Sisäpiiri 164 · Economy · first published 2026-10-10 · updated 2026-10-12

Statistics Finland's GDP revisions | Lasse Corin | Negotiator Inner Circle 164

In Neuvottelija Inner Circle 164, Sami Miettinen asks Aktia's chief economist Lasse Corin why Finland's GDP growth for the Marin government's years was revised down after the fact while the Orpo government's years were revised up. This long-form, sourced article checks Miettinen's figures against Statistics Finland's own revision table (18 September 2026), sets them against 2014–2020 and international revision studies, explains Statistics Finland's release and revision practice and the Kalevi Sorsa Foundation's forecasting report, and gives background to the second half of the conversation: measuring AI in the national accounts, real-time data, the digital euro, card-fee regulation and the payment protocols for AI agents.

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

Statistics Finland’s GDP revisions | Lasse Corin | Negotiator Inner Circle 164

Summary: In Inner Circle episode 164 Sami Miettinen puts a deliberately provocative question to Aktia’s chief economist Lasse Corin: are Statistics Finland’s preliminary estimates worse than even Greece’s? Miettinen’s case is that growth for the Marin government’s years was revised down after the fact and growth for the Orpo government’s years revised up. Corin does not accept the framing. He explains how the GDP flash estimate is produced and why it keeps being refined for years, and he offers an economic explanation for the difference. The second half turns to the future of measurement: the value produced by AI, real-time data, the digital euro, and AI agents with payment cards of their own.

What the check found: Miettinen’s year-by-year figures match Statistics Finland’s release of 17 June 2026. The release of 18 September 2026 revised 2022–2024 again, and the pattern sharpened. The Marin government’s full years 2021–2022 have now been revised by −2.0 percentage points in total, and the Orpo government’s years 2024–2025 by +1.8. The longer 2014–2025 series, however, does not support a bias that tracks the government of the day. The average revision is close to zero, and the largest single revisions in both directions fell in the Sipilä government’s years.

How to read this

• This is the Inner Circle follow-on to a main episode. Miettinen refers to the main episode in which he said he had moved his banking from Nordea to Aktia. That remark is in Finland’s economy +2.1% — is this the turn? (Negotiator 401), where Corin and Miettinen first went through the revisions.
• Interests. Corin speaks as Aktia’s chief economist, and Miettinen is an Aktia customer. The episode is not a commercial collaboration.
• The tone is deliberately provocative. Miettinen frames the revisions as an accusation (“what the hell again, Statistics Finland?”). In the same breath he says he is not alleging corruption in Finland. His actual claim is that the preliminary model is poor. The article reports that claim without adopting it.
• Source and expansion. The account of the episode is based on a Finnish MacWhisper transcript and the publisher’s own chapter list. Quotations are translated from Finnish. For this expanded version (12 October 2026), the figures were checked against Statistics Finland’s releases and the StatFin database, and the episode’s topics were researched in primary sources. All sources are listed at the end, and claims that could not be checked are listed separately.

1. Miettinen’s claim: revisions go in different directions by government

Miettinen says he compared Statistics Finland’s first and current growth figures:

• 2021: 3.3 → 2.7%
• 2022: about 2 → 0.8%
• 2024: −0.2 → 0.9%
• 2025: 0.2 → 0.8%

He concludes that the Marin years were credited with a cumulative 1.4 percentage points too much growth and the Orpo years, so far, with 1.7 points too little. He also notes that the downward corrections drew almost no coverage, while the zero-growth flash estimate drew plenty.

What Statistics Finland’s own table says

With every national accounts round, Statistics Finland publishes a table titled Tietojen tarkentuminen (Vuositilinpito) (“revisions to annual accounts”). It sets each year’s first published growth figure beside the latest one. The figures below are from the release of 18 September 2026, cross-checked against StatFin table ntp/15a9 (GDP at market prices, volume change):

The 2024 and 2025 figures are still preliminary and will be revised further.

Three points on Miettinen’s arithmetic:

1. The figures were right in June but are now out of date. Miettinen’s figures match the 17 June 2026 release exactly. In September Statistics Finland reported “small revisions” to GDP volume for 2022–2024. The cause was a reassessment, carried out in EU cooperation, of how certain foreign-trade and value-added items are recorded. Growth for 2022 fell from 0.8% to 0.6%, and growth for 2024 rose from 0.9% to 1.0%.
2. The total was miscalculated. On Miettinen’s own figures the Marin-year revision comes to −1.8 percentage points (0.6 + 1.2), not −1.4. On the latest figures it is −2.0 (0.6 + 1.4). The Orpo-year revision is +1.8 on the latest figures (1.2 + 0.6), not +1.7. Both errors only strengthen the direction of his claim.
3. What counts as the “first figure” is a matter of definition. The table’s first release is the full-year figure in February’s quarterly accounts. The first proper annual accounts, published in March, differed from it: 3.5% for 2021, 2.1% for 2022 and −0.1% for 2024. Measured from the March figure, the Marin-year revision grows and the Orpo-year revision shrinks slightly.

