Sven van Holten Charria

Standard heterogeneous agent models assume households perceive income risk correctly. Using density forecasts from the NY Fed SCE, I show they don't: perceived risk has its own persistent dynamics, and the persistence is steeply income-graded. The same gradient does not appear in realised income data, suggesting a belief distortion rather than differences in the actual income process. Income surprises raise perceived risk at the bottom of the distribution but not at the top, and the effect lingers far longer for low-income households. I embed these dynamics in a two-agent model where excess precautionary saving from biased risk beliefs suppresses consumption even after income itself has recovered.

Standard post-double selection relies on purely data-driven penalisation to choose controls. This paper integrates economically motivated prior beliefs on which controls matter into the Lasso penalty structure before selection, not after. Monte Carlo simulations show that belief-informed penalties reduce variance in treatment effect inference while maintaining comparable bias. An empirical re-examination of the effect of abortion on crime rates yields tighter inference, more interpretable control sets, and fewer spurious selections.

This paper derives an analytical equivalence between GARCH(1,1) and ARMA(1,1) to recast volatility forecasting as a time-varying autoregressive parameter problem, then uses macroeconomic random forests to estimate that parameter in a high-dimensional setting. The approach isolates the causal effects of Brexit and COVID-19 on USD/GBP baseline volatility and persistence, revealing smoother regime transitions and sharper identification of macroeconomic shock transmission than conventional parametric models.