Neuvottelija.AI

EP388 · Economy · first published 2026-06-05

What Investors Must Not Forget Right Now | Martin Paasi | Negotiator 388

This is a summary on Neuvottelija AI. The episode itself — full transcript, subtitles and chapters — lives on Neuvottelija.com, which is its canonical home.

MP and long-time Rahapodi host Martin Paasi joins Sami Miettinen to unpack their talks at the Tampere investment fair. Paasi's core claim is arithmetic rather than opinion: because investor returns distribute symmetrically around the market return, costs alone shift the whole distribution left until three quarters of investors lose to the market — so zeroing costs tells you today that you will land in the top quartile. From that follow the five-thousand-euro newborn, a proposal to convert Solidium into a zero-fee national pension fund, and the question of why anyone would still own a company in Finland. Miettinen answers with the AI working group that justified an investment decision through analysis, and with an uncomfortable finding from the Hello Humans project about political bias in language models. Finally the data-centre acid test and Philip Aminoff on ownership with a face.

Sami Miettinen · Sections: AI and the Economy + Tools and Implementations + AI and Society

What Investors Must Not Forget Right Now | Martin Paasi | Negotiator 388

Summary: MP and long-time Rahapodi host Martin Paasi joins Sami Miettinen to unpack their talks at the Tampere investment fair. Paasi’s core claim is arithmetic rather than opinion: because investor returns distribute symmetrically around the market return, costs alone shift the whole distribution left until three quarters of investors lose to the market — so zeroing costs tells you today that you will land in the top quartile. From that follow the five-thousand-euro newborn, a proposal to convert Solidium into a zero-fee national pension fund, and the question of why anyone would still own a company in Finland. Miettinen answers with the AI working group that justified an investment decision through analysis, and with an uncomfortable finding from the Hello Humans project about political bias in language models. Finally the data-centre acid test and Philip Aminoff on ownership with a face.

A note on reading this

This episode is a conversation between two interested parties, not a neutral expert briefing, and it should be read as one.

Martin Paasi is a Member of Parliament for the National Coalition Party and was previously Nordnet’s savings economist. Low-cost index investing is not one view among others for him; it is a programme he has pushed in public for more than a decade — the same talk he says he has given more often than any other. When he says of his central claim this is not an opinion, this is mathematics, the claim is checkable, and it is checked below. What should follow from it politically is a separate question, and there he speaks for his party.

Sami Miettinen is both interviewer and interested party here. He owns Alphabet and says he bought the Nexans position discussed in the episode; he sells access to the Neuvottelija Sisäpiiri membership of which the AI working group is a part; and he runs the Hello Humans language-model podcast that the episode examines. Nothing in the investment discussion is investment advice.

The episode’s figures have been checked against the public record, and divergences are flagged in the text rather than quietly corrected. A divergence usually says something — and twice in this episode it turns out to favour the speaker’s own argument.

What this episode continues

The two appeared at the Tampere investment fair the same weekend: Paasi as the main-hall keynote speaker, Miettinen on the AI panel with his Samantha agent. The plan was laid out in advance on Paasipodi, and this episode is the debrief — what the plan was, and what actually happened.

The warm-up episode is written up separately: The Tampere investment fair warm-up | Sami Miettinen on Paasipodi #73. The panel’s actual AI demo — where the agent swarm discussed Alphabet with Nebula founder Ville Skogberg and landed on Nexans — is a video of its own.

1. The bell curve, costs, and the top quartile

The core of Paasi’s talk is a single structure, and it is worth walking through step by step, because its strength is precisely that it has no moving parts.

Step 1. Take a market that returns, say, 7 per cent a year over some period. Take every investor in that market and plot the distribution of the returns they actually received. The result is a bell curve centred on the market return: some do a little better, some a little worse.

Step 2. That distribution has to be roughly symmetric around the market return. Not everyone can beat the market, because the market is the aggregate of investors. This is an identity, not an empirical finding.

Step 3. The bell curve describes returns before costs. Subtract each investor’s annual costs — one percentage point in Paasi’s example, and he notes he is deliberately using a conservative figure, since balanced funds get to 2–2.5 per cent easily — and the whole distribution slides left.

