Eva, 32, has €50,000 invested and adds €1,000 a month. One online calculator gave her a single number for age 62. A simulation told her “90%”. Neither told her what the number depends on, or what 90% of what.
A projection takes today’s numbers, a growth assumption and a contribution plan, and plays them forward year by year. It is not a prediction. It shows where today’s choices lead if the inputs hold, so you can decide whether you like the direction. Gut feeling is bad at this: people think in straight lines about something that compounds, and badly underestimate both the end result and the cost of starting late.
What a projection makes visible that gut feeling misses
The scale of the end result. €500 a month at a 5% real return is worth about €408,000 in today’s money after 30 years, far more than most people guess.
The cost of a delay. Starting five years later does not cost five thirtieths of the result; it costs closer to 30%, because the lost early years would have been multiplying the largest balance at the end.
When returns take over from contributions. Early on, most growth is what you put in; later, most of it is returns on what is already there. The crossover, often somewhere between years 15 and 25, puts a windfall or a temporary dip in savings in context.
What goes into a projection
- Starting balance by type of asset. Cash, shares, bonds and property grow at different long-run rates, so lumping them together is wrong from the start.
- Contributions. How much goes in, to which asset, whether it rises with salary, and when it stops.
- A growth rate for each asset, as a range rather than one number (below).
- Inflation, so the result can be shown in today’s money.
- The horizon and scheduled events: a house purchase, a career break, a mortgage paid off, an inheritance.
Get these right and a simple model is close enough; get them wrong and no model rescues the output.
One growth assumption moves the result by over a million euros
The output is extremely sensitive to the growth rate. For Eva’s €50,000 plus €1,000 a month over 30 years, in today’s money:
| Real return a year | After 30 years |
|---|---|
| 4% | about €847,000 |
| 6% | about €1,260,000 |
| 8% | about €1,910,000 |
The gap between 4% and 8% is about €1,060,000, more than a factor of two, from an input many people type without thinking.
That is why a single line misleads: it implies a precision the arithmetic cannot deliver. The fan is the honest picture. Many planners test the plan against the low line, so it does not depend on luck, and treat the high line as upside. How cautious to be depends on how much room you would have to adjust if the low line turned out right.
A quick way to sketch the fan for any plan, 30 years out, in today’s money: multiply the monthly amount by about 700, 1,000 and 1,400, and existing savings by about 3, 6 and 10, for 4%, 6% and 8% a year. For Eva: about €850,000, €1.3 million and €1.9 million. It is a rough guide for comparing assumptions, not a substitute for a proper calculation.
How to choose a growth assumption
A common approach has four parts:
- Use a range, not one number: a cautious, a central and an optimistic rate.
- Anchor on long-run history across many markets. Developed stock markets have returned about 4 to 6% a year after inflation since 1900, and the Global Investment Returns Yearbook puts world equities at 5.2% from 1900 to 2024. So 4% sits at the low end of that range, 6% at the high end, and 8% is optimistic against the long-run average, though some markets and periods did better. Bonds sit lower; cash lower still.
- Work in real returns, with inflation already taken off, so the numbers show what the money will buy. The purchasing power post covered why. A projection using nominal returns over decades overstates what you can afford by a wide margin; the inflation figure is itself an assumption that differs by country.
- Take off the fees you pay. A 7% market return is 6.5% after a 0.5% expense ratio and 6% after 1%.
Why growth rates get set too high
The overconfidence post explains the bias; in projections it shows up three ways.
Anchoring on the best recent decade. US shares from 2010 to 2020 did exceptionally well by long-run standards. Using that decade as the baseline for the next 30 years extrapolates from a lucky sample; some decades are much better and some much worse.
Confusing the return you need with the return you can expect. “I need 8% to hit my goal, so I will model 8%” turns the projection into a wish list. The market does not care what return you need.
Forgetting fees, which come straight off the return every year.
The counter to all three is the same: a range, long-run averages that span many market conditions, and a yearly update rather than defending an old projection.
Quick check: Eva needs 8% a year to reach her goal, so she models 8%. What is the problem?
