Guide

Monte Carlo simulation for portfolio and retirement planning explained

Your retirement spreadsheet says a 7% average return and a 4% withdrawal rate leave you safe for thirty years. That math assumes markets deliver their long-run average in neat annual slices — which they never do. A crash in year one of retirement can drain a portfolio that would have survived if the same returns arrived in reverse order. Monte Carlo simulation replaces one imaginary future with thousands of randomized paths drawn from historical or modeled return distributions, then counts how often your plan survives. This guide explains what Monte Carlo adds beyond deterministic projections, how to set inputs (returns, volatility, inflation, correlations), how to read success probability and percentile outcomes, common modeling mistakes, and how simulation pairs with sequence-of-returns risk, asset allocation, and compound growth.

What Monte Carlo simulation is — in plain language

Monte Carlo simulation is a computational technique that repeats a financial model many times (often 1,000 to 10,000 iterations), injecting randomness at each step according to probability distributions you specify. Instead of asking "what happens if I earn exactly 7% every year," you ask "what happens across thousands of plausible sequences of good years, bad years, and sideways decades?"

For retirement planning, each iteration typically:

Starts with your current portfolio balance and planned contributions until retirement.
Draws a random annual return for each asset class (or the whole portfolio) from a distribution — often normal or log-normal.
Applies inflation to expenses and optionally to Social Security or pension income.
Subtracts withdrawals each year in retirement.
Records whether the portfolio hit zero before your planning horizon ends — a failure — or still had assets at the end — a success.

After all iterations, you report the success rate (e.g. "87% of paths left money at age 95") and often show percentile bands: median ending wealth, 10th percentile (bad luck), 90th percentile (good luck). The name comes from the Monte Carlo casino — random sampling as a substitute for closed-form math when reality is too messy for a single equation.

Why average-return spreadsheets fail retirees

Deterministic projections — "I have $1.2M, withdraw $48k (4%), earn 7%, done" — ignore three forces that dominate real outcomes:

Sequence-of-returns risk

During accumulation, bad early years hurt less because you are still contributing. During decumulation, selling shares into a bear market permanently reduces the share base that must fund decades of withdrawals. Two portfolios with identical average returns can end in opposite places depending on order. Monte Carlo naturally samples many orderings; a single-line projection hides the worst ones. See our sequence-of-returns guide for the mechanics.

Volatility drag and withdrawal interaction

High volatility lowers compound growth even when the arithmetic mean return is unchanged — a phenomenon sometimes called volatility drag. When you combine volatile returns with fixed dollar withdrawals, failures cluster in paths that hit a deep drawdown early. Monte Carlo surfaces that tail risk; a flat 7% assumption does not.

Inflation uncertainty

Real spending power matters. Modeling CPI as a random variable (or correlated with equity returns) shows whether a nominal "safe" withdrawal still funds healthcare and housing in high-inflation decades.

Key inputs — and how to set them responsibly

Garbage in, garbage out. A Monte Carlo engine is only as honest as its assumptions.

Return and volatility assumptions

Most planners use long-run historical averages for broad indices (e.g. U.S. equities ~10% nominal, investment-grade bonds ~5%) paired with standard deviations (~15–20% for stocks, ~5–8% for bonds). Some tools use geometric (compound) means rather than arithmetic means because investors experience compound paths. Be explicit about which you use.

Conservative planners haircut equity returns by 1–2 percentage points ("lower expected future returns") to account for high starting valuations or demographic headwinds. That is judgment, not physics — document it.

Correlation between asset classes

Stocks and bonds are not independent. In many crisis years, correlations spike toward +1 when diversification is needed most. A diversified asset allocation reduces but does not eliminate this. Monte Carlo should model a correlation matrix; assuming zero correlation overstates diversification benefits.

Withdrawal strategy

Fixed real withdrawals (increase with inflation each year) are the standard stress test behind the famous 4% rule research. Dynamic strategies — cut spending 10% after a 20% portfolio drop, cap raises, guardrails — materially raise success rates in simulation because they reduce sequence risk. Model the strategy you will actually follow, not the most optimistic one.

