Monte Carlo Simulation: The Maths Behind the Method

Imagine you are trying to predict the weather six months from now. You could pick a single forecast — "it will probably be mild, around 14°C" — and present that as your best estimate. Or you could run thousands of simulations using different starting conditions and atmospheric variables, and present a probability distribution: there is a 60% chance of mild temperatures, a 25% chance of cold, and a 15% chance of an unusually warm autumn.

The second approach is more honest, more useful, and considerably more complex. It is also, in essence, what Monte Carlo simulation does for retirement planning.

The name comes from the famous casino district in Monaco — a reference to the randomness and probability at the heart of the method. But the technique itself was born not in a casino but in a weapons laboratory in New Mexico, developed by some of the most brilliant mathematicians of the 20th century.

A brief history — from Los Alamos to your pension

The mathematics — what is actually happening

At its core, Monte Carlo simulation for retirement planning involves three components: a model of how investment returns are distributed, a mechanism for generating random sequences of returns from that distribution, and a retirement scenario to run those sequences through.

Step 1: Define the return distribution

Historical data tells us that annual investment returns are not fixed — they vary, sometimes dramatically. Equity markets have historically produced a long-run average of roughly 7–8% per year in real terms[7], but individual years range from gains of 40% or more to losses of 50% or more. This variability can be characterised statistically — typically using a normal or log-normal distribution defined by its mean (the average return) and its standard deviation (how spread out the returns are).

Step 2: Generate random return sequences

Using that distribution, the simulation generates a random sequence of annual returns — perhaps 30 or 40 years long, representing a full retirement horizon. This is done using a random number generator seeded with different values for each simulation run, producing a unique sequence each time. Crucially, some sequences will start with a run of good years; others will start with a crash. The distribution of starting conditions reflects the statistical properties of the historical data.

Step 3: Run the retirement scenario

Each random return sequence is applied to the retirement scenario: starting portfolio value, annual withdrawal, asset allocation, tax treatment, State Pension income, and so on. The simulation tracks the portfolio value year by year and determines whether it survived the full planning horizon — or ran out of money, and if so, when.

Step 4: Aggregate the results

After running thousands of simulations — odoPT runs 5,000 — the results are aggregated. The percentage of simulations in which the portfolio survived is the probability of success. The distribution of portfolio values at each point in time produces the fan chart — the spread of possible futures, from pessimistic to optimistic.

Why Monte Carlo is genuinely good for retirement planning

The sequence problem — what linear calculators miss

The most important strength of Monte Carlo simulation for retirement planning is its ability to capture sequence of returns risk. A linear calculator that assumes a fixed 6% return every year will produce the same projected outcome regardless of whether good returns come early or late in retirement. Monte Carlo does not make this assumption — it tests every possible ordering, and the results show clearly that the order matters enormously.

Two retirees with identical average returns over 30 years can have radically different outcomes depending on when the bad years arrive. Only probabilistic simulation can reveal this — and reveal how much it matters for your specific situation.

Honest uncertainty

Perhaps the most important thing Monte Carlo does is refuse to pretend that the future is knowable. A linear calculator says "your money will last until you are 87." Monte Carlo says "in 82% of simulated futures, your money lasts 30 years. In 18%, it does not." The second answer is less comfortable. It is also far more useful, because it tells you there is work to do — and gives you the tools to do it.

Where Monte Carlo can mislead you

Monte Carlo simulation is more honest than linear projection. It is not infallible. There are several significant limitations that users of any Monte Carlo tool — including odoPT — should understand.

The independence assumption

Standard Monte Carlo implementations assume that each year's return is statistically independent of the previous year's return. In reality, markets exhibit clustering[8] — bad years tend to follow bad years, and crises tend to be more severe and prolonged than random sampling from a normal distribution would suggest. A standard Monte Carlo model draws each year's return independently, which can underestimate the probability and severity of extended downturns.

The input assumption problem

Monte Carlo simulation is only as good as its inputs. The mean return and standard deviation used to define the return distribution are typically derived from historical data. But past returns do not guarantee future returns. A model calibrated on 20th century US equity returns — which were exceptionally strong by global historical standards — may overstate the probability of success for future retirees facing different return environments.[10]

This is particularly relevant for UK investors. UK equity market returns have historically been lower than US returns, and gilt yields have followed a different path. Using US-calibrated assumptions for a UK retirement plan will produce systematically optimistic results.

