Historical Context

Rolling Returns: What They Measure and How to Read the Chart

Rolling returns provide a powerful way to analyze investment performance over different time periods without being anchored to a specific start or end date. Instead of looking at returns from a fixed point in time (like "returns since 2000"), rolling returns calculate performance over a moving window (e.g., 5 years) at each point in the historical timeline. This approach eliminates the bias that can come from cherry-picking favorable or unfavorable start dates, giving a more complete picture of how an investment strategy has performed across different market conditions.

Here, rolling returns compound the selected series' monthly returns over a moving window. A positive reading means that compounded series gained over the window; a negative reading means it lost. For a value-minus-growth factor, this is the result of compounding monthly return differences. It is not the same calculation as separately compounding the value and growth portfolios and subtracting their final returns. Keeping that distinction clear helps readers compare the chart with other performance reports without confusing different calculations.

The chart on this page displays rolling annualized returns calculated over windows of 1, 3, 5, or 10 years, depending on your selection. Longer windows (5-10 years) smooth out short-term volatility and reveal longer-term trends, while shorter windows (1-3 years) capture more recent dynamics. When rolling returns remain consistently positive over long periods, it suggests a robust value premium. When they fluctuate between positive and negative, it indicates that the value-growth relationship is more cyclical and regime-dependent.

Drawdowns: What They Tell You About Value vs Growth

A drawdown measures the decline in a compounded return series from its previous peak. It depends on the sequence of returns, not just on the latest month. For a value-minus-growth series, repeated negative months can produce a prolonged drawdown even if its long-run average is positive. A monthly return distribution describes the individual observations used in that calculation. It cannot show when a peak occurred, how deep the cumulative decline became, or how long recovery took.

The minimum on this page is the lowest monthly return in the selected series, rather than its maximum drawdown. A percentile ranks the latest monthly return against the historical observations. A reading near the 10th percentile places that month toward the lower end of the sample. It does not identify the current price of value stocks relative to growth stocks. To study a cumulative loss from a peak, use the drawdown charts on the factor analytics page.

Skewness describes asymmetry in the monthly return distribution; positive skew indicates a longer or more pronounced right tail, rather than a guarantee of frequent gains. Kurtosis describes the influence of extreme observations relative to the center. These statistics can help explain why an average and standard deviation leave out parts of the return history. They do not determine what the next observation will be or establish that a particular investment is attractive.

Historical Regimes: What High and Low Spreads Have Meant Historically

Different historical intervals can produce very different factor returns. The examples below use High Minus Low (HML) from the Fama-French three-factor dataset, with source history available through July 2026. Average monthly returns describe the individual observations; calendar-year returns compound them. These fixed historical examples refer to HML, regardless of the metric selected in the chart above.

The 1970s and early 1980s: From January 1970 through December 1984, HML averaged 0.69% per month, with positive readings in 113 of 180 months. That combination shows a positive average alongside many months when value lagged growth. A long-window chart can smooth those interruptions, so examining the monthly observations adds useful context. This record describes the return difference earned by the research portfolios over those dates; it does not establish which economic conditions caused it.

The late 1990s and early 2000s: HML averaged -0.18% per month from 1995 through 1999, then 1.85% per month from 2000 through 2002. The contrast illustrates how strongly the chosen interval can shape a performance summary. In the earlier period, 31 of 60 monthly readings were positive despite the negative average, showing why the frequency and size of gains both matter. The later gains are part of the realized record, not evidence that the reversal was predictable.

The 2010s: From January 2010 through December 2019, HML averaged -0.20% per month, and only 48 of 120 months were positive. That decade gives a concrete example of weak value-factor returns persisting beyond a short market setback. Positive individual months still occurred within the broader decline. The record helps explain the difficulty of holding a factor exposure through a prolonged disappointing period, without establishing the underlying stocks' valuation at the end of it.

The 2020 to 2022 reversal: Compounding the monthly HML observations gives a return of -30.66% in 2020, followed by 22.20% in 2021 and 31.74% in 2022. These are calendar-year factor returns, not returns on an investor's account. The sharp change shows why a single year may describe a very different experience from a longer rolling window. It also illustrates the difference between documenting a reversal afterward and having a rule that could identify it beforehand.

Comparing historical intervals helps reveal changes in realized factor performance. The monthly spread can change sign quickly, while a rolling measure changes more slowly because it combines many months. Read both in their own units and keep their time windows clear. Economic narratives can suggest questions to investigate, but the return data alone do not establish a cause, a current valuation, or a forecast of the next change in performance.

Using Distribution Statistics for Investment Research

The distribution statistics presented on this page provide a comprehensive view of how value spreads have behaved historically. The mean and median offer measures of central tendency, though they may differ significantly if the distribution is skewed. The standard deviation quantifies the typical variability of spreads, helping investors understand the normal range of fluctuations. When current values fall outside of ±2 standard deviations from the mean, they represent statistically unusual conditions that may warrant attention.

The percentile locates the latest monthly return within the full recorded sample for that series. A reading near 95 places it toward the high end, while a reading near 5 places it toward the low end. This remains a description of one realized observation, even when the rank is extreme. It does not measure stock valuations or show that a reversal is due. Comparing successive months reveals how the ranks changed without turning them into independent trading signals.

The z-score expresses the latest return's distance from the historical mean in standard deviations. It makes the scale of that deviation easier to compare across series with different volatility. A large absolute score describes an unusual return relative to that sample; it does not by itself establish statistical significance or an investment opportunity. Factor returns can have asymmetric distributions and extreme observations, so normal-distribution probabilities should not be assumed.