Taxes, fees, and income sources

Include expense ratios, advisory fees, RMD-driven taxable withdrawals, and partial income from Social Security or pensions. Pre-tax vs Roth account ordering changes after-tax sustainability.

Planning horizon and legacy goals

Success is defined relative to a horizon (age 90, 95, 100) and sometimes a legacy floor ("leave at least $200k"). Longer horizons and higher legacy targets lower success rates — state both clearly.

Reading the output — success rate and percentiles

The headline number is probability of success: the share of simulated paths that never ran out of money before the horizon. Common planning targets:

90%+ — conservative; many advisors aim here for clients who cannot return to work.
75–85% — moderate; acceptable if you have flexibility to cut spending or earn part-time income.
Below 70% — a warning sign; adjust savings, retirement age, allocation, or spending before relying on the plan.

Success rate alone hides magnitude. Pair it with:

Median terminal wealth — typical leftover balance.
10th percentile outcome — bad-luck scenario; if this is near zero, you are fragile.
90th percentile — upside; useful for legacy and charitable goals.
Failure age distribution — when paths run out (year 8 vs year 28 implies different fixes).

A plan with 85% success but failures clustered in year 12 of retirement needs a different fix than 85% success with failures only after age 92.

Monte Carlo vs other retirement frameworks

Approach	Strength	Weakness
Deterministic average return	Simple, intuitive	Ignores sequence risk and tails
Historical backtesting (e.g. 1926–present)	Uses real correlated paths	Limited sample size; past may not repeat
Monte Carlo (parametric)	Many paths; tunable assumptions	Depends on distribution choice
Safe withdrawal rate rules (4%, guardrails)	Actionable starting point	One-size-fits-all; may be too conservative or aggressive

Best practice combines them: use Monte Carlo for personalized probability, historical periods for sanity checks, and guardrails for spending policy when markets move. Tools like cFIREsim, Portfolio Visualizer, and many advisor platforms implement variants of these methods.

Modeling pitfalls — fat tails, mean reversion, and overconfidence

Standard normal or log-normal returns assume extreme crashes are rarer than they have been in real markets (fat tails). Underestimating tail risk overstates success rates. Some advanced models use Student's t-distributions or regime-switching (bull vs bear states) to fatten tails.

Mean reversion — the idea that bad decades are followed by good ones — is debated. Monte Carlo that draws independent annual returns assumes no memory; models that embed mean reversion can look more optimistic after crashes. Neither is provably "right"; sensitivity analysis across assumptions is the antidote.

Other common mistakes:

Using arithmetic mean returns without converting to compound equivalents.
Ignoring investment fees and tax drag on rebalancing.
Modeling 100% success as the goal — chasing certainty forces extreme frugality or over-saving.
Running Monte Carlo once and never updating as life changes (marriage, health, inheritance).
Letting a high success rate justify behavioral overspending — probability is not a guarantee.

A practical workflow for DIY planners

Inventory — balances by account type, expected contributions until retirement, fixed and discretionary spending.
Choose a tool — spreadsheet with a Monte Carlo add-in, open-source retirement calculators, or fee-only advisor software.
Set conservative base case — haircut equity returns, include fees, model inflation at 2.5–3%.
Run sensitivity — retire two years earlier, spend 10% more, equity allocation +/- 10%. Watch which lever moves success rate most.
Define spending rules — write down what you cut if portfolio drops 20% (simulations with dynamic withdrawals are only meaningful if you will follow them).
Revisit annually — after major market moves, job changes, or health events. Monte Carlo is a living model, not a one-time certificate.

Production checklist

Model returns with appropriate mean (geometric vs arithmetic) and volatility per asset class.
Include correlation matrix; stress-test crisis correlation spikes.
Separate pre-tax, Roth, and taxable accounts with realistic withdrawal ordering.
Include inflation on expenses; model partial guaranteed income.
Use a withdrawal policy that matches real behavior (fixed vs dynamic).
Report success rate plus 10th/50th/90th percentile terminal wealth.
Run sensitivity on retirement age, spending, and allocation.
Compare parametric Monte Carlo against at least one historical period backtest.
Document assumptions so future-you can audit decisions.
Pair probability with an Investment Policy Statement and behavioral guardrails — math does not execute itself.