The false precision problem

Seeing "82% probability of success" feels precise. It is not. The actual probability depends entirely on the assumptions built into the model — return distribution, inflation, longevity, tax rates. Change any of those assumptions and the number changes. An 82% result from a generous set of assumptions might correspond to a 68% result under more conservative ones. The number should be treated as an indicator, not a measurement.

The 10% tail problem

A 90% success rate is often treated as "safe." But a 10% probability of failure is not negligible — it means that in one in ten simulated futures, the portfolio runs out of money. For a retiree facing a 30-year retirement, a 10% chance of financial ruin in old age is a risk that deserves serious attention, not a footnote. Good Monte Carlo tools show the downside scenarios explicitly, not just the headline success rate.

How odoPT is designed to address these limitations

odoPT is built with an awareness of Monte Carlo's limitations. Here is how the tool addresses each of the major concerns:

The crisis sequence approach — why it matters

The single most important enhancement odoPT makes to standard Monte Carlo is the ability to model structured crisis scenarios rather than relying entirely on randomly generated return sequences.

Here is why this matters. Standard Monte Carlo generates crashes — some simulated sequences will include severe years, and the distribution will produce some very bad outcomes. But the crashes generated by random sampling tend to be short, sharp, and followed by rapid recovery. Real market crises are often different: they are prolonged, they correlate across asset classes, and they arrive with particular timing patterns that random sampling does not fully replicate.

The 2008 global financial crisis took approximately 55 months to reach its peak drawdown and roughly 50 months to recover.[9] The dot-com crash took longer. These are not the short, sharp shocks that a normal distribution generates most naturally — they are sustained, structural events with specific profiles.

You can also stack crises — modelling a scenario where a crash is followed by a period of elevated inflation, or where two crises arrive within a few years of each other. These combined shock scenarios test the robustness of a retirement plan in a way that standard Monte Carlo cannot replicate.

Withdrawal strategies — the other half of the equation

Monte Carlo simulation models what markets do to your portfolio. Withdrawal strategies model what you do in response — and the interaction between the two determines the outcome.

A fixed withdrawal strategy — taking the same amount every year regardless of market conditions — is the simplest approach and the one most basic retirement calculators assume. It is also the most rigid, and the most vulnerable to bad sequences. If markets fall 30% and you continue withdrawing the same fixed amount, you are selling units at the worst possible time with no flexibility.

More sophisticated strategies — including the cash buffer band approach, floor-plus-upside, and expense-matched withdrawal — allow spending to respond dynamically to portfolio performance. In down years, spending falls slightly. In good years, it rises. This flexibility can dramatically improve retirement sustainability without requiring a lower average withdrawal rate.

odoPT models seven distinct withdrawal strategies and shows you how each one performs across the full range of simulated futures. The differences between strategies can be striking — and the right choice depends on your specific income needs, flexibility, and risk tolerance.

For a full comparison of the withdrawal strategies available in odoPT and guidance on which is most appropriate for different situations, see the withdrawal strategies guide.

What Monte Carlo cannot do — and what that means for you

Even the most sophisticated Monte Carlo implementation cannot predict the future. It can tell you the probability of different outcomes under a defined set of assumptions — it cannot tell you which future you will actually live in.

This is not a failure of the method. It is a fundamental property of complex systems. Markets are influenced by events that are not in the historical data — pandemics, wars, technological disruptions, policy changes — and no model can fully anticipate them.

What Monte Carlo can do is give you a structured way to think about uncertainty, to stress-test your plan against a range of plausible scenarios, and to make decisions that are robust across those scenarios rather than optimised for a single projected future.

A plan that survives 85% of simulated futures is more robust than one that survives 65% — even if neither can guarantee survival. The goal is not certainty, which is unavailable. The goal is a plan that is resilient enough across enough scenarios that you can retire with justified confidence rather than false certainty or unnecessary anxiety.

Monte Carlo simulation, used honestly and with an understanding of its limitations, is the best tool currently available for building that kind of plan. It is imperfect. It is still far better than the alternative.