Timing the Momentum Factor Using Its Own Volatility: A Historical Perspective
The momentum factor is one of the most well-known and persistent anomalies in finance. Traditionally, it exploits the tendency of stocks that have performed well in the past to continue outperforming in the short to medium term. But while the momentum premium is compelling, its returns can be highly volatile — with notorious drawdowns during market reversals.
What if we could improve the performance of momentum by timing our exposure based on its own risk?
In this post, we explore a simple yet powerful idea: applying volatility-based timing to the momentum factor itself. Inspired by the low volatility anomaly — which shows that lower risk often coincides with better risk-adjusted returns — we ask:
Can we enhance the performance of the momentum factor by going long only when its recent volatility is below a certain threshold?
To answer this, we use the daily momentum factor data from the Kenneth R. French Data Library, one of the most respected academic sources for asset pricing factors. The database includes daily momentum returns from 1926 to the present, giving us nearly a century of data to test this idea.
We’ll:
- Estimate historical volatility using a rolling lookback window
- Define simple thresholds to classify “low volatility” regimes
- Compare the cumulative and risk-adjusted returns of the volatility-filtered strategy vs the standard momentum factor
Let’s see if less risk really means more reward — even for a high-octane factor like momentum.
Measuring the Momentum Factor’s Volatility
To time the momentum factor, we first need a way to measure how volatile it is at any given time. For this, we use a simple approach: we calculate the rolling historical volatility of the daily momentum returns.
Specifically, we look at the past 252 trading days (roughly one calendar year) and calculate the standard deviation of those daily returns — a common way to estimate how “bumpy” a ride the factor has been. Then, we annualize that number to make it easier to compare over time.
Here’s what we get: a time series that tells us how much the momentum factor has fluctuated, day by day, over the past century.
We plot this rolling volatility over time, and unsurprisingly, it spikes during major market events like the Great Depression, the dot-com crash, and the 2008 financial crisis.
Momentum Returns Across Volatility Regimes
Before jumping into a timing strategy, we wanted to understand a basic question:
Does the momentum factor perform differently depending on how volatile it is?
To test this, we grouped the historical daily momentum returns into five buckets based on their trailing 1-year (252-day) volatility. These groups, or quintiles, range from the calmest periods (Q1) to the most turbulent ones (Q5).
Here’s what we found:
| Volatility Quintile | Avg. Daily Momentum Return |
|---|---|
| Q1 (Lowest Vol) | +0.0589% |
| Q2 | +0.0307% |
| Q3 | +0.0282% |
| Q4 | +0.0293% |
| Q5 (Highest Vol) | –0.0238% |
Tuning the Volatility Threshold for Optimal Momentum Exposure
After observing that the momentum factor tends to perform better during periods of lower volatility, we asked:
What’s the best volatility threshold to follow momentum and avoid excessive noise?
To explore this, we created a simple strategy:
- Go long the momentum factor when its trailing 1-year volatility is below a given target.
- Stay out of the market when volatility exceeds the target.
We varied the volatility target from 0 to 50 and tracked the strategy’s average daily return and Sharpe ratio at each level.
📈 Results: Performance Across Volatility Targets
Here are the key findings:
- Return peaks around a volatility threshold of 16–17%, reaching a daily average of ~0.032%.
- Sharpe ratio peaks earlier, around 7–8%, with a Sharpe above 1.2, indicating strong risk-adjusted returns.
- After a certain point (above 30%), returns and Sharpe both begin to decline, suggesting excess volatility degrades momentum effectiveness.
🔍 Visualization
Refining Momentum Timing with Dynamic Volatility Scaling
Previously, we explored a binary timing approach—either holding the momentum factor or staying out of the market entirely, based on whether recent volatility was below a threshold. While this method showed performance improvements, it comes with a significant downside: frequent in-and-out trading, which can generate high transaction costs and reduce net returns.
To address this, we implemented a dynamic position sizing strategy. Instead of a binary decision, this method adjusts exposure continuously based on the current level of volatility relative to a target. Specifically, we scale the daily momentum return using the ratio:
scaled_return = (vol_target / recent_volatility) * momentum_returnThis way, when volatility is low, the position size increases; when volatility rises, the strategy automatically reduces exposure. This smooth transition minimizes whipsawing and better aligns with the idea of risk parity.
📊 Performance Comparison
| Strategy | Mean Daily Return | Standard Deviation | Sharpe Ratio |
|---|---|---|---|
| Standard Momentum | 0.0254% | 0.7855% | 0.5142 |
| Volatility-Scaled Momentum | 0.0438% | 0.6638% | 1.0476 |
The results are striking: by dynamically adjusting exposure, the strategy not only increased average returns but also cut volatility—doubling the Sharpe ratio compared to the standard approach.
This supports the idea that even high-octane factors like momentum can benefit from risk-based position sizing. It's a powerful example of how simple rules, rooted in sound principles, can significantly improve performance.
