A Monte Carlo retirement plan with a 1,000,000 portfolio, a 4% withdrawal rate, and a 30-year horizon. Run it at 7% expected real return: 91% success rate, median ending portfolio of 1,400,000. Run the same plan at 5% expected real return: 73% success rate, median ending portfolio of 510,000. Same retiree, same portfolio, same withdrawal strategy. Different number for one input.
The expected return is the single most consequential assumption in any Monte Carlo retirement model. It dwarfs the impact of asset allocation choices, withdrawal rate calibration, and even most spending strategy decisions. Getting it wrong does not produce a plan that is slightly off - it produces a plan that may not be a plan at all.
The default in most retirement calculators is to use long-run US historical averages, which produces optimistic numbers because the US 20th century was the most favorable equity market in the developed world. Forward-looking research suggests significantly lower returns are appropriate for the next 30 years. The right answer is somewhere between, and the way you handle the uncertainty matters more than which specific number you pick.
Why Expected Return Is the Most Important Input
Run a sensitivity analysis on a typical 30-year retirement plan, varying one input at a time:
| Input change | Approximate effect on success rate |
|---|---|
| Expected return: -1 percentage point | -15 to -20 points |
| Withdrawal rate: +0.5 percentage point | -8 to -12 points |
| Asset allocation: -10% equity | -3 to -8 points |
| Inflation: +1 percentage point | -5 to -10 points |
| Volatility: +5 percentage points | -3 to -6 points |
The expected return moves the success rate more than any other lever. A retiree who assumes 7% real returns and gets 5% does not have an 87% success plan that ends up at 80%. They have a plan that thought it was safe and is not.
This sensitivity is why most retirement calculators are quietly aggressive. They default to historical US averages, produce reassuring numbers, and let the user make decisions that depend on those numbers being right.
The Three Sources of Expected Return Estimates
Three approaches dominate how retirement planners estimate expected returns. Each has strengths; none is sufficient on its own.
Historical averages. The simplest approach: assume the next 30 years look like the average of the last 70-100. For US equities, this gives roughly 7% real. For US bonds, 2-3% real. The advantage is that the data is concrete and the math is transparent. The disadvantage is that the US 20th century is not necessarily representative of the future. Survivorship bias is a real factor: the US won. Many other countries had wars, revolutions, currency collapses, and prolonged equity bear markets that the US data set does not contain.
Forward-looking models. Approaches that compute expected returns from current conditions rather than historical averages. The most common:
- CAPE-based: expected return = 1/CAPE + dividend yield + earnings growth. With CAPE in the mid-30s, this gives roughly 4-5% real for US equities.
- Dividend discount: expected return = dividend yield + earnings growth + valuation change. Similar output.
- Building blocks: expected return = real risk-free rate + equity risk premium. Output depends on the assumed equity premium, typically 4-5% real for US equities.
These models converge on lower numbers than historical averages because current valuations are above historical norms.
Asset manager forecasts. Major asset managers publish 10-year and 30-year capital market expectations. The cluster as of 2025-2026:
| Asset class | Historical (US) | BlackRock | Vanguard | GMO | JPMorgan |
|---|---|---|---|---|---|
| US equities (real) | 7.0% | 5.0% | 4.5% | 2.5% | 5.0% |
| Intl developed (real) | 5.5% | 6.5% | 5.5% | 5.0% | 6.0% |
| US bonds (real) | 2.5% | 1.5% | 1.5% | 1.0% | 2.0% |
GMO is the outlier on the bearish side; the others cluster in the 4.5-5.5% range for US equities. International equities are estimated above US equities in most forecasts because non-US valuations are lower.
The Survivorship Bias Problem
The most common mistake in expected return modeling is treating US history as if it represents "the stock market" rather than one specific country's experience.
Dimson, Marsh, and Staunton's "Triumph of the Optimists" study computed real returns across 21 countries from 1900 to 2000:
| Country | Real annual equity return |
|---|---|
| South Africa | 7.0% |
| Australia | 7.5% |
| Sweden | 7.6% |
| United States | 6.7% |
| Switzerland | 5.0% |
| Netherlands | 5.8% |
| Canada | 6.4% |
| United Kingdom | 5.8% |
| France | 3.8% |
| Germany | 3.6% |
| Japan | 4.5% |
| Italy | 2.7% |
| Belgium | 2.5% |
The US is in the top quartile, not the average. The 21-country average is closer to 5%. A plan that assumes US-style returns is implicitly assuming the next 30 years go to the country that already won.
For a globally diversified portfolio, the relevant historical reference is the global average, not the US average. That alone reduces the expected return assumption from 7% to roughly 5-5.5% real before any forward-looking adjustment.
How to Choose Your Number
The temptation is to pick one number and run with it. The honest approach is to recognize that the number you pick has a wide error bar and to model the range explicitly.
A reasonable framework:
Bear case: historical 21-country average minus current valuation adjustment. For US equities, roughly 4% real. Run this to test plan robustness.
Central case: asset manager forecast cluster. For US equities, 5-5.5% real. This is the planning baseline. Run the success rate at this assumption.
Bull case: US historical average. For US equities, 7% real. Run this only to see what the plan would look like if everything goes well.
