Mark Verhagen | Searcharxiv

arXiv2026

Births are difficult to predict even with rich survey and full-population register data

Major life events have proven difficult to predict. Does this reflect limits of theory, data, and algorithms, or the large role of chance? We examine one outcome - having a child within three years - through a near-ideal setting for prediction: a data challenge where 147 researchers predicted births for Dutch residents aged 18-45, using survey data and full-population registers. Methods ranged from logistic regression to a large language model and transformers. Predictions were moderately accurate (best F1: register 0.59, survey 0.76); advanced models did not outperform classical ones; and the larger registers did not beat the survey. Simulating the stochastic biology of conception and pregnancy, we estimated a predictive ceiling (survey F1 ~ 0.86-0.94, register 0.88-0.96). Observed performance falls short of this ceiling, implicating imperfect data, methods, and unmodelled chance, while the ceiling itself shows that chance in reproduction alone sets a non-trivial limit on predicting individual lives.

cs.LG↗

arXiv2025

Tax reform as a constrained optimization problem: a piecewise-linear framework and software implementation

In many countries, income tax codes have grown into a complex tangle of interacting brackets, benefits, and deductions. Despite widespread calls for systematic reform, successful attempts at reform are rare. Part of the problem is the difficulty of designing viable reform proposals. Politically viable reform must offer hard guarantees on income effects, marginal rates, and budgetary cost. Existing microsimulation tools can evaluate a reform proposal but cannot generate one by themselves. We develop a framework that casts tax reform as a constrained optimization problem. We show that any statutory tax code satisfying four mild assumptions reduces to a finite-dimensional piecewise-linear function for each taxpayer group, so reform becomes a linear or mixed-integer linear program whose decision variables are legislatable parameters: rates, bracket cutoffs, and lump-sum transfers. We are able to recover current tax systems and generate provably optimal reform candidates within the modeled space, or a certificate that no reform satisfying certain policy design constraints exists. Behavioral effects can also be incorporated, producing a nonconvex mixed-integer formulation. We demonstrate the framework through a near-complete reconstruction of the Dutch income tax code, generating reforms that smooth marginal-rate spikes, cap household income losses, and roughly halve the number of active rules through a lexicographic procedure. Developed in close collaboration with the Dutch Ministry of Finance, the methodology is currently in active use there. An open-source software implementation is available as \texttt{TaxSolver}.

q-fin.GN↗