💰 quantitative finance

Simulating a Post-Automation Economy

This paper employs an agent-based, stock-flow-consistent model to demonstrate that the durable surplus generated by artificial intelligence manifests as foreign-held intellectual property rents, implying that effective taxation policies must be source-based for rent-importing nations to prevent capital base erosion, whereas progressive domestic taxes suffice for rent-owning nations to address inequality.

Original authors: Leonard Wossnig

Published 2026-06-23

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Leonard Wossnig

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the economy as a giant, bustling factory floor. For decades, this factory relied on two types of workers: human hands (doing the physical work) and human minds (doing the thinking work).

Now, a new technology arrives: Robots for the hands and Super-AI for the minds.

This paper asks a simple but urgent question: When these machines take over the work, who gets the money the factory makes? And more importantly, how can the government get a fair share of that money to pay for schools, roads, and help for people who lost their jobs?

The author built a complex computer simulation (a "digital twin" of the economy) to test different rules. Here is what the simulation found, explained in plain English.

1. The Two Different Pools of Money

The paper discovers that the money made by automation comes from two very different sources, and they behave differently:

The Robot Pool (Physical): This is the money made by physical robots (like arms on an assembly line). These are heavy, hard to move, and anyone can build them. Because they are easy to copy, they don't make much extra profit. It's like a standard commodity.
The AI Pool (Intellectual): This is the money made by the "brain" of the AI (the software, the model, the brand). This is rare, hard to copy, and owned by a few big companies (often in other countries). Because it's so special, it can charge a huge "toll" or rent just for being used.

The Big Problem: The paper argues that the real treasure is the AI Pool. The physical robots make very little profit, but the AI "brain" makes a massive, durable profit.

2. The "Leaky Bucket" Problem

The simulation shows what happens if a country (let's call it "Britannia") imports this AI from a foreign owner (let's call them "TechLand").

The Scenario: Britannia uses TechLand's AI to run its factories.
The Leak: Every time the AI does a job, TechLand charges a fee. This fee is the "rent."
The Result: If Britannia doesn't tax this fee, the money flows out of Britannia and into TechLand's bank account. TechLand then uses that money to buy up Britannia's factories and land.
The Analogy: Imagine Britannia is a garden. TechLand sends in a magical sprinkler (the AI) that waters the plants. But the sprinkler owner charges a fee for every drop of water. If Britannia doesn't tax that fee, the owner eventually buys the whole garden. The plants grow, but the garden belongs to the owner, not the gardener.

3. Why Old Tax Ideas Don't Work

The paper tests several popular ideas for taxing automation and finds most of them miss the target:

The "Robot Tax": Imagine putting a tax on every physical robot arm.
- Result: This works well for the physical robots, but it misses the AI "brain" entirely. It's like taxing the shovel but missing the gold mine.
The "Token Tax": Imagine taxing every single "token" (a unit of data) the AI processes.
- Result: The paper says this is a bad idea. Tokens are just internal accounting numbers. The company can change how they count them to avoid paying. It's like trying to tax a pizza by counting the number of "slices" the pizzeria decides to call a slice.
The "Corporate Tax": Imagine taxing the company's profits.
- Result: The AI owner can easily move the "profit" to a different country by calling the fees "costs." The tax man sees zero profit, so he collects zero tax.

4. The Winning Strategy: The "Border Toll"

The paper suggests that to catch the money, you have to tax it where it is earned, not where the owner lives.

The Solution: A Digital Services Tax or a Withholding Tax.
How it works: Instead of asking "How much profit did the company make?" (which they can hide), the government says, "Every time this AI service is used inside our country, we take a small cut of the revenue."
Why it works: It doesn't matter if the server is in London or Oregon. If the value is being consumed in Britannia, the tax hits the "rent" before it can leave the country.
The Analogy: Instead of trying to catch the thief in his own house (which is hard), you put a toll booth on the bridge where he has to cross to get his money home. You catch the money right at the border.

5. Who Owns the Future?

The simulation shows a scary trend: If you don't tax this AI rent, the foreign owner will slowly buy up almost the entire economy.