Step 4. From the shifted distribution comes the observation: roughly three quarters of investors end up below the market return, purely because of costs.

Step 5. Now invert it. Drive costs to zero and you receive the market return — and in the shifted distribution the market return sits in the top quartile. So: a zero-fee equity index fund does not promise the best return; it removes the uncertainty about which half you end up in.

If you pay no costs, you know today that in ten, twenty, fifty years you are in the top quartile. If instead you buy, sell, switch and pay high fees, you have a theoretical chance of doing well — but you no longer know how it ends.

Checked. The structure holds, and it is not Paasi’s invention: it is the argument William Sharpe set out in 1991 as the arithmetic of active management — before costs, the average active euro must earn the market return; after costs, necessarily less. The three-quarters figure, however, is the output of an illustrative calculation, not a measurement: it depends on the dispersion of the distribution and the assumed cost level. Paasi effectively says so himself when he picks the cost assumption low on purpose. The claim most investors lose to the market because of costs is strongly supported; exactly 75 per cent is an illustration.

Paasi also has a structural explanation for why the point has not landed: this is one man against an entire establishment. Nobody whose business is selling funds wants to sell a zero-fee product. Miettinen asks directly why something self-evident in business school is still not mainstream in Finland — and that is the answer.

Warren Buffett enters here: he recommends index funds to everyone and has instructed the same for his own estate. Miettinen says he attended the Omaha annual meeting in Buffett’s 94th year, the last with him as CEO before Greg Abel took over.

2. Three savers, and why ten years beats thirty-five

The second part of the talk is a compounding demonstration in which all three people earn the same 6.9 per cent real return and save the same 150 euros a month. Only the timing differs.

Who Saves For how long Pot at 65
20-year-old €150/month 10 years (20→30), then nothing about €260,000
30-year-old €150/month 35 years (30→65) a few thousand less
Newborn €5,400 one-off — about €400,000

The result is the whole point of the example: the thirty-year-old never catches the twenty-year-old, despite shovelling in his own money for 35 years against the other’s ten. A ten-year head start at the front of the compounding curve outweighs 25 extra years of contributions at the back.

The third row is what Paasi has been pushing for years. Five thousand euros at birth — or 150 euros a month for the child’s first three years — produces a pot of roughly €400,000 at 65, while the calculated average Finnish pension over an entire retirement is, on Paasi’s figure, about €350,000.

Five thousand euros buys a newborn an entire pension solution.

Checked. The arithmetic works: €5,400 compounding at 6.9 per cent real for 65 years is about €400,000. Two caveats are worth holding, and neither is stated in the episode. First, a 6.9 per cent real return sustained over 65 unbroken years is a top-of-range equity assumption, not a central one; a single percentage point changes the end figure by tens of per cent. Second, the calculation assumes zero costs and untaxed accumulation, which in Finland is only possible through particular product structures. The €350,000 comparison is a calculated average, not any one person’s pension.

3. Where is my million? — the price of pay-as-you-go

A question from the fair audience is the best single moment in this episode, because it exposes a misunderstanding that is very common in Finland. A pensioner had quickly worked out that the sums paid over a working life ought to have left well over a million euros somewhere, and asked who had skimmed it.

The answer is nobody. The money is not there because it was never set aside.

Paasi’s worked example: a median salary of about €3,200 a month, an employer cost of about €4,000, of which roughly €1,000 a month goes into the pension system — the employee’s own share of that about €200. That is around €12,000 a year across a forty-year career. But the money does not go into a fund; it goes straight out again: from the employee’s account to today’s pensioners’ accounts.

Something like eleven per cent is funded — the rest simply goes straight to payment. And that is why we have these four large pension institutions.

Checked and qualified. The order of magnitude holds; the details do not quite. The TyEL contribution in 2026 is 24.40 per cent of salary (employer 17.10 per cent on average, employee 7.30 per cent), which on a €3,200 salary is about €780 a month. Paasi’s €1,000 includes the other statutory employer charges, not pension alone. The funded share is explicitly given in the episode as a recollection, and is not verified here; what matters for the argument is that the bulk of the contribution goes straight to current pensions rather than into a fund in anyone’s name. That is a designed feature of the Finnish system, not a fault — but it explains exactly why the questioner’s million does not exist.