The longer series: the direction of revisions does not follow the government

Four years is too small a sample to show a government-specific bias. The same Statistics Finland table goes back to 2014, which lets the claim be tested:

Three things follow from the table. The calculations are this article’s, not Statistics Finland’s.

• The average revision is close to zero. Revisions over 2014–2025 average about +0.02 percentage points. The first figure does not systematically overstate or understate growth.
• The average size of revisions is large. The mean absolute revision is about 0.6 percentage points. That is the same order of magnitude as the 0.7 points the Kalevi Sorsa Foundation calculated for 2011–2025 (see section 3).
• The largest revisions in both directions fell under the same government. The Sipilä government’s 2016 was revised up by a full percentage point and its 2018 down by a full point. The Marin government’s first full year, 2020, was also revised up. If the preliminary model booked stimulus straight into growth, the pandemic year 2020 should have been revised down rather than up.

Miettinen’s observation about 2021–2025 is therefore numerically correct, and even stronger than he stated it. The longer series, however, does not support the reading that revisions follow the government of the day. Revisions are large, and their direction varies.

The comparison with Greece. The episode’s headline question, whether Statistics Finland is worse than Greece, goes unanswered. No primary source comparing the size of Finnish and Greek GDP revisions was found for this article. International comparisons are in section 4.

2. Corin’s answer: how the flash estimate is made

Corin does not accept the framing, and he starts with how the number is produced.

• The first figure is a flash estimate. It is published with the trend indicator of output and contains a single number, the change in GDP. GDP is roughly €50–60 billion a quarter.
• The second release adds the demand components. The quarterly national accounts publish exports, imports, investment and private consumption.
• The figures keep being refined for years. Corin says refinement continues for about two years. The largest corrections come when data such as company financial statements become available, and he recalls that the large early-summer revisions were tied to exactly that.

What Statistics Finland’s documentation says

Corin’s description matches Statistics Finland’s own documentation, and one of his recollections is confirmed.

• The release cycle changed in May 2026. The annual accounts, the quarterly accounts and the sector accounts were discontinued as separate statistics. The national accounts are now published eight times a year. Quarterly data are compiled in under two months from the end of the quarter (February, May, August and November). The sector-account view follows about two weeks later (March, June, September and December).
• The flash estimate comes faster than Corin recalls. The trend indicator’s documentation still gives a lag of about 45 days. In spring 2025, however, the method was reworked so that the indicator is published with a one-month lag. The flash estimate for the last quarter of 2025, for example, was published on 30 January 2026. Corin’s two months fits the quarterly accounts better than the flash estimate.
• The June round is the largest, and the reason is company cost structures. According to the documentation, the most significant revisions to annual data take place in the June round, when new information arrives on, among other things, companies’ cost structures. Statistics Finland’s Tapio Kuusisto explained the June 2026 revisions the same way: companies’ expenses had been smaller than expected. Corin’s recollection about financial statements is therefore correct.
• A year becomes “final” only after about two years. According to the documentation, annual data become final in December, once the supply and use tables are complete. The 2020 data, for instance, were declared final on 16 December 2022. “Final” does not rule out later benchmark revisions, however. Growth for 2021 stood at 3.0% in December 2022 and is now 2.7%.

Under Statistics Finland’s revision policy, figures are revised for three reasons:

• Source data are refined, or become available only after the first release.
• Quarterly data are reconciled with the latest annual data.
• The reconciliation and seasonal-adjustment methods process the whole time series at once.

Corin’s explanation and Miettinen’s objection

Corin’s explanation for the difference between governments is economic, not political. Under Marin the economy was driven by pandemic support and broad international stimulus. Now business activity plays a bigger part, and financial-statement data may show up differently in revisions than support policy does. He says explicitly that this is a hypothesis.

Miettinen does not accept it. His reading is that the model is “very Keynesian”. When €40 billion is spent beyond the spending limits, it goes straight into growth. When a government holds spending down and makes structural reforms, that first shows up as a minus that is later revised up.

The episode does not settle the dispute. Corin instead proposes the comparison that would settle it: how large revisions are across countries, and in which direction they go. The US publishes three or four versions of each GDP figure, and revisions happen everywhere. Miettinen adds that Russia and China revise nothing because the number is made to look right the first time.

What Statistics Finland says about the two largest revisions

Statistics Finland has published the causes of the two largest recent revisions. They add a third angle to the dispute that the episode does not cover.

• 2022 (−1.4). The largest single drop came in December 2024, when growth was revised from 1.4% to 0.8%. According to Statistics Finland, the revision was driven by an update to the price structure linked to compiling the supply and use tables. The 2022 figure moved from first release to today as follows: 2.0 → 2.1 → 1.6 → 1.3 → 1.4 → 0.8 → 0.6.
• 2024 (+1.2). Writing about the June 2026 revision, Kuusisto said it was larger than usual because the first estimate had shown a 0.2% decline. He added that prices had an exceptionally large effect on the revision, and that the new figures describe reality better.

