AI market developments depress coal and gas returns over long horizons but lift crude-oil returns, with effects that flip across market states and time scales. Climate-policy uncertainty, meanwhile, strengthens long-run returns for coal and natural gas, implying higher risk premia for carbon-intensive fuels when policy is unpredictable.

Main Finding

AI market developments and climate policy uncertainty (CPU) affect fossil-fuel returns in markedly different, nonlinear ways that depend on fuel type, market state (quantile), and investment horizon. Specifically, AI developments are largely negatively associated with coal and natural gas returns over longer horizons, while they are positively and more persistently associated with crude oil returns (especially at medium and long horizons). CPU shows mixed short-term effects but is positively associated with coal and natural gas returns over longer horizons; its relationship with oil is weaker and more segmented.

Key Points

• Heterogeneity across fuels: coal, natural gas, and crude oil respond differently to AI developments and CPU.
• AI effects:
  • Coal & natural gas: predominantly negative associations over longer horizons.
  • Crude oil: positive and more persistent associations, particularly at medium and long horizons.
• CPU effects:
  • Short-term: mixed and localized across market states.
  • Long-term: positive association with coal and natural gas returns.
  • Oil: weaker, segmented relationships across market quantiles.
• Relationships are nonlinear and vary across market quantiles (i.e., different effects in bearish vs. bullish market states) and across short-, medium-, and long-term frequencies.

Data & Methods

• Data: daily observations from August 20, 2019 to June 9, 2025.
• Variables:
  • Dependent: returns on three fossil fuels — coal, natural gas, crude oil.
  • Regressors: measures capturing AI market developments and climate policy uncertainty (CPU).
• Methodology: wavelet quantile-on-quantile regression — combines wavelet-based multi-scale decomposition (short/medium/long horizons) with quantile-on-quantile analysis to map how the conditional quantiles of fuel returns respond to different quantiles of AI/CPU indicators across time scales. This identifies nonlinear, state-dependent, and frequency-dependent associations.

Implications for AI Economics

• AI as a financial/economic risk factor: AI market developments should be considered a distinct driver of energy-asset returns and included in asset-pricing, risk-management, and portfolio-allocation models, with fuel-specific treatments.
• Investment strategy and hedging:
  • Investors need horizon- and state-dependent strategies: AI exposure appears detrimental to coal and gas over long horizons, but beneficial for oil; hedges should reflect these asymmetries.
• Policy design and climate transition:
  • CPU raises long-term coal and gas returns, suggesting that climate-policy uncertainty can increase risk premia for carbon-intensive fuels—highlighting the value of credible, predictable transition policies to reduce market frictions and undue long-term price support for fossil fuels.
• Modeling recommendations for AI economics research:
  • Use multi-scale and quantile-based methods to capture nonlinear, state-dependent effects of technological (AI) shocks.
  • Disaggregate effects by fuel/sector; aggregate conclusions across fossil fuels can hide important heterogeneity.
• Further research: explore causal channels (e.g., demand, efficiency, substitution to electricity/renewables), extend to other asset classes, and evaluate policy interventions that alter the identified relationships.