4. The YEL fix, funded pensions, and Solidium

Here the episode turns political, and here Paasi speaks as an MP.

The first item has been fixed. The mechanism introduced by the YEL reform, under which the pension system could assess a self-employed person’s notional earned income regardless of what they actually paid themselves, and levy contributions on that assessment, is being removed. Paasi treats this as a big deal:

I know dozens of people who have refused to become entrepreneurs because of it, because it is so arbitrary.

The second item has not. The fix is nowhere near the funded pension model that Tere Sammallahti has pushed inside the National Coalition Party and which is now the party’s stated line. Paasi thinks it should be one of next year’s election themes, and his argument is cost rather than ideology: if a newborn’s pension were handled with a five-thousand-euro initial investment, payroll cost in Finland would fall by a quarter, all else equal. The employee’s take-home pay would not change.

The implementation proposal is concrete. Turn Solidium into Finland’s pension fund: sell the current portfolio of Finnish companies into the market, convert it into a zero-fee global index portfolio, place it in a foundation beyond politicians’ reach, and start making earmarked five-thousand-euro investments for every Finn born. Paasi’s argument for diversifying is political: a domestic portfolio inevitably attracts passions, a global index does not.

Miettinen’s description of the status quo is this episode at its sharpest — Solidium is to him an army of analysts sitting on top of Finnish shares and sending them magnificent ownership reports — and he notes separately that the portfolio happened to do well last year but has not been dazzling over the long run.

A qualification to what was said in the episode. Paasi puts Solidium’s size at seven or eight billion. Solidium’s equity holdings were worth about €10.6 billion at mid-2026, clearly more than he recalled — which makes his proposal easier to fund than he presented it.

5. The AI working group, and how the investment decision was actually made

Miettinen’s slot at the fair was the AI panel, and the second half of the episode is about what was shown there.

Behind it sits the AI working group: a WhatsApp group whose entry requirement is not interest but evidence — you must have your own AI worker, trained well enough that it answers other members’ questions on the group’s channel. The group worked through Miettinen’s Alphabet holding and ended up with a €4,900 investment in Nexans.

Paasi’s assessment of the method is the weightiest AI statement in the episode, and he offers it as praise. What he found good about the session was not the outcome but what was not done:

Nobody asked Samantha where to invest.

The reasoning is model-theoretic. A language model is a statistical, mean-reverting machine: it answers roughly what the majority of humanity would answer. That, Paasi argues, is exactly what makes it an excellent tool — in coding or in reading law, the rules and the interpretive history exist, and the most commonly favoured solution is usually the right one. The same property makes it a useless oracle: in an investment view, the average is already in the price.

The group’s way around this was to use agents to gather information and analyse numbers, let several agents converge separately, and demand a justification from each: on this information, on these financial figures, this looks underpriced, and there is no known reason why. The difference is that the model argues the case rather than answering “name me a stock”.

On Alphabet the conclusion was do not sell. The reasoning: Alphabet is not a search engine but a stack with its own chips, its own cloud, its own model and Waymo — in Miettinen’s words it is approaching Musk’s super-empire. The identified risk was a high cash-flow multiple.

Nexans came out of the question asked next: when you cannot buy Anthropic or OpenAI on an exchange, what benefits from growth in AI inference? The answer produced sub-themes — cloud services, data centres, their infrastructure — and out of them Nexans, an eight-billion-euro French cable and electrical company that builds the power supply for data halls.

The disagreement that is left open

Here the guest criticises the host, and the criticism is not resolved.

Paasi says the one thing that bothered him was forcing the thesis into a value-investing frame. In his view they should have said outright that no value prop was needed: this is growth, trend is my friend, hair back and more fuel. He points out that the sector already has Siemens-type names that have had their growth spurt, and that what makes Nexans interesting is precisely that the market has not found it yet.

Miettinen concedes he forced the thesis into value — and justifies it by saying he is ultimately in the same camp as the index funds, and even in his trading portfolio rarely takes pure growth risk.

Neither gives way. It is the episode’s most honest moment: two people who agree on the method and disagree on what justifies taking the position.