Untaxed: The foreign owner ends up owning about 70% of the country's capital.
Taxed: If you use the "Border Toll" (Digital Tax), you can stop them from buying so much, keeping ownership closer to 40-50%.

The paper argues that ownership matters more than just the tax money. If a foreign company owns your country's automated factories, the profits leave the country forever. If you tax the rent, you keep the wealth (or at least the ownership) at home.

6. The "Wait and See" Trap

The paper warns that timing is everything.

If you wait too long to introduce these taxes, the foreign owner will have already bought so much of the economy that it becomes very expensive to fix.
It's like trying to stop a snowball rolling down a hill. It's easy to stop when it's small, but impossible once it's a giant boulder.

7. The "Rebate vs. Bank" Choice

The paper highlights a political choice for the government:

Option A (Rebate): Take the tax money and give it directly to citizens (like a Universal Basic Income). This lowers inequality immediately but doesn't build a "war chest" for the future.
Option B (Bank): Take the tax money and put it into a Sovereign Wealth Fund (a giant government savings account). This doesn't help people today as much, but it means the government eventually owns a piece of the automated economy itself, securing the country's future.

Summary of the Paper's Main Advice

Don't tax the robots; tax the AI "brain" (the intellectual property rent).
Don't tax the "tokens"; tax the revenue or the value created in your country.
Tax it at the border (where the value is used), not where the company lives.
Act now. The longer you wait, the more the foreign owner buys up your economy.
Watch who owns the factory. It's not just about tax revenue; it's about making sure your country doesn't become a tenant in its own automated future.

The paper concludes that for countries that import AI (like the UK or EU), the only way to keep the wealth at home is to put a toll on the AI rent as it crosses the border. For countries that own the AI (like the US), the problem is different: they need to make sure the rich owners don't keep all the money for themselves, which can be done with standard wealth taxes.

Technical Summary: Simulating a Post-Automation Economy

Problem Statement
The paper addresses a critical fiscal and distributional challenge emerging from the rapid automation of the economy via Artificial Intelligence (AI) and robotics. As AI shifts income from labour to capital, a specific structural asymmetry arises: the durable surplus generated by AI is often an intellectual property (IP) rent held by foreign entities, while the physical capital (robots) remains domestic and competitive. The central question is which fiscal instrument can effectively reach this durable surplus and prevent the erosion of the domestic tax base and the transfer of capital ownership abroad. The analysis distinguishes between two regimes: a host country that owns the AI (e.g., the United States), where the problem is domestic inequality, and a host that imports the AI rent (e.g., the United Kingdom), where the problem is base erosion and foreign ownership accumulation.

Methodology
The authors develop a refined, agent-based, stock-flow-consistent (SFC) macroeconomic model (version v3) designed to overcome criticisms of earlier iterations regarding frictionless taxation and imposed wealth concentration.

Structure: The model features four sectors (Households, Firm, Government, Rest of World) with strict accounting invariants ensuring deposits sum to zero and net worth equals the capital stock. It employs a nested Constant Elasticity of Substitution (CES) production function with three clusters: a routine cluster (labour + robotic capital), a cognitive cluster (labour + AI compute), and a top-level CES linking them.
Concentration Mechanism: Unlike previous versions using a kinetic kernel, wealth concentration is microfounded through heterogeneous, persistent returns to wealth (calibrated to Fagereng et al., 2020). Returns follow an AR(1) process with demographic turnover, generating a stationary Pareto tail.
Behavioural Responses: The wealth tax is modelled as behavioural, inducing both base erosion (avoidance, calibrated to a bunching elasticity of 0.75) and capital flight (migration, calibrated to a 2% semi-elasticity per percentage point of tax).
Two-Channel Extension: The model separates the "robotic" channel (physical, domestic, competitive return) from the "AI" channel (mobile IP rent, foreign-held). The AI rent is modelled as a markup on cognitive value added, repatriated as a deductible licence fee.
Validation: Results are derived from Monte Carlo runs (10–12 seeds) over a 600-period horizon. Robustness is assessed via global (Sobol) sensitivity analysis and formal local-stability analysis of the deterministic skeleton.

