Factor performance changes. A signal that has worked consistently for years can weaken as market conditions shift, investor behavior changes, or more capital begins pursuing the same opportunity.
For quantitative investors managing multiple signals, that creates an important allocation question: Should every signal receive the same weight at every point in time?
Equal weighting offers a robust starting point for signal aggregation. Our latest research explores whether investors can improve upon that baseline by using information about the environment surrounding each signal to dynamically adjust its weight.
We focused on two sources of information: market regimes and factor crowding.
Reading the market environment
Economic conditions can influence which investment signals are most effective.
ExtractAlpha’s research uses a broad set of macroeconomic and market variables to identify periods with similar characteristics. These include the VIX, financial conditions, economic policy uncertainty, residual stock-return variance, equity returns, bond yield spreads, Treasury yields, commodity prices, and stock-bond correlations.
Rather than treating each market environment as entirely new, regime classification allows us to ask a more useful question:
When conditions looked like this before, which signals performed well?
The framework uses those historical relationships to make modest adjustments around an equal-weighted portfolio, assigning greater weight to signals that historically performed better in comparable regimes.
The results are meaningful. Across the parameters tested, applying the regime model improved Sharpe ratio, skewness, and maximum drawdown relative to the equal-weighted baseline.
Measuring when a signal becomes crowded
Market environment tells only part of the story.
A factor can also become vulnerable when too many investors pursue the same opportunity. Crowding can contribute to faster alpha decay, greater volatility, and more severe downside when positions unwind.
The challenge is determining when crowding is occurring.
We constructed a proprietary composite crowding indicator using nine measures derived from granular stock-level data. The framework incorporates signals including information coefficient decay, abnormal trading activity, intra-decile correlations, Sharpe behavior, and options market activity.
Across the nine signals studied, higher crowding scores were associated with greater drawdown probability, lower subsequent IC, higher volatility, and less favorable return skewness.
That information can then become actionable. Signals exhibiting lower crowding risk receive greater weight, while allocations to more crowded signals are reduced.
What happens when you combine the two?
This is where the research becomes particularly interesting.
We combined regime classification and crowding indicators into a single factor-timing framework and compared its performance with an equal-weighted portfolio.
The resulting model increased the Sharpe ratio from 2.51 to 2.85 and aggregate return from 14.9% to 16.3%. Maximum drawdown improved from -12.9% to -9.8%, accompanied by lower volatility and improvements in skewness and kurtosis.
The drawdown analysis on page 8 of the research is especially revealing. During periods of acute market stress, including the Global Financial Crisis and the COVID-19 sell-off, the dynamically weighted portfolio experienced shallower drawdowns than the equal-weighted baseline.
For investors combining multiple alpha signals, the implication deserves attention. Signal selection is only one part of the portfolio construction problem. The weight assigned to each signal, and when that weight changes, can materially affect portfolio outcomes.
Read the Research
ExtractAlpha’s research note, “Enhancing Signal Aggregation through Regime Classification and Crowding Indicators,” details the methodology behind both models, the construction of our proprietary crowding indicator, testing across classic quantitative factors and ExtractAlpha signals, and the performance of the combined factor-timing framework.
Interested in seeing the full research?