Many key economic objects are latent: central-bank news, the expected path of short rates, and firms’ true decarbonization potential. This thesis offers three essays that recover such signals by aligning measurement with information and data properties. First, it uses large language models to summarize the Fed’s documentary information set and predict the policy rate; deviations from the announced rate provide narrative monetary shocks that avoid common puzzles and forecast bond returns. Second (with Carlo Favero), it extends affine term-structure models with a stochastic trend in short rates linked to slow-moving fundamentals, producing stationary and countercyclical term premia. Third (with Giacomo Bezzi, Mariano Massimiliano Croce, and Gimede Gigante), it develops a signaling model of ESG disclosure showing that forward-looking progress-to-target scores can separate genuinely green firms from imitators and earn alpha. Together, the essays bridge the gap between what agents know and what econometricians observe.