Rolling Returns Explained

Rolling returns are the statistical foundation of quantitative investing. They answer the question: if an investor had bought at any point in history and held for a specific period, what would the results have been? The answer is not one number but a distribution—showing both the frequency of positive outcomes and the magnitude of losses in worst-case scenarios.

Rolling returns transform "what happened once" into "what happens consistently." They replace a single historical narrative with a distribution of all possible outcomes.

This method removes luck from the analysis. It replaces the story of "I bought in 1980 and became rich" with the empirical fact: "In approximately 92% of ten-year periods since 1926, an S&P 500 investor made money."

Key takeaways

• Rolling returns calculate returns for every possible starting date over a specified holding period
• A 5-year rolling return series has roughly as many data points as (total years of data - 5)
• The rolling return distribution shows range, median, and worst-case outcomes
• Longer rolling periods show less volatility and more consistent positive outcomes
• Rolling returns reveal that average returns mask significant period-to-period variation
• The analysis requires using total return indices that include reinvested dividends

Visual representation of rolling returns

The mechanics of rolling returns

Imagine a spreadsheet with 100 years of monthly market data. To calculate rolling ten-year returns:

1. Take all returns from month 1 through month 120 (ten years). Compound them. Record the total return.
2. Move forward one month. Take all returns from month 2 through month 121. Compound them. Record the total return.
3. Repeat until you run out of data (ending with month 1081 through month 1200).
4. Now you have approximately 960 ten-year rolling returns (100 years of data minus 10 years).

This creates a distribution. Some periods had 15% annualized returns. Others had 2%. A few had negative returns. The distribution shows all of them.

Sample Rolling Return Calculation (simplified):

Data: S&P 500 monthly returns from Jan 1980 to Dec 1989
Start date: Jan 1980
End date: Dec 1989
Holding period: 10 years

Rolling return = (Price Dec 1989 / Price Jan 1980) - 1
               = (468 / 135) - 1
               = 246% cumulative
               = 13.7% annualized

Next rolling period:
Start date: Feb 1980
End date: Jan 1990
Holding period: 10 years

And so on...

The key requirement is that you use total return indices, not price indices. A price index shows capital appreciation only. A total return index includes reinvested dividends. For the U.S. stock market, this makes a dramatic difference. Dividends have contributed roughly 40-50% of total returns over the long run. Ignoring them produces misleading results.

What the distribution reveals

Once you have 960 ten-year rolling returns, you can ask questions the market data alone cannot answer:

• What is the median result? (The typical case, not an average which can be skewed)
• What is the worst case? (The most negative rolling ten-year return ever observed)
• What is the best case? (The most positive rolling ten-year return ever observed)
• What fraction are positive? (The probability of success)
• What is the standard deviation? (The volatility of the distribution)

For S&P 500 ten-year rolling returns since 1926:

• Median return: ~9.5% annualized
• Worst case: ~-0.8% annualized (the period starting in 1929, encompassing the Great Depression and recovery through 1939)
• Best case: ~19% annualized (the 1950-1960 period)
• Probability of positive returns: ~92%
• Standard deviation: ~4.2% annualized

This distribution is far more informative than a single number (the long-term average of 10% annualized). It shows that a ten-year investor should expect something in the range of 5-14% annualized, with rare exceptions in either direction.

Rolling returns vs. buy-and-hold

A buy-and-hold analysis examines one outcome: someone who bought in 1926 and held until today. This is a single data point. It tells you what happened once.

Rolling returns examine all possible buy-and-hold scenarios. It tells you what typically happens.

These can diverge sharply. An investor who bought near the market peak in 1929 had a terrible experience through the 1930s. But a rolling analysis shows this was an outlier. Most ten-year periods, even including some crashes, produced positive returns.

Similarly, an investor who happened to retire in 1982 had exceptional returns through 2002. This was fortunate timing, not typical. The rolling distribution shows this was an optimistic outcome, not the median.

The overlay effect: overlapping periods are correlated

Rolling returns for adjacent periods overlap heavily (a ten-year rolling return starting in January differs from one starting in February only by the month of data added and the month removed). This creates correlation in the rolling return series.

Statistically, this means you cannot treat a rolling return series as a true random sample. You have ~960 rolling ten-year return observations, but they are not 960 independent observations; they are much closer to 120 true independent annual samples.

This is important for statistical rigor. A rolling return series is excellent for descriptive analysis (what happened in the distribution) but requires care if you're testing hypotheses or creating confidence intervals.

