Seeing is believing: Forecasting oil market returns with artificial intelligence-powered visual climate change perception

Abstract

This study proposes a novel framework for forecasting oil market returns by quantifying climate change perception based on visual media. A vision-language model processes 746,435 news images from The New York Times between May 2006 and December 2023 to construct the Visual Climate Change Perception Index (VCCP), along with two sub-indices. The VCCP exhibits significant predictive power for one-month-ahead WTI spot returns, outperforming text-based climate sentiment and macroeconomic benchmarks. The Physical Climate Impact Visual Index contributes to short-term return predictability, while the Transitional Climate Policy Visual Index captures longer-horizon dynamics. Out-of-sample analyses confirm the robustness and economic relevance of VCCP-based models, enhancing forecast accuracy and improving asset allocation performance. These findings underscore the role of emotionally salient visual cues in shaping market expectations and highlight the importance of multimodal climate signals in the pricing of high-carbon assets.