I thought I was saving for my own pension

The premise

If pensions depend on how many people are paying in, then as the number of payers and the number of pensioners draw closer, adding more people may not be the whole answer.

That’s a condition, not a conclusion.

My contributions went to someone else the same month

None of it went into a fund. Belgium’s state pension is pay-as-you-go: what I paid in this month went to someone drawing a pension this month. The money isn’t kept anywhere.

I thought I was saving for my own pension. For years I assumed the money deducted from my pay was building up somewhere in my name, and that I’d get it back when I retired.

There is something in my name: a record of how many years I’ve worked, out of a possible 45, and what I earned in each of them. That record decides what I’ll be paid. The record is real, but there’s no pot of money behind it. That’s why it looks so much like a savings account.

So it comes down to numbers

If it were savings, it wouldn’t matter how many other people there are. You would draw on what you had put in.

With pay-as-you-go, that’s what matters most: how many people are paying in, and how many are drawing a pension.

Five in work for every three pensioners

In Belgium in 2022, 5.03 million people were in work and 3.07 million were drawing a pension.

So for every three pensioners, there are five people in work.

By 2070, on the official projection, there will be 5.31 million people in work and 4.22 million pensioners — five people in work for every four pensioners.

As the numbers get closer, each worker carries more

A word of caution first. It’s tempting to say that if one worker supports one pensioner, they’ll have to hand over half their pay. That’s wrong. Pensions are smaller than wages, and the money comes from employers’ contributions and from taxes as well as from workers.

What you can say is this: as the two numbers get closer, keeping pensions at today’s level means each person paying in has to carry more.

The European Commission’s figures show it. Belgium’s public spending on pensions rises from 10.6% of GDP in 2022 to 12.1% in 2070. The report breaks that rise into four parts. Just one of them pushes spending up: the number of people drawing a pension relative to the number working. The other three pull it down.

The usual answer: more people

Immigration usually comes up first. It adds people of working age — that much is simple arithmetic.

But being of working age isn’t the same as paying in. Language, getting qualifications recognised and residence status all stand in between, and each of them takes time. I didn’t measure how long.

There’s another snag before that. The European Labour Authority’s 2025 report lists 2,617 occupations with shortages and 2,177 with surpluses, in the same document and for the same year. Some places are short of people while others have more than they need. The surpluses are mostly in clerical and elementary jobs.

So the issue is less a shortage of people than a mismatch between people and jobs. Raising the headcount doesn’t guarantee that the gaps get filled.

Businesses are betting on something else

What businesses are spending money on isn’t just people. It’s machines and software that let one person do the work of several. Eurostat reports that more and more companies are using artificial intelligence.

So the direction things are actually moving in is this: rather than adding people, get more out of each one.

That raises the question at the heart of this piece. Will pensions be paid for by the number of people working, or by what those people produce with the help of machines?

That only works if two things happen

“Output will pay for it” holds only if both of these are true:

Output actually rises
The gains come back as wages, and so as money for pensions

One without the other isn’t enough. Output could rise tenfold, but if the extra doesn’t reach the people working, the money going into pensions doesn’t grow.

So far, neither has happened

Output first. Labour productivity per hour in the EU rose 0.2% in 2024 and 1.4% in 2025. And Eurostat’s own assessment is that average productivity growth slowed between 2008 and 2019, and kept slowing between 2019 and 2025.

Companies are taking up AI quickly, but it doesn’t yet show in the overall figures. Both can be true at once. So this piece doesn’t claim that machines have raised productivity, or that they can’t.

Then the gains. The International Labour Organization found that the share of global income going to workers fell by 0.6 percentage points between 2019 and 2022, and has been flat since. Had it stayed at its 2004 level, it says, workers would have earned $2.4 trillion more in 2024 alone.

It also says that technological change, including automation, has played a part in this, and that workers aren’t getting a fair share of the gains from higher productivity.

So the question is about shares, not headcounts

Until now, pensions have depended on how many people are working. From here on, things seem to be moving towards pensions depending on how much each person produces.

But if output rose tenfold and the gains didn’t come back to workers or to pensions, technology wouldn’t solve the pension problem.

So the question to ask is less how many of us there are than where the gains from higher output end up.

I didn’t measure that here. Two earlier pieces stopped at the same point. That three separate lines of inquiry have ended up in the same place is as far as this piece goes.