Using rolling returns to evaluate strategies

Rolling returns are powerful tools for comparing investment approaches:

Approach 1: Buy and hold the S&P 500 for ten-year periods.

Approach 2: Buy and hold a 60/40 stock-bond portfolio for ten-year periods.

Approach 3: Use a tactical allocation strategy that shifts between stocks and bonds.

For each approach, calculate the rolling return distribution. Compare the medians, the worst cases, and the probability of positive returns. The approach with the highest median and lowest worst-case drawdown is superior, even if outcomes vary by period.

Real-world interpretation

Imagine an investor says, "I'm planning to retire in seven years and invest in the stock market." What does rolling return analysis tell her?

Looking at S&P 500 seven-year rolling returns:

• Median return: ~8.5% annualized (cumulative 71%)
• Worst case: approximately -4% annualized (cumulative -24%)
• Best case: ~23% annualized (cumulative 260%)
• Probability of positive returns: ~78%

This tells her several things:

1. She has a 78% historical chance of being profitable over her seven-year window—quite good but not certain.
2. If she is unlucky with her starting date, she might face a -24% outcome, which would still leave her with money, but would feel uncomfortable.
3. Her expected return is around 8-9% annualized, which compounds to a 70% gain over seven years on an initial investment.
4. She should ensure her portfolio and life plan can tolerate the -24% drawdown scenario.

Common pitfalls in rolling return analysis

Pitfall 1: Forgetting to reinvest dividends. A total return index is essential. Using a price index overstates the volatility and understates the returns of long-term holding.

Pitfall 2: Using insufficiently long data history. If you have only twenty years of data and want to examine ten-year rolling returns, you have only ten independent observations. More data is required for reliable statistical conclusions.

Pitfall 3: Assuming worst-case equals future worst-case. A rolling analysis shows what happened in the past, not what will happen in the future. The worst seven-year period from 1926-2024 may not bound the worst seven-year period from 2025-2100.

Pitfall 4: Ignoring survivorship bias. Rolling return analysis using S&P 500 data only includes companies that survived to be in the index. Companies that went bankrupt are not included. This biases results optimistically.

FAQ

Q: How much historical data do I need for reliable rolling returns?
A: At least 3-4 times the holding period you're analyzing. For ten-year rolling returns, fifty years of data is a minimum. Ideally 100+ years.

Q: Why not just look at the long-term average?
A: The average hides variation. One investor might get 15% annualized while another gets 2%. The rolling distribution shows both possibilities; an average shows neither.

Q: Do rolling returns account for market regime changes?
A: Not explicitly. A rolling return analysis assumes the future behaves like the past. If market structure changes (e.g., the rise of algorithms), historical rolling returns may be poor predictors.

Q: Can I use rolling returns to compare stocks?
A: Yes, though with caveats. Rolling returns for a single stock are much more volatile than for an index. A single-stock rolling return analysis requires very long data history (50+ years) to be reliable.

Q: How do I account for the correlation between rolling periods?
A: For descriptive purposes, ignore it. For hypothesis testing, reduce your effective sample size by dividing by the overlap period.

Q: What if I want to invest for three years but only have data through ten-year periods?
A: You can aggregate data. Take every three-year slice of your ten-year rolling returns. This is more granular and provides more data points for analysis.

Related concepts

• Probability of positive returns: The percentage of rolling periods achieving profitability; derived from rolling return distributions
• Volatility drag: The impact of price fluctuations on cumulative returns; visible in the standard deviation of rolling return distributions
• Buy-and-hold: A single outcome; contrasts with rolling returns which show all possible outcomes
• Worst-case scenarios: The negative tail of the rolling return distribution; critical for risk planning
• Mean reversion: The tendency of rolling returns to cluster around a central value; visible in rolling return distributions
• Survivorship bias: The tendency of rolling return analyses to overstate results by excluding failures; a critical limitation

Summary

Rolling returns are the empirical foundation of quantitative investing. They answer the question: across all possible periods in history, what outcomes did long-term investors experience? This generates a distribution—showing typical returns, worst-case outcomes, and the frequency of success. Unlike a single historical narrative or a theoretical average, rolling returns reveal what investors should realistically expect. This is why they are essential to understanding the mathematics of long horizons.

Next: The Magic of the 10-Year Horizon

Rolling returns reveal patterns. One pattern stands out consistently in historical data: the ten-year holding period functions as a turning point. Shorter than ten years and volatility matters substantially. Longer than ten years and volatility shrinks and probabilities improve dramatically. Why is ten years the magic number?