Using Historical Rolling Returns
If you invested $100,000 in U.S. stocks on January 1, 1980, you earned roughly 10% annualized through 2024. But if you invested on January 1, 2000, you earned 8% annualized through 2024 (a worse return despite 24 more years). If you invested on January 1, 2008 (right before the financial crisis), you still earned 8% annualized despite the crash.
Same asset class (U.S. stocks). Different returns. Why? Timing. The year you start and stop dramatically affects your outcome. Rolling returns—measuring returns over fixed-length periods starting on every possible date—reveal this pattern. They show you not the "average" return, but the distribution of returns you might have earned depending on when you started.
Rolling returns are the most honest tool for understanding long-term investment outcomes because they sidestep survivor bias and reveal the range of experiences real investors had.
Quick definition
Rolling returns measure the annualized return for every possible fixed-period window in historical data. For example, 10-year rolling returns starting in January 1920, February 1920, March 1920, and so on through every month of data. This creates a distribution showing the range of 10-year outcomes possible depending on starting date, capturing the impact of timing.
Key takeaways
- Rolling returns reveal that 10-year outcomes vary from −2% to +15% annualized for U.S. stocks, depending on start date
- Bad timing (investing at market peaks) produced painful results; good timing (investing at crashes) produced spectacular returns
- For retirement withdrawals, rolling returns show that a 4% withdrawal rate succeeded in ~95% of historical periods, but failed in ~5% (typically those periods starting at market peaks)
- Sequence of returns risk is visible in rolling returns: a period with the same average return but different ordering produces different outcomes
- Rolling returns are the foundation of sustainable withdrawal rate research (Trinity Study) and Monte Carlo validation
How Rolling Returns Work
Imagine you have 125 years of stock market data. You calculate:
- 10-year return starting January 1900, ending December 1909: +4.2% annualized
- 10-year return starting January 1901, ending December 1910: +5.1% annualized
- 10-year return starting January 1902, ending December 1911: +3.8% annualized
- ... and so on, every month forward, until
- 10-year return starting January 2014, ending December 2023: +12.3% annualized
Now you have 1,500 data points (125 years × 12 months, minus 120 months for the final window), each showing what a 10-year investor experienced if they started in that month.
You can then plot the distribution:
- Worst 10-year period: −2% annualized (starting in 1929, the Great Depression)
- 10th percentile: 2% annualized
- Median: 9.8% annualized
- 90th percentile: 14.1% annualized
- Best 10-year period: 19.8% annualized (starting in 1949, post-WWII recovery)
This distribution is far more informative than "the average return is 10%." It shows that one-tenth of investors who picked a random 10-year period earned only 2%, while the luckiest earned nearly 20%. The spread is the real story.
The Trinity Study and Sustainable Withdrawal Rates
The most important application of rolling returns is the Trinity Study (Cooley, Hubbard, and Walz, published 1998), which examined whether retirees could safely withdraw 4% of their portfolio in year one, then adjust by inflation, and never run out of money.
The researchers took U.S. historical data from 1926 to 1995 (70 years) and tested every possible 30-year period (rolling window). They paired various asset allocations (100% stocks, 75/25, 50/50, 25/75, 100% bonds) with the 4% withdrawal strategy.
Key finding: A 4% initial withdrawal rate succeeded in 95% of 30-year rolling periods.
This meant:
- In 67 of 70 possible 30-year rolling windows, a retiree starting with 4% spending (adjusted for inflation) never ran out of money
- In 3 periods, the retiree ran out of money
- Those 3 failures were periods starting at stock market peaks (1966, 1973, 1987) followed by prolonged crashes or stagflation
This is rolling return analysis in action. The "4% rule" wasn't pulled from theory—it was extracted from the distribution of outcomes.
Later research extended the study through 2024, confirming the 95% success rate (sometimes 94%, sometimes 96%, depending on exact data and methods). But the failures are always visible: rolling returns make clear when the strategy would have failed.
Rolling Return Outcomes
The diagram illustrates rolling return timing: starting in 1926 (pre-crash) vs. 1932 (post-crash) produces dramatically different outcomes with the same strategy.
What Rolling Returns Reveal About Sequence Risk
Sequence of returns risk—the harm of early losses when spending from a portfolio—is visible in rolling returns.
Two 30-year periods might have the same average return (say, 7%) but different sequences:
Good sequence (accumulation): +10%, +10%, +10%, +5%, +2%, +1%, +5%, +8%, ... (gains clustered early)
Bad sequence (early crash): −20%, −15%, +5%, +5%, +10%, +15%, +12%, +8%, ... (losses clustered early)
Both average 7%. But a retiree withdrawing 4% in the bad sequence faces a catastrophe:
Year 1: Portfolio −20%, then −4% withdrawal → portfolio is −24% (depletes much faster)
Year 2: Portfolio −15%, then −4% withdrawal → portfolio is −19%
Years 3+: Portfolio recovers, but damage is done.
