Monte Carlo Simulation in Personal Finance: Why Projections Fall Short

Type your numbers into almost any retirement calculator and it answers with a single, confident figure. Here is a real one. Take a 40-year-old with $250,000 invested, adding $2,000 a month, holding 80% stocks, and planning to retire at 65. Blend the usual long-run return assumptions and the calculator prints the result in a clean arc: about $4.3 million by retirement. It feels like an answer. It looks like a plan.

It is neither, and the gap between what that number looks like and what it actually means is the whole subject of this article. A projection hands you a destination. What you needed was the odds of arriving. Those are different questions, and the difference is not academic: run this exact plan through a proper simulation and the comforting $4.3M turns out to be one of the better outcomes, not the expected one. To see why, it helps to take the single number apart and put back, one at a time, the three things it quietly leaves out.

A projection is one guess wearing a suit

Start with what the generic projection actually is. Under the smooth curve sits one assumption: that your portfolio earns the same average return every year, on schedule, for twenty-five years. Nine percent this year, nine percent next year, nine percent through the crash you can’t see coming and the boom you can’t predict. The line is smooth because the math is simple: it is compound interest, not a forecast of markets.

That smoothness is the problem. Markets do not deliver the average on a timetable; they deliver something violent and lumpy that averages out to a number only in retrospect. A projection that assumes the average arrives evenly isn’t describing a likely future. It is describing the one tidy future that almost certainly won’t happen, and presenting it with the authority of a single number. It answers an easier question than the one you asked (“how much, if everything goes exactly to plan?”) and lets you mistake the answer for the harder one.

Nobody earns the average

Here is the first thing the line leaves out, and it is pure arithmetic before it is anything about markets. The average of your returns is not the return you compound. A portfolio that gains 30% and then loses 30% has an average annual return of zero, but you are down 9% — the loss bites a bigger base than the gain built. Volatility drags compounding below the average, always, and the more a portfolio bounces around, the wider that gap grows. The “average return” the calculator uses is a number almost no real investor actually earns over a real, jagged path.

So instead of assuming the average, we sampled it. We ran the identical plan 5,000 times through Killion’s simulation engine, letting returns vary the way they actually do, and recorded where each version landed. That turns the single line into a distribution. Below, each faint line is one of those runs: a single world the plan might live through.

The dashed line is the generic projection, unchanged: $4.3M. Notice where it falls. Not through the middle of the pack, where you’d expect an “expected” outcome — but up near the top, at the 71st percentile. The median outcome, the true middle of the distribution, is about $2.9 million — roughly a third less than the projected figure. And the projected $4.3M is actually reached in fewer than three runs out of ten. The single line wasn’t the expected result. It was the optimistic edge, wearing the costume of a forecast.

The worlds also show something a number can never show: the range. A tenth of the time this plan finishes below $1.4M; a tenth of the time above $7.6M. Same inputs, same discipline — a five-fold spread in outcomes, driven entirely by which returns happened to land in which order. That spread is not noise to be averaged away. It is the answer, and the projection deleted it to fit on one line.

Risk doesn’t arrive on schedule

At this point a fair objection arrives: fine, use a Monte Carlo simulation, plenty of tools do. And they do — but most of them make a quieter assumption that is nearly as misleading as the straight line. A basic Monte Carlo draws each year’s return independently from a fixed bell curve, as if every year were a fresh, unrelated coin flip around the long-run average. Bad years don’t cluster. A crash in stocks doesn’t drag bonds down with it. Inflation holds still. Risk, in that model, arrives politely and on schedule.

Real markets don’t work that way, and the difference is not cosmetic. Downturns persist — recessions and crises run in streaks, not single bad draws. Correlations break exactly when you’re relying on them: in a true crisis, the stocks and bonds that are supposed to zig and zag fall together. And inflation has regimes of its own. Killion’s engine models this directly, with distinct market regimes that cluster, spill into each other, and reprice risk together. So we ran the same household both ways — a naive independent-draw Monte Carlo, and the regime-aware engine — and compared the endings.

The naive simulation is more optimistic in every way that matters. Its median outcome is $3.7M against the realistic engine’s $2.9M. Its downside — the 10th percentile, the rough edge of a bad-but-not-catastrophic life — is $1.7M against $1.4M. And the share of futures that finish below a $3M goal climbs from 37% in the naive model to 51% once risk is allowed to behave realistically. A coin-flip, in other words, versus a comfortable “probably fine.”