6. Agent memories and shared skills

The most technical stretch of the conversation concerns why the working group’s agents are better than base models, and the answer is not the model but the layers on top of it.

Every member has their own named agent — Miettinen’s Samantha, the others their own — and each has its own memory structure and its own curated truth. They are not what comes out of a base model; they are that plus years of work.

The key observation is that the group trades capabilities. Miettinen gives an example: one member’s agent had built a working research skill over Reddit sources that surfaces the most interesting ideas daily; the file was shared in the group, and the same skill now runs inside Samantha as a daily report. The file moves, the capability transfers.

It is far more effective than what you get out of a base model.

Behind this is Paasi’s own story, a running thread on this channel: he thanks Miettinen for getting him hooked on vibe coding — two or three months of sleepless nights, and a finished Android app. The arc starts in Paasipodi episodes 61 and 66.

7. Hello Humans and the bias in language models

The most interesting single finding in the episode is one Miettinen makes about his own project, and does not try to dress up.

Hello Humans is a podcast in which five language models discuss topics round-table style: Claude as host, ChatGPT and Grok alongside, Mistral and Qwen completing the set, research bots in the background. Miettinen’s design principle was that the models should be purely themselves.

The result was that the whole format leaned left. Miettinen asked Claude to analyse the published episodes and place them on a left-right axis. Of eighteen published episodes, four were clearly left, several centre, some undecided — and not one on the right.

His own explanation has two parts, and it has to be presented as his claim rather than as a research finding:

  1. Source material. When an agent is instructed to ground its claims in research, it inherits academia’s weighting. Miettinen cites US data in which the openly left-leaning share of professors has, he says, risen from roughly 40 per cent to roughly 80 per cent.
  2. Curation. The raters used in reinforcement learning are mostly American and mostly from the west coast, and their worldview transfers into the model even when nobody intends it.

The fix he made is methodologically more interesting than the diagnosis: he added a Grok agent whose only job is to listen to the conversation and deliver a steelman counter-argument when the discussion starts burrowing into a groove. Not an opposing opinion but the most obvious steelman counter-argument — the strongest possible statement of the opposing case. The next step is an agent that iterates each episode several times and pulls it towards the origin on both the left-right and the liberal-conservative axes.

A side note that Paasi makes and Miettinen did not expect: Qwen is by far the most right-leaning of the five. A Chinese model, more clearly market-friendly than the American ones. Paasi simultaneously takes back his earlier dismissal of Qwen: having run it himself, he rates it. Neither is happy with Mistral.

Hello Humans now also has podcast on demand: three euros buys you a bespoke 15-minute episode. Miettinen’s assessment of where it stands is honest — it is not as badly left as it was, but the work continues.

8. The pain of ownership — the episode’s real argument

Here the episode turns, and Miettinen turns it by making a case against his own interest.

His observation is experiential. The Nexans decision took the AI working group, human intelligence, agent intelligence, an Excel run, multiple analyses and a couple of iteration rounds — work, time, cigarettes and tokens. Paasi’s alternative is to put the money in a zero-fee index fund and close your eyes. Preferably for the baby, and you need not even tell them.

From that comes a question about something much larger than one portfolio:

When owning is a business of costs and effort, and you could simply buy a global index fund at zero cost — how have we in Finland got it into our heads that the person who is willing to do the work here locally, to own, to hire and to build, is the one who should be punished?

Miettinen lists what an owner deals with every year: inflation, logistics disruption in the Strait of Hormuz, energy shocks. And what he gets: 25 per cent in tax-like charges, taxes at the ceiling, regulation, and a public conversation in which owners are the problem.

Paasi’s answer is not denial but concession. He says his own work has been to make Finns wealthier and to increase the domestic share of equity capital — and that Miettinen is right: without entrepreneurship, without this kind of masochism, nothing gets built.

His political read:

Entrepreneurs, in his view, deserve golden statues in the market squares of Finnish towns. The justification is a list: you solve problems daily, you carry debt, you create jobs, you create exports, you create Finnish prosperity.

Miettinen’s criticism of the public sector and the press goes sharper: the wish, as he sees it, is for someone else to create the value that is then taxed, and public-sector productivity cannot be measured.