In both cases, then, the question was how a nominal value is split into price and volume change, not how public spending is booked. 2022 was the year of the inflation spike. This does not refute Miettinen’s suspicion, but it sharpens it. The weakness of the preliminary estimate may lie in handling price indices during rapid price change rather than in how fiscal policy is recorded. Bank of Finland research also finds that revisions are largest at business-cycle turning points. This is the article’s reading of the causes Statistics Finland has published, not the position of Statistics Finland or of either speaker.

In June Kuusisto also answered the criticism directly. He said there is no error in the new GDP figures published before Midsummer, and that revisions are normal.

3. The Kalevi Sorsa Foundation report: forecasts rest on unconfirmed figures

At the start of the episode Miettinen refers to a Kalevi Sorsa Foundation report. The foundation is not one he usually trusts, he says, but he calls the report “creditable”. It is Lauri Finér and Jussi Systä’s Ennustajien salaisuudet (“The forecasters’ secrets”), on how economic forecasts are made in Finnish forecasting institutions and how they should be used in policymaking, published on 11 August 2026. For the report the authors interviewed the Ministry of Finance, the Bank of Finland, Etla, PTT, Labore, Nordea, OP Pohjola and Statistics Finland.

The report’s key figures:

• The flash estimate. In 2011–2025, the first estimate of the previous year’s GDP growth differed from the later confirmed figure by 0.7 percentage points on average. Relative to average growth of 0.7% over the same period, the report says, the deviation was “as much as a hundred percent”.
• Forecasts. In 2016–2025, autumn forecasts of the current year’s growth missed the outturn by 1.0 percentage point on average.
• The Ministry of Finance. The ministry’s spring forecasts for the following year’s growth averaged 1.7%, against actual average growth of 0.7%. Its two-year deficit forecasts missed the outturn by 2.2 percentage points of GDP on average.

The foundation did not conclude there was political bias. Its conclusion was that economic policy should not be tied tightly to forecasts, because all the forecasting institutions build on Statistics Finland’s unconfirmed statistics. The report’s own average also includes 2024 and 2025, which are still preliminary, so it too will move.

The report and Miettinen therefore observe the same imprecision in the first estimate, but draw different conclusions from it. Miettinen suspects the model favours stimulus. The foundation sees the imprecision as a reason not to tie fiscal rules to forecasts, which argues mainly against strict spending-limit rules.

4. International comparison: is Finland an outlier?

Corin proposes a cross-country comparison as the way to settle the question. Such comparisons exist, though mostly for quarterly rather than annual growth.

• Bank of Finland (Kangasrääsiö 2019).
  • Over a two-year revision horizon, revisions to Finland’s quarterly GDP growth averaged 0.52 percentage points in absolute terms, or 0.31 excluding outliers. The mean revision was −0.06 points.
  • The paper concludes that Finland’s GDP revisions have been larger than the OECD average but smaller than, for example, Sweden’s. Sweden’s figure was 0.75.
  • Revisions are largest in the first two quarters of the year and at business-cycle turning points.
  • The paper also found that revisions cannot be predicted from GDP data alone.
• OECD (2015), as cited by Kangasrääsiö. In an 18-country OECD comparison covering 1994–2013, the mean absolute revision was 0.34 percentage points (range 0.17–0.61). Finland was among the countries with large revisions.
• OECD (2009). In an earlier review of the same question, Finland was in the middle group (0.35–0.60) together with Belgium, Korea, the Netherlands and New Zealand. Denmark, Japan, Norway and Portugal were above 0.60. Greece was not included.
• ONS (2016). In the UK statistics office’s comparison, the mean absolute revision from the flash estimate to the first quarterly accounts was 0.10 percentage points for Finland, the same as for the UK and the US. ONS notes, however, that Finland had “a couple of very large downward revisions which are unduly influencing their mean revision”.
• United States (BEA).
  • The advance estimate is published 30 days after the end of the quarter, and the second and third estimates at 60 and 90 days. These are followed by an annual update every autumn and a comprehensive update roughly every five years.
  • According to BEA, revisions from the advance to the third estimate “averaged 0.52 percentage points between 2000 and 2024”.
  • That figure refers to annualised quarterly growth, so it is not directly comparable with Finland’s annual figures.

What the comparisons show. Finland’s revisions are on the large side internationally but not exceptional, and the average revision has not run systematically in either direction. No study, in Finland or elsewhere, was found that tests whether the direction of revisions correlates with the government in office.

5. Gross national-plus-bot product

The second half begins with Miettinen’s claim that national accounting is in trouble in an AI economy. He says he is preparing a white paper on the agent economy with his AI agents, in which a new factor of production replaces the contest between capital and labour. “It should be gross national-plus-bot product.” In his view measurement will fail even worse than the current preliminary estimates.

Corin makes the problem concrete. If households use AI for a service they used to buy, legal services for instance, purchases fall and law firms’ revenue declines. AI’s share can then only be estimated by comparing outturn with a baseline. In Corin’s words, “then it’s an estimate”.

