@article{yadav2026calmmonths,
title = {Calm Months, Crash Months: Why Volatility Management Pays for Momentum and Not the Market},
author = {Yadav, Tanishk},
year = {2026},
journal = {SSRN Electronic Journal},
note = {Preprint, abstract 7113079}
}
Tanishk Yadav
MS Financial Engineering at New York University, graduating May 2027, and looking for full-time quantitative research roles. My master's thesis measures, in real time, when US debt dynamics turn self-reinforcing.
More research
Other papers
RateWalk: Forecasting Fed Decisions
Naive inflation-regime conditioning overfits; empirical-Bayes shrinkage fixes it in the US, UK and Germany, and a blend with a market proxy reaches log-loss 1.056 and a 72.7% hit rate.
VolEdge: The Variance Risk Premium
A full-stack platform that extracts risk-neutral moments from live option chains without a pricing model; a GARCH vol-targeted sleeve beat SPY on risk (Sharpe 0.85 vs 0.68) while naive short-variance harvesting lost money.
Multi-Scale Regime Detection
Four independent stress signals that must agree before de-risking: maximum drawdown -6.2% vs -41.7% for SPY, and the same calibration transfers to factor and international portfolios.