A plan that succeeds in the bear case is robust. A plan that succeeds only in the central case is reasonable. A plan that requires the bull case is fragile.
Most retirees pick a single central number and ignore the range. The Monte Carlo simulation already produces a probability distribution of outcomes given an assumption, but it does not capture the uncertainty about the assumption itself. Running multiple return scenarios captures that second layer of uncertainty.
The Equity Premium and Bond Returns
The expected return assumption is not just one number for "the portfolio." It is a number for each asset class, and the relationship between them - the equity premium - matters as much as the absolute numbers.
Historical US equity premium over bonds: about 4.5%. Forward-looking estimates: 3-4%. International averages: 3-3.5%.
A lower equity premium changes the case for equity allocation in retirement. The 60/40 portfolio's expected return depends heavily on the gap between equity and bond returns. If bonds yield 2% real and equities yield 5% real, the equity contribution to the 60/40 is 1.8% above the bond contribution per year. If equities yield 4% real (lower premium), that contribution drops to 1.2%, which is a 33% reduction in the equity premium.
For a 30-year retirement, this changes the optimal allocation. Standard advice for 60/40 is calibrated to a 4-5% equity premium. At a 3% premium, a higher equity allocation may be needed to hit the same expected outcome - but at the cost of higher volatility. There is no free lunch.
Use distinct return assumptions for stocks and bonds. Use distinct assumptions for US versus international equity. The simulation should reflect actual asset class differences, not blend them into a single portfolio number.
What This Means for Your Plan
Don't default to historical US averages. They produce reassuring numbers based on assumptions that overstate forward returns. The 7% real is the bull case, not the central case. Use 5-5.5% real for US equities as a defensible central estimate.
Run sensitivity scenarios. A robust plan should work at the central return assumption AND a bear case. If it only works at the central case, the plan depends on returns being at least as good as expected, which is a thin margin.
Use distinct assumptions per asset class. International equities have different forward-looking expectations than US equities; bonds different from both. The 60/40 portfolio's expected return is not a single blended number - it is the weighted combination of asset-specific assumptions.
Adjust withdrawal rate to the assumption. A 4% withdrawal rate at 5% real returns is materially different from 4% at 7% real returns. If the central case suggests lower returns, the safe withdrawal rate is also lower - typically 3-3.5% in the lower-return regime, not 4%.
Account for the equity premium explicitly. The case for equity allocation in retirement depends on the gap between expected equity and bond returns. Lower forward-looking premiums change the optimal allocation.
Pair conservative returns with fat-tailed distributions. The two corrections compound. Lower expected returns plus fatter tails produce noticeably worse plans than the historical-average normal-distribution baseline most calculators use. That gap is not pessimism. It is the realistic risk profile that the optimistic baseline was hiding.
Configure the assumptions yourself. Most retirement Monte Carlo tools default to optimistic numbers. Override the defaults. The success rate the model gives you is only as good as the inputs you feed it.
The expected return assumption is where most retirement plans quietly fail. The math is correct; the inputs are too generous. A plan built on conservative central estimates that survives a bear case is the kind of plan that holds up in real markets. A plan built on historical US averages is a plan that will rank well in calculators and may not survive in life.
Frequently Asked Questions
- What expected return should I use in my Monte Carlo retirement plan?
- For US equities, a defensible central estimate is 5-6% real (after inflation), well below the historical 7% that backwards-looking calculators often default to. Forward-looking models from major asset managers cluster in the 4.5-5.5% real range as of 2025-2026. For bonds, 1-2% real is reasonable. Use a slightly lower number for international equities and adjust upward for fat tails to compensate for the volatility your model assumes.
- Should I use historical returns or forward-looking estimates?
- A blend. Pure historical returns assume the next 30 years look like the average of the last 100, which is generous given current valuations. Pure forward-looking estimates depend on which model you trust. The honest approach is to run the simulation at three return assumptions (historical, forward-looking central, and a bear case) and treat the spread as the real range of outcomes. A plan that survives all three is robust; one that survives only the historical case is fragile.
- How sensitive is the success rate to return assumptions?
- Extremely. A 1 percentage point reduction in expected real return - say from 7% to 6% - can drop the Monte Carlo success rate by 15-20 percentage points over a 30-year horizon. The same change increases the median ending portfolio by 30-40%. The expected return is the single most consequential input in the entire model, more so than the withdrawal rate or asset allocation.
- Why are forward-looking estimates so much lower than historical?
- Three structural reasons. Equity valuations are higher than the historical average (CAPE ratio mid-30s versus historical median of 16), which mathematically reduces forward returns. Bond yields are lower than the historical average, reducing the bond contribution. And the 20th century US equity premium was the highest of any major market - applying it forward assumes the next century looks like the most favorable one in the historical record.
- Should I use different return assumptions for stocks and bonds?
- Yes, always. Treating stocks and bonds with the same expected return defeats the purpose of asset allocation modeling. Use roughly 5-6% real for US equities, 4-5% for international equities, 1-2% real for bonds, and adjust based on starting yields and current valuations. The relationship matters as much as the absolute numbers - the equity premium over bonds is what drives the case for stock allocation in retirement.