Comparing Different Lookback Windows in Momentum Scaling
To test the robustness of our volatility-scaled momentum strategy, we evaluated how different lookback windows used to compute volatility affect both returns and risk-adjusted performance. Specifically, we compared 1-month (21 trading days), 6-month (126 days), and 12-month (252 days) lookbacks. The idea is simple: recent volatility might be more relevant for short-term positioning, but longer windows might smooth out noise.
| Lookback Window | Mean Daily Return | Daily Std. Dev. | Sharpe Ratio |
|---|---|---|---|
| 1M (21 days) | 0.0595% | 0.652% | 1.45 |
| 6M (126 days) | 0.0482% | 0.660% | 1.16 |
| 12M (252 days) | 0.0438% | 0.664% | 1.05 |
| Standard Momentum | 0.0254% | 0.785% | 0.514 |
Key insights:
- 🔹 All volatility-scaled strategies outperform the standard momentum strategy in terms of both average return and Sharpe ratio.
- 🔹 The 1-month lookback delivers the highest return and best risk-adjusted performance.
- ⚠️ However, shorter lookbacks imply more frequent rebalancing, leading to higher transaction costs and slippage, especially in less liquid environments.
- 📉 The 12-month lookback offers smoother scaling and lower turnover, at the cost of slightly lower returns.
This highlights the tradeoff between reactiveness vs. stability in momentum timing. The best lookback choice depends on your cost constraints, trading frequency, and portfolio turnover tolerance.
Digging Deeper: Controlling for Other Factors in Momentum Performance
After observing the significant improvements achieved through volatility scaling, a natural follow-up question arises: are these improvements purely a function of volatility management, or could other known market factors be influencing the momentum factor’s behavior?
To answer this, we performed a multivariate regression analysis that includes not only the historical volatility of momentum, but also the five well-established Fama-French factors—Market (Mkt-RF), Size (SMB), Value (HML), Profitability (RMW), and Investment (CMA). These factors are widely used to explain variations in asset returns and are often considered essential components in factor-based portfolio construction.
The goal here is to assess whether the momentum factor remains significant after accounting for these other influences, and more importantly, whether its volatility still provides incremental explanatory power. By regressing momentum returns against these variables, we aim to isolate the unique contribution of each component and identify any potential multicollinearity or hidden dependencies that may be driving performance.
The regression model used was:
Momentum ~ Market (Mkt-RF) + Size (SMB) + Value (HML) + Profitability (RMW) + Investment (CMA) + Momentum Volatility (MomVol)Here are the results:
| Variable | Estimate | Std. Error | t value | Pr(>|t|) |
|---|---|---|---|---|
| Intercept | 0.0835 | 0.0110 | 7.57 | < 0.0001 *** |
| Mkt-RF | -0.0775 | 0.0062 | -12.50 | < 0.0001 *** |
| SMB | 0.0100 | 0.0113 | 0.89 | 0.374 |
| HML | -0.5560 | 0.0120 | -46.28 | < 0.0001 *** |
| RMW | 0.1236 | 0.0156 | 7.94 | < 0.0001 *** |
| CMA | 0.4639 | 0.0192 | 24.16 | < 0.0001 *** |
| MomVol | -0.0795 | 0.0142 | -5.60 | < 0.0001 *** |
Model Summary:
- Residual Std. Error: 0.7242
- Multiple R-squared: 0.1436
- Adjusted R-squared: 0.1433
- F-statistic: 427.9 on 6 and 15,304 degrees of freedom
- p-value: < 2.2e-16
Conclusion: Even after accounting for the five Fama-French factors, momentum volatility (MomVol) remains a statistically significant variable, negatively associated with momentum returns. This reinforces the intuition behind dynamic scaling: when momentum is volatile, expected returns are lower, suggesting that position sizes should be adjusted accordingly. In other words, not only does volatility scaling help reduce drawdowns, but it also aligns with the statistical drivers behind momentum performance itself.
🧠 Takeaway
This exercise highlights how volatility itself can be used as a timing tool for factor exposure. Not only does this strategy improve average returns, but it also improves risk-adjusted performance compared to blindly holding the factor.
The next step would be to test the robustness of this finding across:
- Different lookback windows for volatility,
- Other factors (value, size, etc.),
- Out-of-sample validation or cross-validation windows.
Conclusion
Volatility-based timing offers a promising approach to enhancing momentum investing by systematically reducing exposure during turbulent periods. Our historical analysis shows that applying a simple volatility filter can lead to better average returns and superior risk-adjusted performance, helping to mitigate the large drawdowns that often accompany momentum strategies.
While these findings are encouraging, investors should treat them as a starting point rather than a final solution. Further research and testing, especially in live or out-of-sample settings, are essential before adopting such timing strategies in practice.
By combining classic factor insights with risk management techniques like volatility timing, we can strive toward more resilient investment approaches that better navigate the complexities of financial markets.
No comments:
Post a Comment