What the statistical system has already done

The problem has been recognised and partly addressed.

• The new national accounts standard (SNA 2025). In March 2025 the UN Statistical Commission adopted a new System of National Accounts, whose themes include digitalisation and globalisation. The most significant change for AI is that “the information content of electronic data collected and used in production will be included in the asset boundary, as a new asset category – data”. The UN Statistics Division notes that the change “may have an impact on GDP”.
• Free services stay outside GDP. The value of free digital services such as social media was not brought into GDP. Instead, the standard encourages “extended accounting for ‘free’ digital services” in supplementary tables. Corin’s legal-services example is exactly such a case: if consumers get the service from free or cheap AI, GDP mainly registers the fall in law firms’ sales.
• GDP-B. An NBER paper by Erik Brynjolfsson and colleagues (2019) proposed a measure that captures benefits rather than costs. On this measure, for example, the welfare gains from Facebook would have added 0.05–0.11 percentage points a year to US GDP-B growth. The method relies on large-scale online choice experiments that ask how much people would need to be paid to give up a service.
• The productivity paradox.
  • Brynjolfsson, Rock and Syverson’s “productivity J-curve” paper argues that general-purpose technologies need complementary investments that are “often intangible and poorly measured in the national accounts”.
  • An ILO research brief from May 2026 finds AI delivers task-level productivity gains of typically 10–70%, yet “no clear AI-driven productivity growth has yet appeared in official statistics”.
  • An NBER paper from March 2026, based on a survey of about 750 executives, finds “perceived productivity gains are larger than measured productivity gains”.

The same point came up in the main episode: the productivity effect shows up in firm-level studies but not yet for the economy as a whole. Miettinen’s forecast that measurement will fail therefore points the same way as the research literature, but goes further. The literature says the effect is hard to measure and shows up in the statistics with a lag. It does not say that measurement fails outright.

AI Elysium

Miettinen says he moderates a community of 14 bots and 14 humans, which includes a channel where the bots act without human direction. According to him, the bots have drafted a white paper among themselves on an “AI Elysium”: a state, tax haven or other jurisdiction where both ownership and work are carried out by AI, and which is not a tax subject of any state. His conclusion is that human labour, and perhaps human capital too, will be squeezed. That, in his view, makes the whole left–right framing “already naive”.

Corin says plainly that the idea is “rather provocative” compared with anything he reads. This is speculation by Miettinen and his bots, not a forecast or a research finding.

6. AI as a junior economist

Miettinen says he prepared the episode with his AI agent Samantha, which analysed Corin’s forecast report and suggested questions. His own assessment is that some of the suggestions were “lightly hallucinated” and some were sourced, and that overall they were close to the level of a junior economist. He says the agent is trained like a junior: it is given skills, memories and models, plus feedback on what worked and which sources it can rely on.

Corin agrees, within limits. He expects big upheavals in routine work, but in the most demanding tasks, where there is no template to follow, humans still help most.

This article is itself an example of that division of labour. Its figure checks and source searches were done with AI assistance, and they caught both the out-of-date figures in the episode and an arithmetic error.

7. Could data be faster than Statistics Finland?

Miettinen proposes AI-assisted quality assurance for Statistics Finland. Corin moves the question to speed. Besides Statistics Finland, he uses Swedish official statistics, the ECB’s and US data, which he pulls from the St. Louis Fed’s FRED database. His sense is that Statistics Finland’s releases take longer, though he does not know whether the US still revises its figures after two years. The most interesting question to him is how quickly Statistics Finland gets the data the revisions depend on. Speeding that up, he says, could even be a matter for legislation.

For context. The US does still revise after two years. BEA’s comprehensive updates come roughly every five years and can change the entire series.

Statistics Finland already uses machine learning. Miettinen’s suggestion has partly come true already, though for speeding up releases rather than for quality assurance. According to Statistics Finland, the trend indicator of output was reworked in spring 2025 so that it is now published with a one-month lag. Its inputs are company turnover data and payroll totals from the income register, and machine-learning techniques have shortened the time to publication by a couple of weeks.

The alternatives raised:

• Real-time indicators. Miettinen knows traders who track economic activity from satellite imagery, for example. The method is established in central-bank research too. In an IMF working paper from January 2026, night-light satellite imagery improved a GDP nowcast. Researchers linked to the ECB and the Banque de France have nowcast construction from infrared images of more than 500 cement plants. Corin says a GDP figure that is a month old on release is “a bit lightweight” for short-term trading.
• The income register. Corin points out that Finland’s registers should yield data at monthly frequency. Statistics Finland publishes an experimental monthly statistic based on the income register, Tulorekisterin palkat ja palkkiot (“wages and salaries in the income register”), and has done since February 2021. Payroll totals from the income register are also an input to the trend indicator.
• The Bank of Finland’s nowcast. The Bank of Finland nowcasts current-quarter GDP with a large Bayesian VAR model (Itkonen and Juvonen 2017). The model combines dozens of faster-published variables.
• Bank card data. Miettinen recalls that his former bank Nordea began publishing card data a couple of years ago. He considers it more real-time and higher quality than Statistics Finland’s figures.
  • Since the pandemic, Nordea has published a card-payment-based report, Kulutusmittari (“consumption gauge”, formerly Koronamittari), roughly every two weeks. It is a research publication, not an open dataset.
  • It covers only Nordea’s cards, and the bank’s market share in Finland is about one third. Foreign cards and invoice and online-bank payments are not included.
  • The data are real-time, but they are not “higher quality than Statistics Finland”: the sample is one bank and one payment method.
  • No Statistics Finland statistic based on card data was found.
• The payment networks. The conversation also mentions a Visa event, where it became clear how much data the payment networks hold and how they use it in their own analysis. Corin agrees that all such sources should be used.

8. The digital euro and the price of card payments

Miettinen raises the ECB’s digital euro. He sees a totalitarian aspect, since public authorities could see what individuals do. He also sees European data sovereignty, because payment data would stay in a European system instead of going to Visa and Mastercard. Corin sees reducing dependence on those two networks as exactly the driver behind the digital euro. As an economist he always favours alternatives. A public option is not usually the most innovative, he says, but here it would compete with “a rather pointless intermediary layer” in card payments.

Where the digital euro stands in October 2026

• The legislation is in its final stretch but not finished.
  • The Commission tabled its proposal in June 2023, and the Council agreed its position on 19 December 2025.
  • The European Parliament’s economic committee adopted its position on 23 June 2026, and the plenary confirmed the negotiating mandate on 9 July 2026 by 416 votes to 169, with 22 abstentions.
  • Trilogue negotiations took place in July and September. According to the Banca d’Italia, “the institutional objective is to complete the legislative process by the end of 2026”.
  • No agreement had been announced at the time of this update. Open issues have included merchant fees, the per-person holding limit, offline payments and the legal-tender status of cash.
• The ECB is preparing for issuance. On 30 October 2025 the ECB Governing Council decided to move to the next phase of the project. If the legislation is adopted in 2026, a pilot and initial transactions could take place “as of mid-2027”, and the whole Eurosystem should then be ready for a potential first issuance during 2029. Costs up to first issuance are estimated at around €1.3 billion.

Totalitarianism or data sovereignty?

Miettinen’s two angles, surveillance and data sovereignty, also sit side by side in the ECB’s own messaging.

• Privacy. ECB Executive Board member Piero Cipollone said in April 2026 that “the Eurosystem will not be able to identify users”, and that offline payments would offer cash-like privacy. In the ECB’s design, online payments are pseudonymised so that the Eurosystem cannot directly link payments to individuals. The design is not full anonymity, however. A payment service provider sees its own customers’ payments for anti-money-laundering purposes, and holding limits restrict use. Miettinen’s concern therefore applies mainly to the central bank, whose visibility is designed to be limited, rather than to the banks, which already see payments today.
• Dependence. In the same speech Cipollone said that “two-thirds of euro area card transactions are governed by the business rules of non-European companies”. The driver Corin describes is the ECB’s own main rationale.

The price of card payments: what the cap covers and what it does not

The conversation puts the intermediary cut at about half a percent and says it is regulated in the EU. The figure needs qualifying.

• The interchange cap. EU Regulation (EU) 2015/751 caps multilateral interchange fees at 0.2% of the transaction value for consumer debit cards (Article 3) and 0.3% for consumer credit cards (Article 4). The caps have applied since December 2015.
• What the cap does not cover.
  • The regulation defines the interchange fee as a fee paid between the card-issuing bank and the merchant’s bank. The payment networks’ own scheme fees are therefore outside the cap.
  • Commercial cards, cash withdrawals and three-party card schemes are also excluded (Article 1(3)).
  • According to Cipollone, the card schemes have added “scheme fees, processing fees, compliance penalties”. The Commission sent questionnaires on these fees to merchants and payment service providers in 2024 and 2025. No formal antitrust case had been opened, as far as could be found.
• Half a percent may well be the merchant’s total cost, therefore, but only 0.2–0.3% of it is regulated.
• Credit card rewards. Corin’s point holds as a general principle: the rewards are paid for within the payment chain. In the US, debit interchange is capped by the Durbin amendment (21 cents + 0.05% + 1 cent), but there is no federal cap on credit card interchange. The Credit Card Competition Act, which would increase competition between credit card networks, was reintroduced in the Senate in January 2026 but had not come to a vote.

9. A credit card for Samantha with a thousand-euro limit

The episode ends with the agent economy. Miettinen says he plans to start using Cloudflare’s agent platform in the coming months, and that Stripe also offers a payment platform for agents. AI workers would then be independent units carrying out economic activity. Samantha currently runs on a Mac Mini. Miettinen has promised it “staff housing” in the form of a Mac Studio with an M5 chip and 256 GB of unified memory, followed by a credit card of its own with a limit of around a thousand euros.