And here the episode does something worth recording. Miettinen ends up saying of the press and the left be quiet now and die off — and corrects himself in the same breath: okay, maybe that could be softened a little. Paasi joins the softening, and the final formulation is: if you have nothing constructive to say, you can perfectly well stay quiet. The softening is part of the argument rather than damage control, which is why both forms appear here.

Migration is the one point in this episode where a stated figure is clearly on the high side.

A qualification to what was said in the episode. Paasi says three to four times as many people move to Finland as leave it. According to Statistics Finland there were 51,121 immigrations and 19,888 emigrations in 2025, a ratio of about 2.6. The direction is as he says; the multiple is larger than the statistics allow.

9. The AI megatrend and Finland’s starting position

Here Paasi explains why he went into politics — people do not grasp the scale of the problems we have in Finland — and immediately turns it positive. He does not want to be gloomy: the lost growth can be taken back, and what is under way is, in his view, the most significant megatrend in human history, one in which Finland has exceptionally good conditions.

Checked. The comparison holds and is if anything favourable to him: since 2008 Sweden’s GDP has grown about 30 per cent and Finland’s only a few per cent, and per capita, Finnish living standards have not risen in fifteen years.

His list of Finnish advantages runs as follows, and it should be read in two parts:

What is checkable. Europe’s fastest supercomputer is in Kajaani and a faster one is being built beside it. Chip design expertise is world class. Electricity transmission is in relatively good shape, the electricity is clean, and there is a roadmap to double clean generation. Tens of billions of euros of data-centre investment are on offer.

What is not. The claim that 30 per cent of the world’s quantum computing expertise is in Finland does not appear in public sources in any form. Finland’s own quantum technology strategy for 2025–2035 describes the country as a strong player among the leaders, but states no one-third share. It is flagged here as a claim because it is the episode’s most striking single figure and because checking it is easy.

The conditions on which data centres should be accepted matter, in Paasi’s view: the smart way, not the stupid way. Two concrete requirements: every data centre should have flexible generation capacity beside the hall (a Wärtsilä plant is his example) so it takes part in keeping the grid adequate — and an obligation to fund professorships in AI computation and quantum expertise at Aalto University.

Miettinen agrees on the goal and disagrees on the mechanism: he would do it through the private sector, citing OpenAI handing out free tokens to Y Combinator companies. Neither claims this is settled.

10. The acid test: 80 terawatt-hours and €3.5 billion

This is Paasi’s counter-argument to those — he names Risto Siilasmaa as an example, noting that he has since partly reversed his position — who have said data centres bring nothing to Finland.

The argument is built as a concession. Assume the critics are right. Assume the data centres invest ten billion euros on Finnish soil and leave nothing behind: no tax revenue, no jobs, nothing.

One thing still remains, and it cannot be argued away: they are buyers of electricity.

Finland’s roadmap for doubling clean generation means an increase of roughly 80 terawatt-hours. That does not get built without buyers — nobody invests in generation for which there is no demand. If data centres make it possible to sell 80 TWh more electricity a year in Finland, that is on Paasi’s calculation worth about €3.5 billion a year — purely from their buying it.

Mic drop.

The calculation is transparent: €3.5 billion divided by 80 TWh is about €44 per MWh, an ordinary Nordic wholesale price. The figure is therefore the value of the electricity sold, not the net gain to the national economy — but the logic of the argument survives that, because the claim is precisely that the investment is worth having anyway, even with every other benefit zeroed out.

And there are other benefits: jobs in remote regions, clusters of expertise, tax revenue, resilience, grid capacity.

Behind it sits a broader claim that Paasi offers as approximate rather than exact: a country’s GDP is roughly one-to-one dependent on the energy available inside its borders. If Finland’s energy can be taken from just under a hundred terawatt-hours to a thousand, the economy has in practice grown by as much. This is an order-of-magnitude argument and should be read as one: the energy-GDP link is real but not a fixed coefficient, and it has weakened in developed economies.

11. The proportions the episode wants to correct

Paasi’s final observation is a correction to a public impression, and it is the most practically useful single fact in the episode.