In the good sequence, gains come early, so the portfolio grows while withdrawals happen from the gains, not the principal. The retiree is fine.
Rolling returns capture this: the rolling return for the bad sequence is lower than the good sequence, even though the average return is identical. This is why looking at annualized returns is insufficient—you need to see the specific year-by-year order.
How to Interpret Rolling Return Charts
A typical rolling return chart for U.S. stocks (10-year periods) shows:
- X-axis: Starting date (1926 to 2015)
- Y-axis: 10-year annualized return (−5% to +20%)
- Each point: The 10-year return if you invested on that date
Reading it:
- If the chart shows a point at (1929, −2%), this means investing in 1929 and holding 10 years yielded −2% annualized (the Great Depression period).
- If the chart shows a point at (1949, +15%), investing in 1949 and holding 10 years yielded +15% (post-war recovery).
- The scatter tells you the range of outcomes.
- Clustering (points all in the 8–12% range) suggests consistency.
- Wide spread (points from −5% to +20%) suggests high variability in timing.
U.S. stocks show a wide scatter for 10-year periods, indicating significant timing effects. 30-year periods show less scatter (fewer really bad 30-year periods) because time smooths volatility. This is why long-term investing "works"—the worst possible 30-year period still had positive returns.
Rolling Returns for Different Periods
1-year rolling returns: Highly variable, −50% to +50% range. Almost useless for planning because any single year is noisy.
5-year rolling returns: Still variable, −10% to +25% range. Some bad periods visible (2008–2012 period includes the crash and slow recovery), some great periods visible (1995–1999 tech boom).
10-year rolling returns: More stable, −2% to +20% range. Most periods cluster in the 8–12% range. Bad periods (1929–1938, 1966–1975) are visible outliers. Useful for understanding decade-long outcomes.
20-year rolling returns: Very stable, typically 6–13% range. Few truly bad 20-year periods in U.S. history. 1928–1947 (includes Depression and WWII) is the worst.
30-year rolling returns: Nearly all periods succeeded (positive returns), except a few that barely crossed zero or went negative (1929–1958 is notably weak). This is why "30 years is enough to recover from any crash."
Rolling Returns and Withdrawal Sustainability
The Trinity Study tied rolling returns directly to retirement sustainability. The finding: a strategy that fails in 5% of 30-year rolling periods is a strategy where about 1 in 20 retirees (starting on different dates) would have faced financial stress.
This connects rolling returns to Monte Carlo: Monte Carlo generates synthetic return sequences (based on real volatility patterns). Rolling returns show the actual sequences that occurred. A Monte Carlo result of "92% success rate" should roughly match rolling return data (e.g., "92% of 30-year periods succeeded"), validating the simulation.
Mismatches reveal problems:
- If Monte Carlo shows 92% success but rolling returns show only 85% success, the Monte Carlo assumptions are too optimistic.
- If rolling returns show 95% success but Monte Carlo shows 80%, the simulation assumptions are too pessimistic or the volatility model is wrong.
International Rolling Returns
Rolling returns for non-U.S. markets reveal how much worse timing risk was elsewhere.
Japanese stocks:
- 10-year returns starting 1980: +18% annualized (the bubble)
- 10-year returns starting 1989: −1% annualized (the crash and long decline)
- 20-year returns starting 1989: +2% annualized (the lost decade plus slow recovery)
A Japan investor in 1980 felt genius; a Japan investor in 1989 felt foolish for 20+ years. Same country, same market, different timing. Rolling returns make this visible.
Emerging markets: Rolling returns for emerging markets (Brazil, Russia, India, China) show extreme variability: periods of 15%+ returns followed by periods of 0–5% returns. The distribution is much wider than U.S., indicating that starting date matters more. An investor starting in 1999 (before Brazil's 2000s boom) had great returns; an investor starting in 2012 (after the boom) had mediocre returns.
This is why emerging market return forecasts are unreliable: the range of outcomes is huge, and timing determines whether you're in the boom or the bust.
How to Use Rolling Returns for Planning
Step 1: Find rolling return data for your asset mix.
Various sources provide rolling return data:
- Morningstar Principia or the Morningstar website
- Vanguard's historical return data
- DFA (Dimensional Fund Advisors) research
- Academic databases (Shiller, CRSP, Fama-French data library)
For example, search "S&P 500 rolling returns 30-year" or "60/40 portfolio rolling returns."
Step 2: Note the distribution.
What's the worst 30-year return? The best? The median? The 10th percentile? This distribution is your planning range.
Step 3: Test your plan against the pessimistic scenarios.
If the worst 30-year rolling return for your portfolio mix is 5% annualized, can you live on that? If you're planning on 7% returns, you're assuming above-median outcomes. That's okay if you have flexibility, but dangerous if you don't.
