Monte Carlo simulation
Running thousands of possible futures, built from past returns, to see the range of outcomes.
5 min readLesson 16 of 57
In short
Instead of one guess, it shows a spread: a middle path, the outcomes most paths land between, and how often money is lost. It's only as good as the past it's built from, so treat it as a picture of uncertainty, not a forecast.
Why it matters
Any projection of a portfolio's future runs into the same problem: nobody knows the order of the good and bad years. A single line growing at an average rate hides that. Two investors with the same average return can end up far apart, depending on when the crashes come.
A Monte Carlo simulation replaces the single line with thousands of possible paths, and shows the spread. It turns a vague worry, "what if it goes badly?", into numbers: what a typical outcome looks like, what a bad one looks like, and how often the money ends up lower than it started.
How it works
The name comes from the casino: the method relies on random draws. Each simulated path is built by repeatedly picking returns at random and stringing them together. Do that thousands of times and the paths fan out, with most bunched in the middle and a few out at the extremes.
Where the returns come from matters most. MarketLab's Projection uses a block bootstrap: it takes your portfolio's own past daily returns at today's weights, using up to the last three years, and draws whole month-long blocks of 21 sessions at a time, in random order, until it has ten years. Drawing whole blocks keeps calm stretches and stormy stretches together, the way they come in real markets, rather than mixing single days.
It builds 2,000 such ten-year paths. At each month it sorts them and shows two bands: the middle 50% of paths (from the 25th to the 75th percentile) and the middle 90% (from the 5th to the 95th), with the median, the middle path, as a line. It also counts what share of paths are below the starting value after 1, 5 and 10 years: the chance of a loss.
The result is not a forecast. It assumes the future will look like a reshuffled version of the recent past. If the last three years were unusually strong, the paths will be too optimistic, and they can't include a kind of crash that wasn't in the history.
A worked example
Take a $10,000 portfolio and build two short paths of three months each. Each month is one block drawn at random from the portfolio's history, and each block's 21 sessions compound to one monthly return.
- Path 1 draws blocks that returned +2%, −5% and +3%. It grows by 1.02 × 0.95 × 1.03 ≈ 0.998, ending at about $9,981: a small loss.
- Path 2 draws blocks that returned +3%, +2% and +2%. It grows by 1.03 × 1.02 × 1.02 ≈ 1.072, ending at about $10,716.
- Repeat until there are 2,000 paths, then sort their end values from lowest to highest.
- The middle 50% band runs from the 500th to the 1,500th path, and the 90% band from the 100th to the 1,900th.
- If 260 of the 2,000 paths end below $10,000, the chance of a loss over that horizon is 260 ÷ 2,000 = 13%.
Result: The same history, reshuffled, gives very different endings: the two paths end over $700 apart after only three months. Over ten years the bands spread far wider, which is the point: the simulation shows a range of plausible outcomes rather than one number.
How to read it
- The median line
- Half the paths ended above it and half below. It's a middle outcome, not the expected one, and not a promise.
- The 50% band
- Where the middle half of paths land. Think of it as a fairly ordinary range of luck, good or bad.
- The 90% band
- Where nine in ten paths land. The bottom edge is a bad but not extreme outcome; real life can still fall below it.
- Chance of a loss
- The share of paths below the starting value. It usually falls as the horizon lengthens, for a portfolio that rose in the past, because good and bad stretches have more time to even out.
In the real world
- Financial planners use Monte Carlo tools to judge retirement plans, often reporting a "probability of success": the share of paths in which the money lasts. Pension funds use the same idea to see how likely they are to meet what they owe.
- Different tools draw returns in different ways. Some assume a bell curve with a chosen average and volatility; others, like MarketLab's, resample real history. Resampling keeps the fat tails of real returns, but it can only repeat the kinds of days that already happened.
- Results are very sensitive to the inputs. A model fed a strong decade will show rosy bands, and one fed a decade with a crash will look grim. Professionals usually run several sets of assumptions rather than trusting one.
Common mistakes
- Treating the median as a forecast. It's the middle of a set of reshuffled past returns, not a prediction of what will happen.
- Reading the bottom of the 90% band as a worst case. One path in 20 ends below it, and the future can contain falls worse than any in the history used.
- Forgetting that recent history drives it. A portfolio of stocks that soared lately will show high projected returns, which rarely repeat that neatly.
- Changing holdings until the projection looks good. That's Overfitting the past, applied to a projection instead of a backtest.
See it in MarketLab
Portfolio projection →
On the portfolio's Risk tab, open the Projection view to see 2,000 possible ten-year paths for your portfolio, the 50% and 90% bands, the median after 1, 5 and 10 years, and the chance of a loss at each.
Goal planner →
In the same Projection view, add monthly contributions or withdrawals, a target amount and inflation, and see how the paths, and your chance of reaching the target, change.
Value at Risk →
Read about Value at Risk for the one-day view of the same question: how much a bad day can cost.
Test yourself
Optional: 4 quick questions on the words, using the idea, and how it plays out in real markets.
Loading your answers…