Monte Carlo Simulation: Stress-Testing Your Portfolio Without Guessing

Why one backtest is not enough

Investors love a clean chart. It feels decisive. But a backtest is usually one path through history, and history only happened once. If you test a 60/40 portfolio from 1995 to 2024, you are seeing one sequence of bull markets, recessions, inflation shocks, and rate cycles. Change the order of those events and the ending can change a lot, even if the long-run average return is similar ^[1][2].

That is the core reason Monte Carlo simulation exists in portfolio work. It does not try to predict the future. It tries to map the range of plausible futures given assumptions about returns, volatility, correlations, and withdrawals ^[2][4]. In retirement planning, that is more useful than a single “best case” or “worst case” line because the investor’s real problem is not whether the portfolio can survive one historical path. It is whether it can survive many possible paths.

Why this matters: a backtest can make a strategy look safer than it is if the historical sequence happened to be kind. Monte Carlo forces you to confront the ugly truth that the same average return can still produce a wide spread of outcomes.

What Monte Carlo simulation actually does

At its simplest, Monte Carlo simulation generates thousands of possible return paths by repeatedly sampling from assumed distributions. Each path is one possible future. The result is not one answer but a cloud of answers: ending wealth, drawdowns, ruin probability, and the chance of meeting a spending goal ^[2][4].

For retirement, the logic is straightforward:

1. Choose starting wealth.
2. Choose an asset mix, such as 60% stocks and 40% bonds.
3. Specify assumptions for expected return, volatility, inflation, and correlation.
4. Apply a withdrawal rule, such as a fixed real withdrawal.
5. Run the sequence thousands of times.

The output is usually summarized as percentiles: the 10th, 25th, 50th, 75th, and 90th percentile outcomes. That is where the fan chart comes from. The middle line is the median path. The shaded bands widen over time because uncertainty compounds ^[2][5].

For readers who want the mechanics of risk measurement itself, our risk measurement guide is a useful companion. And if you are comparing risk-adjusted metrics, Sharpe vs. Calmar is worth reading before you start treating one ratio as a verdict.

A simplified 60/40 retirement example

Let’s use a deliberately simplified example. Assume a retiree starts with $1,000,000, holds a 60/40 stock-bond portfolio, and withdraws 4% of the initial balance in year one, then increases that dollar amount with inflation each year. That is close to the classic safe-withdrawal framing popularized by Bengen’s 1994 work, which found that historical withdrawal sustainability depended heavily on the starting valuation and sequence of returns ^[1].

Now compare two ways of thinking about the same portfolio:

Illustrative example only. The table below is not actual performance data. It uses assumed annual returns and volatility to show how a retirement path can branch out. Assumptions: 30-year horizon, 60/40 mix, annual rebalancing, 4% initial withdrawal indexed to inflation, 6.0% nominal expected return, 10.0% annual volatility, 2.5% inflation, 0.60 stock-bond correlation, zero taxes, and zero transaction costs. These are educational assumptions, not forecasts.

The point is not the exact dollar range. The point is the spread. A retiree can have a “reasonable” expected return and still face a meaningful chance of failure if the first decade is poor. That is sequence-of-returns risk, and it is one of the main reasons retirement research moved beyond simple average-return thinking ^[1][2].

How the fan chart tells the story better than one line

A fan chart is the visual shorthand for Monte Carlo. Picture a line chart of portfolio value over time, then shade the area around the median path. The shading is narrow at the start and widens as the horizon extends. That widening is the message: uncertainty compounds.

Here is a practical way to read it:

Common mistake: investors often stare at the median line and ignore the lower bands. That is backwards. If you are planning withdrawals, the lower tail is the part that can force lifestyle cuts, not the median.

For a broader context on how uncertainty and diversification interact, see correlation and diversification and rebalancing.

Worked example: what a 30-year withdrawal path can look like

Below is a worked example using a simplified annual model. It is not a forecast and not actual backtest performance. It is a teaching device designed to show how the same starting portfolio can end in very different places.

