Machine learning is useful because heat damage is unlikely to follow one stable, linear relationship across companies. A small increase in temperature may have little effect until an operating threshold is reached. Beyond that point, heat can reduce productivity and equipment efficiency, increase cooling needs, or constrain production.
The effect can also differ widely across companies facing similar weather. An automated plant with modern cooling and greater operational flexibility may continue operating. A labor-intensive business with older equipment and greater worker exposure to heat may experience a sharp loss of output. Averaging the two can make the overall effect appear modest even when one company faces a material financial loss.
Machine learning can search for these threshold effects and differences across firms more effectively than a model built around one average relationship. A May 2026 working paper by Christian Breitung, Gerard Hoberg, and Sebastian Müller, “Machine Learning the Impact of Climate Change on Firms Worldwide,” illustrates how a machine-learning framework can capture those differences. Their models estimate how abnormal seasonal temperature and precipitation affect sales, efficiency, profitability, and costs, with effects varying widely across firms. The adverse effects are concentrated in more exposed industries, labor-intensive and older firms, and companies operating in less developed regions.
Investors therefore need what could be called a company’s “heat-response function”: an estimate of how production, costs, margins, and cash flow change once temperatures exceed thresholds that matter to its operations. Absent that relationship, investors do not have a valuation input.