The return she needs is not the return she can expect. 8% is well above the long-run world-equity average, so the plan would depend on a particularly strong outcome. Testing it against 4% and 6% shows how big the gap could be.
“In 90% of runs” is not a 90% promise
Even the fan assumes returns arrive smoothly: 6% every single year. Markets deliver an average as a mix of good, terrible and ordinary years, and that unevenness changes the outcome, not just the path. A Monte Carlo simulation runs the plan thousands of times, each with a different sequence of yearly returns, and shows how the results spread out.
Here is Eva’s plan run 10,000 times. Each year’s return is drawn at random from real history: the yearly returns, after inflation, of a portfolio spread equally across up to 16 developed stock markets from 1900 to 2020, lowered by 0.7 points so they average the 6% of the middle line while keeping their year-to-year swings.
scripts/simulate-projection-paths.mjs (seed 43).Reading the percentiles. The 10th percentile is the bad-luck path: only one run in ten did worse. The 50th is the middle; the 90th the good-luck path. For Eva they land at about €510,000, €1.08 million and €2.2 million. So in 90% of runs her pot passed about €510,000.
Why the middle sits below the single line. About three in five runs (61%) finish below the €1.26 million a steady 6% produces. A 20% loss followed by a 20% gain leaves you 4% down, so the same average return compounds to less when it arrives unevenly: these historical years average 6% but compound at about 4.9%.
Saying it honestly. “In 90% of runs the pot passed €510,000” is what the model says. “A 90% chance” drops the assumptions and turns an estimate into a promise about the future, which it cannot be: inflation, growth and how markets move together can all differ from the past. A common way to read it: look at the 10th percentile and ask whether you could live with it.
How the simulation was built and what it leaves out
The yearly returns come from the Jordà-Schularick-Taylor Macrohistory Database (release 6): an equal-weighted mix of up to 16 developed markets, 1900 to 2020, after inflation. Those years averaged 6.7%; each is lowered by the same 0.7 points so they average 6%, keeping their spread. The calculation can be rerun with scripts/simulate-projection-paths.mjs (seed 43). Monte Carlo methods were developed in the 1940s for physics problems that could only be solved by random sampling, and have been standard in retirement software since the 1990s.
What it leaves out: drawing from history includes the crashes that happened, but not worse ones that have not happened yet. It draws each year independently, so it understates how bad years clustered, as in the 1930s or the 1970s. Real investors sell in panics and chase what just went up; the model does not. And personal shocks, a job loss, an illness, a divorce, are not market events at all; they belong to the emergency fund and insurance.
Why order matters in retirement
While you are saving, the order of returns still affects the final balance, because contributions go in at different times: early falls let later contributions buy more cheaply, while late falls hit a larger pot. Once money is coming out, sequence-of-returns risk matters a great deal. Someone drawing €40,000 a year from €1 million who meets a 30% fall in the first year still needs €40,000, but now sells from €700,000, so a larger share of what they own goes at low prices and is not there for the recovery. Two retirees with the same balance, withdrawals and average return can end very differently if one meets the bad years early. A single line cannot show this, because it has no order; a simulation can, which is why it matters more near retirement. Ways of buffering it belong to the drawdown years and are covered later.
Quick check: in 90% of runs Eva's pot passed €510,000. Does she have a 90% chance of reaching it?
No. That is what the model produced from its assumptions, built on past returns. The future can differ, so “in 90% of runs” is the honest reading, not a promise.
Using a projection: the gap and the yearly update
The gap against a goal. A projection turns “will I be OK?” into something specific: “at 4% I am €80,000 short; at 6% I am €40,000 ahead; what would close the gap at 4%?” The usual levers are the savings rate, the date and the target itself, and which one fits depends on the person.
A yearly update. A projection built at 30 with assumptions from one year stops being useful a decade later if it is never touched. Many people update it once a year: the real starting balance, any scheduled events, and how far the last projection drifted from what happened, which shows how tight the assumptions were. A rerun that moves a few points is noise; a change in your situation is information.
Testing a decision before you make it, by changing one input and comparing, is the subject of the next post.
If your plan only works on the highest of the three lines, which of the levers would you reach for first?