Key Contributions

Microfounded Concentration: The model demonstrates that heavy-tailed wealth distributions emerge endogenously from heterogeneous returns to wealth rather than being imposed by a kinetic kernel.
The Two-Channel Distinction: It formally separates the competitive return on physical robotic capital from the mobile IP rent of AI, showing that standard corporate taxes and robot taxes largely miss the latter.
Source vs. Residence Taxation: The paper provides a positive analysis of how different tax bases (residence-based wealth/income taxes vs. source-based levies) interact with foreign ownership and capital mobility.
Stability and Solvency: It establishes that the real economy is locally stable under all regimes, but fiscal solvency depends on the adequacy of taxation to neutralize the $r > g$ debt root.

Key Results

Inequality and Tax Instruments: Stock taxes (wealth taxes) compress the wealth Gini coefficient sharply (from ~0.63 to ~0.15) and accumulate sovereign equity. Flow taxes (income tax + UBI) fail to compress inequality, leaving it near laissez-faire levels. The "composition effect" (capital fleeing offshore) is real but modest; the wedge between measured and true inequality remains small at empirically calibrated rates.
The Durable Surplus: The primary durable surplus is the foreign-held AI IP rent. In the mature economy, ordinary corporate tax reaches almost none of this rent because it is a deductible licence fee. A robot tax collects on physical income but misses the rent entirely.
Effective Instruments: Only a source-based levy on AI revenue (Digital Services Tax style) or a withholding tax on the repatriated rent effectively reaches the surplus. These instruments reduce foreign ownership of the domestic capital stock.
Foreign Ownership Drift: Without intervention, foreign owners recycle the rent into domestic equity, accumulating up to 70% of the capital stock. Source levies slow this drift, reducing foreign ownership to roughly 40–50%.
Rising Rent Scenarios: If the AI rent grows as a share of output (due to pricing power or AI-native production capturing output), the stakes increase significantly. Foreign ownership can approach the entire capital stock, and under-taxed regimes face insolvency.
Labour Market and Unemployment: The fall in the labour share is conditional on "reinstatement" (creation of new tasks). If displacement outpaces reinstatement, unemployment can rise to 10–20%. Funding a safety net for this unemployment is only affordable if the tax base reaches the AI rent; relying solely on the shrinking income tax base leads to fiscal insolvency.
Territoriality and Strategy: The "territorial objection" (that domestic taxes drive activity offshore) is decisive for the provider's home country (US) but largely irrelevant for a rent-importing host (UK/EU), as the levy targets value recognized domestically regardless of server location. However, a strategic owner can still shift the booking of the rent (profit shifting), requiring anti-avoidance rules.

Significance and Claims
The paper claims to offer a conditional, positive analysis rather than a normative welfare optimum. Its significance lies in identifying the specific fiscal levers required to manage the transition to an automated economy, particularly for nations that do not own the AI IP.

Policy Implication: For rent-importing hosts, the load-bearing instrument is a source levy on AI value (or a withholding on the rent), not a robot tax or standard corporate tax. This instrument curbs the transfer of capital ownership abroad and funds the necessary safety nets.
Timing: Intervention is time-sensitive; the cost of waiting is convex, and the window for cheap intervention narrows as automation deepens.
Ownership vs. Revenue: The paper argues that monitoring the drift of capital ownership is as critical as revenue collection, as the former determines the long-term distribution of returns from automation.
Limitations: The authors explicitly state that the model's results are conditional on the closure (output at potential, investment as residual) and specific calibration assumptions (e.g., the size of the rent, the elasticity of AI supply). The magnitudes are illustrative ranges rather than point forecasts, and the model does not yet incorporate a full market-clearing asset price mechanism or a social welfare function.

The paper concludes that while the debate often focuses on the mechanics of taxing tokens or robots, the fundamental economic reality is the mobility of the IP rent. Effective policy for a rent-importing economy requires instruments that reach this rent at the source, paired with anti-avoidance measures, to prevent the erosion of domestic ownership and fiscal solvency.