TreasureX Researchnber.wpNew Method Makes Complex Earnings Models Easier to Estimate
Neele L. Balke, Stéphane Bonhomme, Thibaut Lamadon · October 16, 2026
Economists often build models with hidden variables to explain how people's earnings evolve over time. These models are powerful but hard to fit, because the math requires integrating over unobserved factors — a step that is usually impossible to do exactly.
Variational inference (VI), a tool from machine learning, sidesteps this by turning the problem into a simpler optimization. The catch: when the approximation is too rigid, the estimated parameters can be biased.
Balke, Bonhomme, and Lamadon first test VI on earnings-dynamics models and show that the choice of approximation matters a lot. They then propose indirect variational inference (IVI), which treats the variational step as an auxiliary model and corrects the resulting bias. Crucially, IVI keeps VI's speed because it never needs to compute the full likelihood.
Applying these tools to models with nonlinear persistence, non-Gaussian and serially correlated transitory shocks, and latent heterogeneity, they find that flexible approximations plus IVI produce reliable estimates in both simulated and real data.
Why it matters: better estimates of earnings dynamics feed into tax policy, social insurance design, and inequality research. For investors, improved labor-income models sharpen forecasts of consumer spending, credit risk, and macro trends — though the paper names no specific companies.
Source: National Bureau of Economic Research — https://www.nber.org/papers/w35877#fromrss
Educational content only, generated with AI assistance from public research abstracts. Not investment advice.