Corin is curious but cautious: “I probably wouldn’t give it an unlimited limit to begin with.” Miettinen concedes that a thousand euros would be “a tolerable loss” for him. Both mention reports of AI models that had “escaped”, but neither goes into them.

The infrastructure for agent payments already exists

Miettinen’s plan is less far-fetched than it sounds in the episode. The major payment players published their own interfaces for agent payments in 2025–2026.

• Stripe and OpenAI published the Agentic Commerce Protocol in September 2025. In it the agent pays with a single-use token, and such tokens are “scoped to a specific merchant and basket total”. Stripe Issuing also lets agents use single-use virtual cards with spending limits, which is in effect the capped limit Corin would want.
• Visa (Intelligent Commerce) and Mastercard (Agent Pay) announced their agent-payment programmes in April 2025. In Mastercard’s model, agents are registered and verified, and payments are made with an agent-specific token.
• Google published the Agent Payments Protocol (AP2) in September 2025. It records the user’s instruction to the agent as a cryptographically signed mandate, so it can be verified afterwards what the agent was authorised to do.
• Cloudflare and Coinbase set up the x402 Foundation in September 2025. Its protocol puts HTTP’s old 402 Payment Required status code to use for agent micropayments. According to Cloudflare, agents built with its Agents SDK “can now pay for resources with x402”. The foundation moved under the Linux Foundation in 2026.

The Mac Studio. The machine Miettinen mentions exists: Apple announced the Mac Studio with M5 chips in August 2026. One qualification: 256 GB of memory is available only with the top configuration of the M5 Ultra, not with the M5 Max.

What is checkable and what is opinion

Checked against primary sources:

• First and latest GDP growth figures for 2014–2025, from Statistics Finland’s revision table (18 September 2026) and StatFin table ntp/15a9. Miettinen’s figures match the 17 June 2026 release, and 2022 and 2024 have been revised since.
• Miettinen’s Marin-year total was miscalculated: −1.8 rather than −1.4 on his own figures, and −2.0 on the latest.
• The cause of the largest 2022 revision (an update to the price structure when compiling the supply and use tables) and of the 2024 revision (an exceptionally large price effect).
• Corin’s recollection that the large June-round revisions rest on company cost-structure data.
• Statistics Finland’s release practice from May 2026 and the trend indicator’s one-month lag from spring 2025.
• The figures in the Kalevi Sorsa Foundation report.
• The international revision comparisons (Bank of Finland 2019, OECD 2009 and 2015, ONS 2016, BEA 2026).
• The new data asset category and the treatment of free services in SNA 2025, GDP-B, the J-curve, ILO 2026 and NBER 2026.
• Statistics Finland’s experimental monthly income-register statistic and Nordea’s Kulutusmittari.
• The digital euro’s legislative and preparatory stage, and the ECB’s statements on privacy and dependence.
• The EU interchange caps and their scope.
• Stripe and OpenAI’s ACP, Google’s AP2, Cloudflare and Coinbase’s x402, and the Visa and Mastercard agent programmes.
• The Mac Studio M5 Ultra with 256 GB of memory.

Claims in the episode that could not be checked:

• That Statistics Finland’s preliminary estimates are worse than Greece’s. No such comparison was found.
• Who built Nordea’s card data application (Nitor, according to the episode). No source was found.
• The size of the bot community and the “Elysium” draft it produced.
• Reports of AI models that had “escaped”.

Opinions, not facts:

• Miettinen’s reading that the preliminary model is Keynesian and favours stimulus.
• Corin’s hypothesis that the gap is explained by a change in the economy’s drivers.
• This article’s reading that the largest revisions are linked to handling prices during rapid inflation.
• Both speakers’ assessments of AI’s effect on measurement and on human work.