Now compare that with a different order of the same broad return environment:

The average return over the three years is not the whole story. The first sequence is kinder because the portfolio gets an early lift before withdrawals bite. The second sequence is harsher because losses arrive first, when the account is largest and withdrawals are already underway. That is the retirement version of “the order matters.”

Practical takeaway: if your plan only works when the first decade is favorable, it is not robust. Monte Carlo helps you see that fragility before real money is on the line.

What investors get wrong about Monte Carlo

The biggest mistake is treating Monte Carlo as if it were a truth machine. It is not. It is a structured guess. If the inputs are weak, the output is polished nonsense. That is the classic garbage-in-garbage-out problem, and it is especially dangerous in retirement planning because the model can produce a neat success probability that feels more authoritative than it deserves ^[4][5].

Here are the three errors I see most often:

• Overconfident return assumptions. A 7% expected return is not a law of nature. Small changes in expected return can materially change success rates over 30 years.
• Static regime assumptions. Many simulations assume the future behaves like the past with the same volatility and correlation structure. Markets do not always cooperate. Inflation shocks, rate resets, and valuation compression can break the pattern ^[5].
• False precision. A 10,000-iteration run sounds rigorous, but if the assumptions are shaky, the extra decimal places are cosmetic. More simulations do not fix bad inputs.

This is where regime detection becomes relevant. A model that ignores changing market regimes may be elegant and still wrong. Likewise, backtesting pitfalls apply here too: the danger is not only overfitting the past, but also mistaking a convenient assumption set for a durable plan.

Pfau’s retirement-income research is useful because it emphasizes that retirement outcomes depend on spending flexibility, asset allocation, and the interaction between returns and withdrawals—not just on a single average return number ^[2].

Limitations: what Monte Carlo misses

Monte Carlo is powerful, but it is not omniscient. It usually assumes that returns are drawn from a stable statistical process. Real markets are messier. Correlations can jump in crises. Inflation can surprise. Bonds can fail to diversify when investors need them most. And regime shifts can make yesterday’s “normal” volatility look quaint ^[4][5].

That matters because the model can understate the risk of clustered bad years. It can also miss structural breaks: a new inflation regime, a policy shock, a valuation reset, or a credit event that changes the distribution itself. In other words, Monte Carlo is good at exploring uncertainty within a framework. It is weaker at predicting when the framework itself changes.

There is also a behavioral limitation. Investors often use the output as a yes/no answer: “My plan has an 82% success rate, so I’m fine.” That is too simplistic. A plan with an 82% success rate can still fail in the exact sequence that matters to you. The right question is not whether the number is above some threshold. It is whether the downside is survivable.

Editorial judgment: the honest value of Monte Carlo is not that it gives you confidence. It gives you humility. If a retirement plan only works under a narrow band of assumptions, the model has done its job by exposing that fragility early.

Decision tree: when Monte Carlo is useful, and when it is not

Use this simple decision tree before you trust any simulation output.

This is where the method becomes practical rather than academic. If you are comparing retirement strategies, Monte Carlo is most useful when paired with a spending policy, a rebalancing rule, and a realistic view of inflation. If you are still deciding how to build the portfolio itself, stocks vs. bonds vs. cash and inflation and real returns are the right foundations.

Checklist: how to evaluate a Monte Carlo result

Before you act on a simulation, run through this checklist.

If you want a broader framework for evaluating strategy quality, the benchmarking problem is a useful companion read. And if you are comparing systematic approaches, systematic vs. discretionary helps explain why process matters as much as outcome.

So what should an investor actually do with this?

For retirement investors, the best outcome is not the highest median ending balance. It is a plan that survives a bad sequence without forcing panic decisions. Monte Carlo helps you see whether you have built that kind of plan—or just a plan that looked good in one historical chart.

Closing thought: the future will not arrive as a single line. It will arrive as a range. The sooner you start planning that way, the less likely you are to confuse luck with durability.