Sven van Holten Charria

Do households perceive the income risk that macroeconomic models assign them? Job stayers in households below $50k report about 1.8 times the wage uncertainty of those above $100k, hold these beliefs more persistently, and revise them more strongly following realised wage shocks, while the realised risk estimated from matched same-job spells shows no comparable gradient. Under an estimated learning rule, the bottom group over-perceives its transitory risk, while the middle and top groups under-perceive their permanent risk. I trace both patterns to how legibly pay separates into these two components: the labelled share of transitory income rises from about one-eighth at the bottom to over one-third at the top. In a life-cycle economy, these beliefs reduce liquid wealth by five percent at the bottom and fourteen percent at the top, where households consume over two percentage points more of a windfall within the year. Equating perceived with realised risk therefore overstates precautionary wealth and understates consumption responses, particularly among high-income households.

This thesis examines the ‘Post-Double Selection’ methodology introduced by Belloni et al. (2014b) for inference on treatment effects in high-dimensional datasets. Economically intuitive variables are incorporated directly into the selection process through lower penalization. Sufficient conditions are established for asymptotically valid inference, allowing imperfect beliefs when the size of the amelioration set is controlled. In a simplified setting, sufficient conditions are established under which the Discrete penalty reduces absolute bias and mean squared error relative to regular Post-Double Selection. Monte Carlo simulations demonstrate improvements over traditional Post-Double Selection under certain conditions. An empirical re-examination of the effect of abortion on crime rates, as studied by Donohue and Levitt (2001), highlights the practical benefits of this approach, leading to a more comprehensive control selection and meaningful economic interpretations.

This paper derives an analytical equivalence between ARCH(p) and AR(p) to recast volatility forecasting as a time-varying autoregressive parameter problem, then uses the Macroeconomic Random Forest to estimate those parameters in a high-dimensional setting. The approach traces the effects of Brexit and COVID-19 on USD/GBP baseline volatility and shock persistence, revealing smoother regime transitions and a clearer interpretation of macroeconomic shock dynamics than conventional parametric models.