Monte Carlo simulations and "probability of success", explained
By Investor Wisdom Daily Team·
Imagine a hypothetical retiree, Dana, who is shown two pictures of the same plan. One is a single smooth line: “at 5% a year, your savings last.” The other is a fan of hundreds of wiggly lines with a headline number: “80% probability of success.” Both used a 5% average return. Why do they disagree, and what does that 80% actually count?
This lesson answers that with arithmetic you can do on paper, including a tiny simulation run with a six-sided die.
The idea in one paragraph
Section titled “The idea in one paragraph”A projection applies one return to every year. A Monte Carlo simulation instead draws a different random return for each year from an assumed distribution (a menu of possible returns and how likely each is), runs the plan through those returns, and repeats this many times. Each run is called a path or trial. The probability of success is simply the share of paths in which the money lasted (or a goal was met): successful paths ÷ all paths. Percentiles describe the spread of results: the 10th percentile is the outcome that only 10% of paths fell below. None of these numbers is a forecast. They are the consequences of the assumptions you fed in.
The reason one average line misleads is sequence-of-returns risk: once money is being withdrawn, the order of good and bad years changes the result, not just their average. If volatility and drawdowns are new to you, read Drawdown and concentration risk and the standard-deviation section of Sharpe and Sortino ratios first; this lesson does not repeat them.
Example 1: same average, different order
Section titled “Example 1: same average, different order”Two hypothetical 3-year paths have the same three returns, in opposite order. The simple average is (20 + 10 − 15) ÷ 3 = 5% in both.
- Path A: +20%, +10%, −15%
- Path B: −15%, +10%, +20%
With no withdrawals, order does not matter. Multiplication can be done in any order: 1.20 × 1.10 × 0.85 = 1.122, so $100,000 becomes $112,200 on either path.
Now take $10,000 out at the start of each year, then apply that year’s return:
| Year | Path A | Path B |
|---|---|---|
| 1 | (100,000 − 10,000) × 1.20 = 108,000 | (100,000 − 10,000) × 0.85 = 76,500 |
| 2 | (108,000 − 10,000) × 1.10 = 107,800 | (76,500 − 10,000) × 1.10 = 73,150 |
| 3 | (107,800 − 10,000) × 0.85 = 83,130 | (73,150 − 10,000) × 1.20 = 75,780 |
Same returns, same average, same withdrawals, yet Path A ends $7,350 ahead (83,130 − 75,780). In Path B the loss hit the largest balance and the withdrawals were then taken from a shrunken pot, so less money was left to enjoy the later gains.
For comparison, a flat 5% projection gives 94,500, then 88,725, then $82,661.25. It looks like neither path. A single line cannot show you the Path B possibility at all.
What a Monte Carlo simulation does
Section titled “What a Monte Carlo simulation does”Example 1 hand-picked two orders. A simulation picks a return distribution (an average, a typical swing and a shape), draws one return per year for each path, runs the plan (withdraw, grow, check whether the money ran out), repeats this thousands of times, then counts successes and sorts the ending balances to read percentiles. The computer only does the repetition.
Example 2: a simulation by hand with a die
Section titled “Example 2: a simulation by hand with a die”Here is a deliberately tiny model you can reproduce. Roll one six-sided die for each year and look up that year’s return:
| Die roll | 1 | 2 | 3 | 4 | 5 | 6 |
|---|---|---|---|---|---|---|
| Return | −20% | −5% | +5% | +10% | +15% | +25% |
Each face has a 1 in 6 chance, so the average return is (−20 − 5 + 5 + 10 + 15 + 25) ÷ 6 = 5%.
The plan: start with $60,000 and withdraw $20,000 at the start of each year for 3 years. A path succeeds if every withdrawal is covered. The flat-5% projection says yes: (60,000 − 20,000) × 1.05 = 42,000; (42,000 − 20,000) × 1.05 = 23,100; (23,100 − 20,000) × 1.05 = $3,255 left.
