Behind the forecast: how do you build a model for the age of automation?
From scenario assumptions through government accounting to sensitivity tests — and what still cannot be concluded.
A good model separates accounting identities, behavioral assumptions and empirical evidence; a numerical result is not causal proof.
Start with a question, not a number
The question is how automated productivity, job loss and the distribution of ownership affect real welfare and the government's financing capacity. An unemployment rate of 20% is a starting point for examination.
First layer: accounting
The incomes and expenditures of households, firms and government must be shown to be consistent with one another. If a public grant is paid, its source of financing and its effect on debt or on a paying party must also be identified.
Second layer: economic responses
Prices, wages, investment, taxation and consumption respond to one another. Some responses are chosen as simulation assumptions; others require calibration from studies and data.
Third layer: sensitivity tests
Assumptions about capital ownership, price declines, debt and benefit financing are swapped to check whether the conclusion holds. High sensitivity indicates that the finding is conditional on the assumption.
What is still missing?
Identified, causal data, a suitable cross-country sample, labor market transition mechanisms and external validation are needed before the results can be used as a policy forecast. Documenting the steps on the site allows readers to examine the limitations.
Sources and transparency
This publication explains the existing research in accessible language. Its accounting example and scenarios are neither official forecasts nor investment recommendations.
What do you think?
Which model assumption would you test first using real data?