Step 4: Stress-test with different starting points.
Run your plan not just with "average" returns, but with returns from the worst historical rolling period. If your plan survives that, you're likely safe.
Common Mistakes
Mistake 1: Assuming rolling returns represent the future
Rolling returns show what happened 1926–2024. There's no guarantee the future matches that distribution. Regime changes (technology, demographics, policy) could create new patterns. Use rolling returns as one input, not the sole predictor.
Mistake 2: Cherry-picking the best rolling periods
If you show a client the 1949–1979 rolling 30-year period (great returns) and ignore the 1929–1958 period (terrible returns), you're lying with data. Always show the full distribution.
Mistake 3: Confusing rolling returns with expected returns
A rolling return of 9% in a historical period doesn't mean you should expect 9% going forward. Expected returns depend on valuations today; rolling returns tell you what was earned in the past. They're related but not identical.
Mistake 4: Not adjusting for inflation
Rolling returns are often quoted nominally (not adjusted for inflation). A 4% nominal return during high inflation is a 0–1% real return. Always ask: are these nominal or real?
Mistake 5: Ignoring that rolling returns exclude taxes and fees
Historical rolling returns are usually calculated before taxes and fees. A 9% rolling return becomes 7% after a 1% expense ratio and −1% tax drag. Plan on lower than historical rolling returns.
FAQ
Q: If rolling returns show the worst 30-year period returned 2%, should I plan on 2%?
A: No. Plan on the median or a modest percentile (25th–50th). The worst historical period is one data point; the median is more defensible. But know the worst-case exists and keep contingencies (flexible spending, part-time work) if needed.
Q: Why do rolling returns for U.S. stocks show all positive 20+ year periods, but negative 10-year periods?
A: Volatility clusters and means-revert, but over long enough periods (20+ years), diversification and economic growth dominate. The Depression (1929–1939) was brutal over 10 years, but by 20 years (1929–1949), the post-WWII recovery had lifted returns into positive territory.
Q: How do rolling returns compare to Monte Carlo simulations?
A: Rolling returns are real data. Monte Carlo is simulated data. Rolling returns have survivorship bias (they only include markets that survived). Monte Carlo can include hypothetical meltdown scenarios. A good approach: validate Monte Carlo outputs against rolling returns (they should roughly match), then extend Monte Carlo into scenarios beyond historical experience.
Q: Should I use rolling returns or historical average returns for planning?
A: Both. Historical average (e.g., 9% for U.S. stocks) is a central tendency. Rolling returns reveal the range around that tendency. Plan on a return lower than the historical average (due to valuation changes), but within the historical distribution (because regime change is possible but not certain).
Q: If I'm young, can I ignore bad rolling periods and just plan on the average?
A: Not entirely. A young person with 40+ years has time to recover, so a bad 10-year period is less painful. But if that bad period occurs in their 20s, and it's bad enough to trigger panic selling or job loss, the behavioral impact is real. The mathematical recovery happens, but the opportunity cost (years of being out of the market) is material.
Q: How far back should rolling return data go?
A: For U.S. stocks, 100+ years is ideal, capturing multiple bull markets, recessions, and regime shifts. For international or emerging markets, 50+ years. If data only goes back 20 years, you're seeing a limited distribution and likely missing tail risks.
Related Concepts
- Sequence of returns risk – Why the order of returns affects outcomes, visible in rolling return variation
- Sustainable withdrawal rates – Rolling returns establish the empirical foundation for 4% rule
- Dollar-cost averaging – How regular investing smooths rolling return variability
- Market timing – Why entering at different points in the cycle produces different outcomes
- Recency bias – Assuming recent rolling returns will continue (they don't)
Summary
Rolling returns are the empirical foundation of long-term investing wisdom. They show not the "average" return, but the distribution of returns you might have earned depending on when you started. A 10% average return is accurate for U.S. stocks, but hiding within that average are periods where 30-year investors earned 2%, and periods where they earned 14%. Timing matters profoundly.
The Trinity Study, which validated the 4% withdrawal rule, was built on rolling returns: testing every historical 30-year window and counting successes. Rolling returns revealed that a retiree spending 4% in year one (adjusted for inflation) succeeded in 95% of periods—not because theory says so, but because the data shows it.
For your own planning, use rolling returns to understand the range of plausible outcomes. Find data for your asset allocation (60/40, 80/20, whatever you're using), look at the worst 30-year rolling period in the history available, and ask: could my plan survive that outcome with some flexibility (spending cuts, continued work, home sale)? If yes, you're likely safe. If no, adjust your plan downward until it passes the worst-case test.
Rolling returns are honest: they don't hide timing risk, they expose it. Use them to build plans robust to bad timing, because you won't know good timing in advance.