Climate Risk and Corporate Financial Asset Allocation: An Analysis Based on Double Machine Learning

Abstract This paper examines how climate physical risk reshapes the financial assets that non-financial firms hold. We build an economic model, simulate it, and estimate it on firm-level data. In the model, climate risk raises corporate financialization through three channels that work in parallel. Firms substitute away from physical capital whose expected return has fallen. They also hold financial assets as a precaution, valuing their liquidity more highly as climate-driven cash-flow risk rises. And because physical investment is irreversible, greater uncertainty makes them hold back. The total effect splits cleanly into these three parts. Simulating the model, we find that each channel matters, that the response concentrates among financially flexible firms, and that its composition shifts from substitution and precaution toward real options as firms grow rigid. A Monte-Carlo experiment then confirms that double machine learning (DML) recovers the structural effect from a synthetic panel. In a panel of 3,415 Chinese A-share firms over 2006 to 2019, with climate risk measured by the volatility of local daily precipitation, DML puts the average effect at 0.027. The effect is linear in risk and concentrated in the top tercile of firms, where it reaches 0.054, and it survives a horse-race across eight learners and a firm-cluster bootstrap. Climate risk is nonetheless a weak out-of-sample predictor, a warning against reading corporate financialization as a simple gauge of climate exposure. JEL: G11; G32; Q54; C14; C55; C63