Why the usual figure tells a different story

The figure most often quoted is a different one. On 1 January 2024 the EU’s old-age dependency ratio was 37.0%: fewer than three people of working age for every person aged 65 or over. In 2004, it was just under four.

But that ratio counts people by age. A pension system counts people in work and people drawing a pension. Two groups make the difference.

Some people of working age aren’t working. In 2025, 76.1% of 20-to-64-year-olds in the EU had a job — 197.7 million people, the highest share since the series started in 2009. Even in the best year so far, the rest are counted as working age but aren’t paying in.

Some pensioners are under 65. In Belgium in 2022, 3.07 million people drew a pension, but only 2.29 million were 65 or over — a gap of about 780,000. This is where early retirement, survivors’ and disability pensions come in.

So in Belgium, counting by age gives about three people of working age for every person aged 65 or over. Counting the pension system gives five people in work for every three pensioners. The age-based figure looks almost twice as good. The report even has a name for the difference: the last line of its table is system efficiency, with a value of 1.8.

Where I stand

I hadn’t seen these figures side by side until I researched this piece.

And I don’t think it was just carelessness on my part. The system shows me a personal record: how many of my 45 years I’ve worked, and what I earned when. Look at that long enough and it starts to feel like a savings account.

This piece doesn’t say what should be done. What I do know is that I was counting something different from what the system counts.

What I got wrong

Version 1 put the arithmetic first. It explained two ways of calculating the dependency ratio before asking the question, so readers were working through figures without knowing why. It showed the checking instead of asking the question. This version turns that round
Version 1 was missing a link between “the numbers are getting closer” and “so do we add people or raise output?”. While researching, I had cut “does AI raise productivity?” because an earlier piece covered the same subject — but not the same question. That piece asked what counts as work; this one asks what pays for pensions
At first I wanted to write that “many people think of their pension as savings”. I had no source for that. The sentence only worked once I made it about myself. The problem wasn’t a missing source; it was the subject of the sentence
A secondary summary gave a series supposedly from this report: “contributors per pensioner, 1.9 → 1.3 → 1.2”. The report’s table has no such row. Opening the report itself caught it
The material I started from mentioned “a time when six workers supported each pensioner”. The EU wasn’t at that level: in 2004 there were already fewer than four people of working age for each older person

What I checked, and what I couldn’t

Checked

Belgium’s state pension is pay-as-you-go and defined-benefit, which covers 99% of spending. The amount is calculated by applying an accrual rate of 1.667% or 1.33% to earnings over a career of up to 45 years, with an earnings ceiling of €71,519.98 in 2022
Belgium in 2022: 3.07 million pensioners, 5.03 million in work, 2.29 million aged 65 or over, 6.79 million aged 20–64. Projected for 2070: 4.22 million pensioners and 5.31 million in work. The report gives these figures in thousands
The ratios above are rounded. The exact divisions give 1.64, 1.26 and 2.96
The last line of that table is system efficiency, with a value of 1.8. The report’s breakdown of spending treats the dependency ratio and the employment rate as separate multiplying factors
Belgian public spending on pensions: 10.6% of GDP in 2022, 11.8% in 2050 and 12.1% in 2070. Of the 3.5-point rise between 2022 and 2070, the dependency ratio accounts for 6.5 points; the other three parts are negative
The EU old-age dependency ratio: 37.0% on 1 January 2024, 26.8% in 2004
The EU employment rate for 20-to-64-year-olds in 2025: 76.1%, or 197.7 million people, the highest since 2009
The European Labour Authority’s 2025 report lists 2,617 shortages and 2,177 surpluses
EU productivity growth per hour slowed after 2008 and kept slowing between 2019 and 2025
The global labour income share fell 0.6 points between 2019 and 2022 and has been flat since
World population passed 2 billion in 1927 and 8 billion on 15 November 2022

Couldn’t check

Where the gains from higher output went. That’s the question this piece ends on, and it stays open
Which occupations make up the 2,177 surpluses. This piece names none of them
How the roughly 780,000 pensioners in the gap break down
How long it takes people of working age who arrive through immigration to start paying in

What this piece does not claim

That the system will collapse
That artificial intelligence raises productivity, or that it can’t
That higher productivity can make up for fewer people
That my own mistake is one other people make
That immigration is the answer, or that it isn’t