Now five paths. These are the rolls we recorded; roll your own and you will get different paths, which is exactly the point.
| Path | Rolls | Returns | End of year 1 | End of year 2 | End of year 3 |
|---|---|---|---|---|---|
| 1 | 3, 4, 5 | +5, +10, +15 | 42,000 | 24,200 | 4,830 |
| 2 | 1, 1, 2 | −20, −20, −5 | 32,000 | 9,600 | runs out |
| 3 | 6, 2, 4 | +25, −5, +10 | 50,000 | 28,500 | 9,350 |
| 4 | 2, 6, 1 | −5, +25, −20 | 38,000 | 22,500 | 2,000 |
| 5 | 4, 3, 6 | +10, +5, +25 | 44,000 | 25,200 | 6,500 |
Each cell is (previous balance − 20,000) × (1 + return). For example, path 4: (60,000 − 20,000) × 0.95 = 38,000; (38,000 − 20,000) × 1.25 = 22,500; (22,500 − 20,000) × 0.80 = 2,000. Path 2 starts year 3 with $9,600, which is $10,400 short of the $20,000 withdrawal, so it fails.
Reading the results:
- Probability of success: 4 of 5 paths lasted, so 4 ÷ 5 = 80%.
- Sorted ending balances: $0, $2,000, $4,830, $6,500, $9,350. The median (middle) is $4,830.
- The worst path failed in year 3, something the single projection never showed.
- “Success” is not comfort. Path 4 counts as a success while ending with just $2,000.
With only five paths, each one moves the success rate by 1 ÷ 5 = 20 percentage points. Real tools run thousands of paths so that one lucky or unlucky path barely moves the answer.
Example 3: what changes if the swings are wider
Section titled “Example 3: what changes if the swings are wider”Keep the same 5% average but widen the die table:
| Die roll | 1 | 2 | 3 | 4 | 5 | 6 |
|---|---|---|---|---|---|---|
| Return | −30% | −10% | +5% | +10% | +20% | +35% |
The average is still (−30 − 10 + 5 + 10 + 20 + 35) ÷ 6 = 5%. Reuse the same rolls:
| Path | Returns | Year 1 | Year 2 | Year 3 |
|---|---|---|---|---|
| 1 | +5, +10, +20 | 42,000 | 24,200 | 5,040 |
| 2 | −30, −30, −10 | 28,000 | 5,600 | runs out |
| 3 | +35, −10, +10 | 54,000 | 30,600 | 11,660 |
| 4 | −10, +35, −30 | 36,000 | 21,600 | 1,120 |
| 5 | +10, +5, +35 | 44,000 | 25,200 | 7,020 |
Success is still 4 of 5, but the spread widened: the best ending rose from $9,350 to $11,660, the thinnest success fell from $2,000 to $1,120, and the failing path entered year 3 with $5,600 instead of $9,600. A flat-average projection shows no change at all. With thousands of paths, wider swings at the same average typically lower the success rate, as Try it shows.
What the model assumes
Section titled “What the model assumes”Every simulation rests on choices someone made. The main ones:
- The distribution: average return, size of swings (standard deviation) and shape.
- Independence: each year is drawn fresh, ignoring any clustering of bad years.
- Constant mean and volatility: the same menu for every year of a 30-year plan.
- Fees and taxes: a yearly fee comes out of every path; check whether a tool includes taxes (the linked simulator does not).
- Inflation: results in today’s dollars (returns after inflation) or future dollars. Mixing the two overstates what the money buys.
- Behaviour: fixed spending, although real people often cut back after a bad year.
Limits: what the number cannot tell you
Section titled “Limits: what the number cannot tell you”- False precision. “73.8%” sounds exact but inherits every guess in the inputs. Read it as a rough band.
- Fat tails. Real markets can produce more extreme years than a smooth bell curve suggests, so the worst simulated paths may not be bad enough.
- Regime changes. Long stretches of high inflation or poor returns break the “same menu every year” assumption.
- Hidden failure depth. A success rate counts whether money ran out, not how early or by how much.
- Not a probability about the world. It is a share of simulated paths. As the SEC’s beginners’ guide puts it, “All investments involve some degree of risk.”
Try it
Section titled “Try it”Open the retirement simulator. It runs a real Monte Carlo simulation in your browser: 10,000 paths, each year’s stock and bond returns drawn at random (lognormal, independent of each other), rebalanced yearly, fees taken off, spending withdrawn at the start of each year and returns after inflation, so everything is in today’s dollars. The “simulation number” fixes the random draws, so the same inputs always give the same answer.