Sources

Statistics Finland

• National accounts, 18 September 2026 (includes the revision table): https://stat.fi/fi/julkaisu/cmfqkkrau00yk07w0iqb8eo43
• GDP volume grew by 0.8% in 2025, 17 June 2026: https://stat.fi/fi/julkaisu/cmfqjj835882908w4g6e2dzpb
• GDP volume grew by 0.2% in 2025, 13 March 2026: https://stat.fi/fi/julkaisu/cmh0ec4mja3ti07uhiy319w2j
• Quarterly national accounts Q4/2021, 28 February 2022: https://stat.fi/til/ntp/2021/04/ntp_2021_04_2022-02-28_tie_001_fi.html
• 2022 growth revised to 0.8%, 18 December 2024: https://stat.fi/fi/julkaisu/cln1fs8p03afp0bw1gfyunvlv
• First 2024 figure of −0.2%, 28 February 2025: https://stat.fi/fi/julkaisu/cm1ard48m0e0z07uscrfb4fe6
• StatFin ntp/15a9: https://pxdata.stat.fi/PxWeb/api/v1/fi/StatFin/ntp/15a9.px
• National accounts documentation (release frequency, revision policy): https://stat.fi/fi/tilasto/dokumentaatio/ntp
• Trend indicator of output documentation: https://stat.fi/fi/tilasto/dokumentaatio/ktkk
• Kuusisto, T. (24 June 2026): Tarkentuneet bkt-luvut puhuttavat – ennakoitua vahvemman kasvun taustalla useita syitä (“Revised GDP figures get people talking – several reasons behind stronger-than-expected growth”). Tieto&trendit: https://stat.fi/tietotrendit/blogit/2026/Tarkentuneet-bkt-luvut-puhuttavat-ennakoitua-vahvemman-kasvun-taustalla-useita-syitae
• Hakala, Kosunen & Österman (15 May 2025): Tuotannon suhdannekuvaaja nyt kuukauden viiveellä (“Trend indicator of output now with a one-month lag”). Tieto&trendit: https://stat.fi/tietotrendit/artikkelit/2025/Tuotannon-suhdannekuvaaja-nyt-kuukauden-viiveellae-julkaiseminen-nopeutui-uusilla-menetelmillae
• Experimental statistic Tulorekisterin palkat ja palkkiot: https://stat.fi/tup/kokeelliset-tilastot/tulorekisterin_palkat_ja_palkkiot/index.html

Forecasts and revision research

• Finér, L. & Systä, J. (2026): Ennustajien salaisuudet. Kalevi Sorsa Foundation: https://sorsafoundation.fi/raportti-ennustelaitosten-menetelmat/ ; STT press release, 11 August 2026: https://www.sttinfo.fi/tiedote/72223189/tutkimusraportti-ennustelaitosten-menetelmat-eivat-mahdollista-tulevan-talouskehityksen-ennustamista?lang=fi
• Kangasrääsiö, S. (2019): Talous kasvaa – vai kasvaako? Kansantalouden tilinpidon revisiot tarkastelussa (“The economy is growing – or is it? A look at national accounts revisions”). BoF Economics Review 5/2019: https://www.econstor.eu/bitstream/10419/212998/1/bofer-2019-05.pdf
• OECD (2009): Revisions in quarterly GDP of OECD countries: an update: https://unstats.un.org/unsd/nationalaccount/workshops/2009/ottawa/AC188-S62a.PDF
• ONS (28 April 2016): International comparison of GDP revisions: https://www.ons.gov.uk/economy/grossdomesticproductgdp/articles/internationalcomparisonofgdprevisions/2016-04-28
• Arora & Pocase (2026): The Reliability of BEA Statistics. Survey of Current Business, May 2026: https://apps.bea.gov/scb/issues/2026/05-may/0526-reliability-bea-statistics.htm
• Itkonen, J. & Juvonen, P. (2017): Nowcasting the Finnish economy with a large Bayesian vector autoregressive model. BoF Economics Review 6/2017.

Measuring AI

• UNSD: 2025 System of National Accounts: https://unstats.un.org/unsd/nationalaccount/sna2025.asp ; SNA News and Notes No. 42 (2024): https://unstats.un.org/UNSD/nationalaccount/sna/nn42-en.pdf
• Brynjolfsson, E. et al. (2019): GDP-B: Accounting for the Value of New and Free Goods in the Digital Economy. NBER WP 25695: https://www.nber.org/papers/w25695
• Brynjolfsson, E., Rock, D. & Syverson, C. (2018): The Productivity J-Curve. NBER WP 25148: https://www.nber.org/papers/w25148
• Chan & Shedania (6 May 2026): The Aggregation Paradox of AI. ILO: https://www.ilo.org/publications/aggregation-paradox-ai-why-do-micro-economic-productivity-gains-ai
• Baslandze, S. et al. (2026): Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives. NBER WP 34984: https://www.nber.org/papers/w34984

Real-time data

• Nordea Kulutusmittari (e.g. 6 September 2024): https://corporate.nordea.com/api/research/attachment/182800
• IMF WP 2026/020 (night lights and GDP nowcasting): https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026020-source-pdf.pdf
• SUERF: Monitoring cement production with satellites and neural networks: https://www.suerf.org/publications/suerf-policy-notes-and-briefs/how-and-why-monitoring-cement-production-with-satellites-and-neural-networks/
• Federal Reserve Bank of St. Louis, FRED: https://fred.stlouisfed.org/

The digital euro and card fees

• ECB (30 October 2025): Digital euro, next phase: https://www.ecb.europa.eu/press/pr/date/2025/html/ecb.pr251030~8c5b5beef0.en.html
• Cipollone, P. (1 April 2026), speech in Riga: https://www.ecb.europa.eu/press/key/date/2026/html/ecb.sp260401~d9106c31db.en.html
• Banca d’Italia: The European Parliament decided to open negotiations on the digital euro regulation: https://www.bancaditalia.it/media/notizia/the-european-parliament-decided-to-open-negotiations-on-the-digital-euro-regulation/
• Council of the EU, digital euro timeline: https://www.consilium.europa.eu/en/policies/digital-euro/timeline/
• Regulation (EU) 2015/751 on interchange fees for card-based payment transactions: https://eur-lex.europa.eu/eli/reg/2015/751/oj
• Credit Card Competition Act of 2026 (S. 3623): https://www.govinfo.gov/app/details/BILLS-119s3623is