The lesson isn’t that Monte Carlo is the magic word that makes a projection trustworthy. It is that how you model risk changes the answer as much as whether you model it at all. A simulation that assumes risk is mild, independent, and well-behaved isn’t neutral; it is optimistic by construction, and it hides exactly the clustered, correlated stretches that do the real damage to a plan. Not all simulations are equal, and the polite ones flatter you.

The investor is part of the model

There is one more thing every projection leaves out, and it is the one with your name on it. The math so far has quietly assumed a perfect investor — someone who sets an allocation at 40 and never touches it through twenty-five years of headlines, drawdowns, and 2 a.m. account checks. That person is rare. The plan on paper and the person living it are not the same, and the distance between them has a cost.

So we modeled the person. Killion’s engine can run the same household, over the same 5,000 market worlds, as two versions of you: one who stays invested, and one who panics — dumping to cash for six months after any 20% drop, the way fear actually behaves. Because both face identical markets, the difference between them is pure temperament.

The result is more interesting than “panic loses.” Look at the floors: they’re close, and panic-you even edges it — $1.8M against $1.4M — because sitting in cash genuinely dodges the deepest declines. The panic-seller is not wrong about crashes. But look at the top of the chart, where it’s forfeited. The median falls from $2.9M to $2.1M. The 90th-percentile upside collapses from $7.6M to $3.4M. And the odds of missing the $3M goal jump from 51% to 86%. Panic-selling rarely ruins you. It quietly caps you at mediocre — by selling out near the lows and missing the recoveries that were the entire reason to hold stocks in the first place. A projection that assumes you’ll sit still can’t see any of this, because it never let you move.

Watch a number become a probability

Stack the three corrections and watch the answer change shape. We began with a point: $4.3M, delivered as if it were the plan. Restore the distribution and the point becomes a spread of worlds with a $2.9M median, the projected figure marooned up at the 71st percentile. Let risk cluster and correlate the way it really does, and the downside thickens and the chance of falling short of $3M climbs past a coin flip. Put the actual investor back in, and a failure of nerve can shave another third off the middle of the range.

What’s left is no longer a number. It is a probability and a shape: a most-likely outcome, a spread around it, a measurable chance of falling short, and — though we held it implicit here — a window in time where that risk concentrates. That is the difference between a projection and a genuine one. The projection answers “how much.” The genuine version answers “how often does this actually work, and how badly does it miss when it misses?” Only one of those is something you can plan against.

The spread is where the decisions live

This matters because every real decision lives in the spread, not at the point. If your plan clears its goal at the median but fails it 40% of the time, the honest response isn’t to feel good about the median — it’s to ask what closes that gap: a higher savings rate, a few more working years, a more deliberate glide toward safer assets as the goal nears, or spending guardrails that flex when markets do. None of those moves is visible from a single projected number, because the number has no downside to defend against. You can only manage a risk you can see.

It also reframes the most boring advice in finance into something you can feel. “Stay invested” stops being a slogan and becomes the highest-leverage decision on the chart — the one that, in the behavior figure, separated a $2.1M life from a $2.9M one over the identical markets. A genuine projection doesn’t just tell you where you might land. It shows you which of your own choices actually move the odds, and which are noise.

The easier question, and the real one

The generic projection isn’t lying, exactly. It is answering a question — “how much will I have if everything averages out, on schedule, and I never flinch?” — and answering it correctly. The trouble is that almost nothing about that question is true. Returns don’t arrive on average, risk doesn’t arrive on schedule, and you are not a spreadsheet.

The real question was always the harder one: how often does this plan actually work, and what could break it? Answering it honestly costs you the comfort of a single number and gives you a range instead — but the range is the only thing you can actually steer. That is the entire case for a comprehensive simulation over a tidy projection, and it’s the kind of run you can do on your own numbers in Killion: not a destination printed in bold, but the odds of getting there, and the few decisions that move them.

Notes on the figures. Every number here was computed from a live run of Killion’s own simulation engine on a single hypothetical household — age 40, retiring at 65, $250,000 invested, $2,000/month in nominal contributions, an 80% stocks / 15% bonds / 5% cash mix — across 5,000 simulations (seed fixed for reproducibility). Figure 1 and the “realistic” column of Figure 2 use the engine’s regime-switching environment, in which returns, correlations, inflation, and cash yields shift with five clustering market regimes. The “basic Monte Carlo” column of Figure 2 uses independent, correlated yearly draws around fixed asset-class assumptions — no regimes. Figure 3 runs a disciplined holder and a panic-seller (sells fully to cash for six months after a 20% drawdown) over the same market paths, so the gap reflects behavior alone. All dollar figures are nominal; the $3M “goal” is illustrative. This article is educational analysis, not investment advice, and does not recommend any security or strategy.