Modelling and forecasting crude oil price volatility with climate policy uncertainty

Accurate prediction of crude oil price volatility is crucial for various stakeholders, including investors, companies, and energy organizations. Existing research has explored the use of various uncertainty indicators, such as economic policy uncertainty (EPU), geopolitical risk (GPR), and equity market uncertainty (EMV), to improve volatility forecasts. This paper aims to enhance the predictability of crude oil volatility using the climate policy uncertainty (CPU) index and its change indicators. Climate uncertainty affects crude oil markets through two primary channels: its impact on economic activity (affecting supply and demand) and its influence on financial markets. To capture the nuances of climate uncertainty, the study employs not only the original CPU index but also two change indicators: the change in CPU (CCPU) and the increase in CPU (ICPU). These indicators address the potential for limited attention and focus on attention-grabbing positive innovations. The paper empirically investigates the predictive power of these three CPU-related indicators.

The literature extensively explores the relationship between crude oil volatility and various uncertainty indicators. Studies have examined the role of EPU, GPR, and EMV in forecasting oil volatility, finding varying degrees of predictive power. This paper distinguishes itself by employing the newly proposed CPU index and its change indicators, providing a new perspective on the impact of climate-related uncertainty on crude oil markets. Existing literature also highlights the effects of climate change on commodity markets through physical and transition risks, as well as the potential for investor flight from the oil and gas industry due to escalating climate concerns. This study complements these works by directly examining the predictive power of CPU on crude oil market volatility.

The study uses realized volatility (RV), calculated as the sum of squared daily returns, as a measure of crude oil market volatility. Logarithmic RV (LV) is used in predictive regressions to address the non-Gaussian nature of RV. The analysis employs autoregressive (AR) models as a benchmark, and these are extended to include CPU-related variables. The Akaike information criterion (AIC) guides the selection of the lag order in the AR model. The predictive power of CPU, CCPU, and ICPU is assessed through coefficient estimates, Newey-West t-statistics, and incremental R². Out-of-sample forecasting performance is evaluated using the Campbell and Thompson (2008) R² statistic and the Clark and West (2007) test. The economic significance of the findings is assessed through an asset allocation exercise, employing a mean-variance framework and the realized certainty equivalent return (CER) to evaluate the economic gains from using CPU-related volatility forecasts. The study also utilizes the method of Bollerslev et al. (2018) to assess the average utility gains from using the forecasts. Finally, the relationship between CPU-related indicators and oil fundamentals (production and consumption) and CO2 emissions is investigated to understand the source of the predictive power.

The in-sample analysis reveals that CPU change indicators (CCPU and ICPU) significantly predict crude oil market volatility, while the original CPU index does not. The out-of-sample analysis, using both expanding and rolling windows, confirms the significant positive predictive power of CCPU and ICPU, while the CPU index exhibits negative predictive power. Asset allocation exercises demonstrate that models incorporating CCPU and ICPU yield substantially higher certainty equivalent returns (CER) for crude oil futures investors compared to models using the original CPU index or the AR benchmark, both with and without transaction costs. The analysis of business cycles shows that the predictive power of CPU-related indicators is concentrated during economic expansions. Furthermore, the predictive ability of CPU change indicators remains significant even after controlling for other economic indicators and uncertainty indexes, suggesting that these indicators contain unique information related to climate risk. The investigation of oil fundamentals indicates that CPU change indicators have a significant negative correlation with oil consumption and CO2 emissions, suggesting that the predictive power stems from the impact of climate policy uncertainty on oil demand.

The findings address the research question by demonstrating that changes in climate policy uncertainty, captured by CCPU and ICPU, offer significant improvements in the predictability of crude oil market volatility. The economic significance of this improved predictability is evident in the substantial economic gains for investors using these forecasts in asset allocation strategies. The identification of oil consumption as the key economic channel linking CPU changes to volatility provides valuable insights into the underlying mechanisms driving the observed relationship. The robustness of the findings, even when controlling for other economic indicators and uncertainty indices, underscores the unique information contained within the CPU change indicators. The concentration of predictive power during economic expansions highlights the importance of considering business cycle conditions in the context of climate policy uncertainty and its impact on oil markets.

This paper contributes significantly to the literature by demonstrating the predictive power of climate policy uncertainty changes for crude oil volatility. Changes in CPU, particularly increases, outperform the original CPU index and other common uncertainty measures. The economic value of this finding is considerable for investors. Future research could explore more targeted measures of CPU, examine differences between physical and transition climate risks, and employ non-linear models to capture potential complexities in the relationship.

The study focuses on the US climate policy uncertainty index and WTI crude oil. The results might not be generalizable to other regions or oil types. The reliance on specific economic indicators and uncertainty indexes may limit the scope of analysis. The model used is linear, and a non-linear model might reveal additional insights.