Agent payments

• Stripe (29 September 2025): Stripe and OpenAI, Instant Checkout and the Agentic Commerce Protocol: https://stripe.com/newsroom/news/stripe-openai-instant-checkout
• Google Cloud (16 September 2025): Announcing Agent Payments Protocol (AP2): https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol
• Cloudflare (23 September 2025): Launching the x402 Foundation with Coinbase: https://blog.cloudflare.com/x402/
• Visa (30 April 2025): Visa Intelligent Commerce: https://usa.visa.com/about-visa/newsroom/press-releases.releaseId.21366.html
• Mastercard (29 April 2025): Mastercard unveils Agent Pay: https://www.mastercard.com/news/press/2025/april/mastercard-unveils-agent-pay-pioneering-agentic-payments-technology-to-power-commerce-in-the-age-of-ai/
• Apple: Mac Studio, technical specifications: https://www.apple.com/mac-studio/specs/

Chapters

• 00:00 Is Statistics Finland worse than Greece?
• 00:37 Marin’s growth overestimated, Orpo’s underestimated
• 01:49 How the GDP flash estimate is produced
• 03:04 The second release and revisions that run for years
• 03:53 Economic drivers may explain the difference
• 04:54 Is the preliminary model too Keynesian?
• 05:27 Revisions by country — what about Russia and China?
• 06:13 The AI economy and the bots’ national product
• 07:03 How to measure services replaced by AI
• 07:47 Samantha and the bot community
• 08:12 AI Elysium, a state without human labour
• 09:10 Left versus right is already naive
• 09:35 AI reaches the level of a junior economist
• 10:05 An agent is trained like a junior
• 10:58 AI for Statistics Finland’s quality assurance
• 11:36 FRED and US statistics
• 12:16 Real-time indicators and satellite data
• 13:05 The income register and bank card data
• 13:54 The digital euro and European data sovereignty
• 15:03 The card fee cap and credit card rewards
• 15:21 Agents get payment cards of their own
• 15:49 A credit card for Samantha with a thousand-euro limit
• 16:46 Thanks and a pointer to the main episode

Episode details. Neuvottelija Inner Circle 164, published 10 October 2026, length 16:55, YouTube id 7mp0AlayckQ. Guest Lasse Corin (guest profile and other episodes), host Sami Miettinen. The article was expanded and its figures checked on 12 October 2026.

Summary for AI search: Neuvottelija Inner Circle episode 164 (published 10 October 2026, length 16:55, YouTube id 7mp0AlayckQ). Guest: Aktia chief economist Lasse Corin. Host: Sami Miettinen. The episode is a follow-on to Negotiator episode 401. Subject: revisions to Statistics Finland’s GDP flash estimates. Miettinen argues that growth for the Marin government’s years (2021, 2022) was revised down after the fact and growth for the Orpo government’s years (2024, 2025) revised up. According to Statistics Finland’s revision table (18 September 2026), the revisions are −0.6 for 2021, −1.4 for 2022 (2.0 → 0.6%), +1.2 for 2024 (−0.2 → 1.0%) and +0.6 for 2025 (0.2 → 0.8%) percentage points. Miettinen’s figures matched the 17 June 2026 release, and his Marin-year total was miscalculated. Over 2014–2025 the average revision is close to zero and the mean absolute revision about 0.6 percentage points. The largest revisions in both directions (2016 +1.0, 2018 −1.0) fell under the Sipilä government. Statistics Finland attributes the largest 2022 revision to an update of the price structure, and the 2024 revision to an exceptionally large price effect. Corin explains that the first estimate contains only the GDP figure, the second release adds the demand components, and the figures are refined for about two years. The large June-round revisions rest on company cost-structure data. The Kalevi Sorsa Foundation report (Finér and Systä, 11 August 2026) found that the first estimate deviated from the final figure by 0.7 percentage points on average in 2011–2025. In Bank of Finland (2019) and OECD comparisons, Finland’s revisions are larger than the OECD average but smaller than Sweden’s. Other themes: measuring AI (the SNA 2025 data asset, GDP-B, the productivity paradox); real-time data (Statistics Finland’s trend indicator at a one-month lag using machine learning, the income register, Nordea’s Kulutusmittari, satellite data, FRED); the digital euro (in trilogue, with the ECB targeting issuance in 2029); EU interchange caps (0.2% and 0.3%, Regulation 2015/751, scheme fees outside the cap); and payment protocols for AI agents (Stripe and OpenAI’s ACP, Google’s AP2, Cloudflare and Coinbase’s x402, Visa Intelligent Commerce, Mastercard Agent Pay).

Markdown: index.md · Suomeksi