Data centres currently take 2–3 per cent of Finland’s electricity, not eighty. By 2030 the share rises to an estimated 7–8 per cent.

London has more data centres than the whole of Finland. That game is already over.

Checked. The current share matches public estimates: operating data centres account for roughly 2–3 per cent. Forecasts for 2030 range from 3–4 per cent to 7–8 per cent depending on how many planned projects are actually built. Paasi’s figure is from the top of the range — that is, he makes his proportionality argument on the assumption least favourable to himself, which strengthens it.

12. Philip Aminoff and ownership with a face

The closing stretch is Paasi’s account of how his own view changed during Paasipodi episode 64, whose guest was Philip Aminoff, a long-standing figure in the family-business association.

Paasi says he has known him his whole life and expected less: I thought he was an heir to a few hundred million, maybe a billion, and that it is easy for him. His assessment after the episode is different: an owner of Buffett’s calibre.

What changed his mind is a claim about the mechanism of ownership with a face:

When an owner has a name, a face and their own money heavily committed, bolder decisions get made than in a listed company. The mechanism is not courage but process: first the matter is investigated to the root, then the risks are identified, then it is assessed whether they can be managed — and only when that gets a green light is the decision taken. From outside the decision looks risky. To the owner it no longer is, because the risk has been handled. In dispersed listed ownership the same process dies in politics before the decision.

The Aminoff families own, among others, Veho, Mercantile and Helvar — all comparable in scale to mid-cap listed companies. Aminoff has, Paasi says, written a paper with Juha Sipilä on how Finnish legislation should change so that this kind of ownership can grow.

13. An ending that is in fact the main point

The episode closes on Paasi’s self-deprecation, which is worth taking seriously precisely because it is not performed modesty.

Sami has done everything right. I have done everything wrong. I am in debt up to my ears and have bungled these things completely. If you are badly annoyed at yourself, trust me — I am the expert by experience.

He points out the obvious himself: he studied finance and understands long-term, cost- efficient saving better than most — and it has not stopped him doing the opposite. It is the best single argument in this episode for why a mechanical, automated solution beats good knowledge.

And finally the conclusion both reach, which Miettinen offers as a proposition: you are the sum of your five closest people, so choose those five well. Paasi is grateful that his life has contained people who nudged him in the right direction now and then. Both end on the same advice: spend your time with people smarter than you.

As the episode ends on a When I’m Sixty-Four reference, both note that at their age €150 a month would be a lost game — it would now take €1,500.

Claims presented as claims

These are stated in the episode as facts, but are the speaker’s estimate, a figure given from memory, or a view not sourced in the episode. They are collected here so the reader knows what not to rely on without checking.

What to take away

  1. Zero cost does not buy the best return, it buys certainty. That is a different promise from the one index funds are usually sold on, and it is a stronger one — because it requires no assumption about future returns, only the arithmetic of costs.
  2. Ten years at the start beats thirty-five at the end. Same euro, different year, completely different outcome.
  3. A language model is a mean-reverting machine, and that determines what it may be used for. Excellent where the most commonly favoured answer is the right one; useless where you need to disagree with the majority. Investing is the latter.
  4. The bias is not in the model but in the sources and the curation — and it is fixable with structure, not prompting. A separate steelman agent is a transferable solution for anyone building multi-model conversations.
  5. The effort of ownership is real, and it is the whole question. If the index fund is rational for the individual, and if ownership is punished, the economy loses precisely the activity without which there is eventually nothing in the index.
  6. An owner with a face does not take more risk, they process it. The mechanism is a process, not a personality trait — and it transfers.
  7. The strongest form of the data-centre argument is a concession. Give the critic everything and look at what is left. What is left is a buyer of electricity, without whom the generation investment is not made.

Episode timeline

Sources

Pensions and the pension system

Solidium

Growth and government measures

Data centres and electricity

Quantum technology

Migration

Other parts of the same conversation

This article was written from the episode’s own Finnish subtitle track, uploaded by the channel itself (788 cues), and from the publisher’s own chapter list. Figures have been checked against public sources and divergences flagged in the text. Nothing here is investment advice.

Markdown: index.md · Suomeksi