- Defaults: $1,000,000 savings, $40,000 a year (a 4% withdrawal rate: 40,000 ÷ 1,000,000 = 0.04), 30 years, 60% stocks. 7,375 of 10,000 paths last, shown as 73.8%. The 10th-percentile ending balance is $0, meaning more than one path in ten ran out.
- Change the simulation number from 1 to 2. The result moves to 73.7%: that wobble is pure chance from sampling.
- Raise the stock swing from 17 to 22, keeping the 5% average. Success falls to 63.9%. Same average, wider swings, lower success: Example 3 at scale.
- Put the stock swing back and lower the stock average from 5 to 4. Success falls to 64.2%. One point of assumed return moved the answer by almost ten points.
- Lower spending to $35,000. Success rises to 86.4%.
(Figures from the simulator’s code with the inputs above, checked on November 19, 2026; they are model outputs under round-number assumptions, not forecasts.)
Common mistakes
Section titled “Common mistakes”- Reading 80% success as “80% chance my plan works.” It is the share of simulated paths under assumed inputs. Change the inputs and the number changes.
- Chasing 100%. A near-certain success rate usually means spending very little now; the trade-off is real either way.
- Ignoring how paths fail. Equal success rates can hide very different failure timing and depth.
- Trusting the average line. As Example 1 shows, withdrawals make order matter, and a single average return hides it.
- Forgetting the inputs are guesses. Averages and swings from one historical window are not guarantees for the next one.
Real-world context
Section titled “Real-world context”Retirement tools that use this method often show a success percentage, a fan chart and a few percentile balances. Look for the assumptions behind them: average return and volatility, before or after inflation, fees and taxes, number of paths, and what counts as “success”. Without those, the headline percentage cannot be interpreted. Asset allocation basics covers how the stock/bond mix you enter is chosen.
Quick check
Section titled “Quick check”- Two paths have the same returns in a different order and no withdrawals. Do they end with the same balance?
Answer to question 1
Yes. Growth factors multiply, and multiplication gives the same product in any order. Order only matters once money is added or withdrawn along the way.
- A simulation runs 1,000 paths. How much does one extra failing path change the success rate?
Answer to question 2
1 ÷ 1,000 = 0.001, or 0.1 percentage points. With 5 paths, one path is worth 20 points.
- You start a year with $50,000 and withdraw $30,000 first. What is the balance after a −10% year, and after a +10% year?
Answer to question 3
(50,000 − 30,000) × 0.9 = $18,000, or (50,000 − 30,000) × 1.1 = $22,000. The return acts only on what is left after the withdrawal.
- Plan X and plan Y both show 85% success. What else would you check?
Answer to question 4
The low percentiles (how bad the bad paths are), the year failures typically start, and the assumptions: average return, volatility, inflation treatment, fees and taxes.
Key takeaways
Section titled “Key takeaways”- With withdrawals, the order of returns matters: the same average can leave very different ending balances.
- A Monte Carlo simulation is many plan runs, each with randomly drawn yearly returns; the success rate is successful paths ÷ total paths.
- Percentiles and the depth of failures tell you more than the headline success rate alone.
- Results are only as good as the assumed distribution, independence, fees, taxes and inflation treatment, and small changes move them a lot.
- All figures here are hypothetical and educational, not a recommendation or a forecast.
Next steps
Section titled “Next steps”- Retirement simulator: run 10,000 paths with your own assumptions.
- Drawdown and concentration risk: how deep falls go and how long recovery takes.
- Asset allocation basics: how the stock and bond mix shapes the swings you simulate.
Sources
Section titled “Sources”- U.S. Securities and Exchange Commission, Beginners’ Guide to Asset Allocation, Diversification, and Rebalancing (August 27, 2009; accessed November 19, 2026): risk and reward, and the quote on investment risk.
- Investor Wisdom Daily, retirement simulator source code (
calc.cjs, model F-11), read November 19, 2026: what the linked tool simulates.
For educational purposes only. This material does not provide investment advice.