Market timing models for a momentum strategy
A market timing model is a rule that decides how much exposure to hold, rather than which assets to hold. For a momentum strategy, timing has a specific purpose: to reduce exposure in the market conditions where momentum is known to perform badly. This article reviews the main families of timing models, the research behind them, and the costs they bring.
Educational content. Timing rules are described as research subjects, not as signals to follow.
Why time a momentum strategy at all?
Momentum's long-run record is strong, but its returns are uneven. Long-only momentum portfolios carry the full drop of a bear market; long-short versions suffer sharp crashes when markets rebound after a long fall. Both problems are linked to the state of the overall market. If that state can be identified, even roughly, exposure can be adjusted. The question is whether the identification is good enough to pay for the errors it inevitably makes.
Family 1: moving-average trend filters
The simplest timing model measures the market's price against its own moving average. Mebane Faber's 2007 paper "A Quantitative Approach to Tactical Asset Allocation" popularised a version using the 10-month simple moving average: hold an asset when its month-end price is above its 10-month average, move to cash when it is below. Applied to several asset classes over long histories, the rule kept returns similar to buy-and-hold while reducing volatility and the depth of drawdowns in his sample.
For a momentum strategy, the filter can be applied to the broad market rather than to each stock: run the momentum portfolio when the market index is above its long moving average, and reduce or remove exposure otherwise. The same idea underlies the 200-day moving average used by many practitioners. We build the mechanics of a moving-average rule step by step in our spreadsheet crossover tutorial.
The weakness of moving-average filters is lag and whipsaw. In a choppy market the price crosses the average repeatedly, generating trades that lose a little each time. In a V-shaped recovery the filter turns positive only after much of the rebound has happened.
Family 2: absolute (time-series) momentum
A close relative of the moving-average filter asks whether the market's return over the past twelve months exceeds the risk-free rate. This is the absolute momentum test used in dual momentum and studied by Moskowitz, Ooi and Pedersen (2012). It has the same strengths and weaknesses as a moving-average filter and tends to give similar signals, because both summarise the same price history.
Family 3: market-state rules
Cooper, Gutierrez and Hameed (2004) defined market states by the cumulative return of the market over the prior 36 months. After "up" states, U.S. momentum profits were significantly positive; after "down" states, they were close to zero. This is a slower signal than a moving-average filter, designed to identify the post-bear-market environment in which momentum is most fragile, and it aligns closely with the crash mechanism found by Daniel and Moskowitz (2016).
A practical rule based on this finding cuts momentum exposure after a three-year market decline and restores it when the market recovers. Because such states are rare, the rule switches seldom, which keeps costs low but means its benefits rest on a small number of historical episodes.
Family 4: volatility-based scaling
Instead of switching on and off, volatility scaling adjusts exposure continuously, in proportion to the inverse of recent volatility. Barroso and Santa-Clara (2015) showed that scaling momentum to a constant volatility target substantially improved its risk-adjusted returns in their sample. This is technically risk management rather than timing, but in practice it has a similar effect: exposure falls in the turbulent markets where crashes occur. Our article on choosing your risk covers volatility targets in more detail.
Family 5: statistical regime models
Regime models assume that markets move between a small number of unobserved states, for example a calm, rising state and a volatile, falling one, each with its own return distribution. A hidden Markov model estimates the probability of being in each state from observed returns. The approach goes back to James Hamilton's 1989 regime-switching model of the business cycle. For timing, the strategy holds full exposure when the model assigns a high probability to the favourable state and reduces it otherwise. We introduce the method in hidden Markov models for market regimes.
Regime models are more flexible than fixed rules, but they have more parameters, and their state probabilities can change when the model is re-estimated with new data. A fair test must re-estimate the model using only data available at each point in time.
Measuring whether timing helps
A timing overlay should be judged against the untimed strategy on several dimensions at once:
| Measure | Why it matters |
|---|---|
| Annualised return and volatility | Timing often lowers both; the question is the balance. |
| Maximum drawdown and its length | The main reason to time momentum is to shorten and shallow the worst periods. |
| Skewness of monthly returns | A useful overlay should reduce momentum's negative skew. |
| Number of switches and costs | Frequent switching can absorb the benefit. |
| Results across sub-periods | An overlay that works only in one decade is likely fitted to it. |
| Sensitivity to parameters | Results should not collapse when the look-back changes from 10 to 9 or 12 months. |
The honest trade-off
No timing model avoids losses without also missing gains. Trend filters cut drawdowns in long bear markets but cost return in sideways markets and fast recoveries. Market-state rules target momentum's specific weakness but rely on few episodes. Volatility scaling is the most broadly supported in the literature but uses leverage in calm periods. A researcher who reports only the version that worked best in the backtest has learned little; reporting the full range of variants is what makes the result credible.
Sources
- Faber, M. (2007). A Quantitative Approach to Tactical Asset Allocation. Journal of Wealth Management, 9(4), 69–79.
- Cooper, M., Gutierrez, R. and Hameed, A. (2004). Market States and Momentum. Journal of Finance, 59(3), 1345–1365.
- Daniel, K. and Moskowitz, T. (2016). Momentum crashes. Journal of Financial Economics, 122(2), 221–247.
- Moskowitz, T., Ooi, Y. H. and Pedersen, L. (2012). Time Series Momentum. Journal of Financial Economics, 104(2), 228–250.
- Barroso, P. and Santa-Clara, P. (2015). Momentum Has Its Moments. Journal of Financial Economics, 116(1), 111–120.
- Hamilton, J. (1989). A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle. Econometrica, 57(2), 357–384.