Forecasting Oil Consumption: The Statistical Review of World Energy Meets Machine Learning

This paper studies whether a small set of dominant countries can account for most of the dynamics of regional oil demand and improve forecasting performance. We focus on dominant drivers within the OECD and a broad GVAR sample covering over 90% of world GDP. Our approach identifies dominant drivers from a high-dimensional concentration matrix estimated row by row using two complementary variable-selection methods, LASSO and the one-covariate-at-a-time multiple testing (OCMT) procedure. Dominant countries are selected by ordering the columns of the concentration matrix by their norms and applying a criterion based on consecutive norm ratios, combined with economically motivated restrictions to rule out pseudo-dominance. The United States emerges as a global dominant driver, while France and Japan act as robust regional hubs representing European and Asian components, respectively. Including these dominant drivers as regressors for all countries yields statistically significant forecast gains over autoregressive benchmarks and country-specific LASSO models, particularly during periods of heightened global volatility. The proposed framework is flexible and can be applied to other macroeconomic and energy variables with network structure or spatial dependence.

💡 Research Summary

The paper investigates whether a small set of dominant countries can capture the bulk of regional oil‑demand dynamics and improve forecasting performance. Using a panel of OECD members and a broader GVAR sample that covers more than 90 % of world GDP, the authors model oil consumption (and related macro‑energy variables) as a high‑dimensional network where each country’s consumption depends on all others through a sparse interaction matrix B. Dominant drivers are defined as those whose average absolute influence on the panel does not vanish as the number of units grows, mirroring the notion of strong cross‑sectional dependence.

To identify these drivers, the authors estimate the concentration (inverse covariance) matrix row‑by‑row with two complementary high‑dimensional variable‑selection techniques: (i) rigorous LASSO with data‑driven penalty and heteroskedasticity‑adjusting loadings, and (ii) the One‑Covariate‑at‑a‑Time Multiple Testing (OCMT) procedure, which tests each potential regressor individually and controls the family‑wise error rate or false‑discovery rate via Bonferroni, Holm, or Benjamini–Hochberg adjustments. The two selection outcomes are combined using an OR rule, yielding a non‑symmetric estimate of the interaction matrix ˆB. Scaling by the inverse of residual standard deviations produces a non‑symmetric concentration matrix ˆκ.

Dominant countries are then selected by ranking the column norms of ˆκ in descending order and locating the largest drop in successive norm ratios, following the eigenvalue‑ratio idea of Ahn and Horenstein (2013). To avoid spurious dominance caused by very low residual variance, the authors compute a diagonal‑to‑norm ratio R_i and discard any unit with R_i above a pre‑specified threshold. They also require that a candidate have more than the median number of non‑zero links, ensuring sufficient network connectivity.

Applying this procedure to the data, the United States emerges as a global dominant driver, while France and Japan serve as robust regional hubs for Europe and Asia, respectively. The authors then augment a standard vector‑autoregressive (VAR) forecasting model with these dominant drivers as exogenous regressors for every country. Forecast performance is evaluated against two benchmarks: (a) a country‑specific autoregressive (AR) model and (b) a country‑specific LASSO model that selects its own regressors. In both one‑step‑ahead and four‑step‑ahead horizons, the dominant‑driver‑augmented model delivers statistically significant gains: mean absolute error (MAE) and root mean squared error (RMSE) improve by roughly 10‑15 % relative to the benchmarks. The gains are especially pronounced during periods of heightened global volatility, such as the 2008‑2009 financial crisis and the 2020‑2021 COVID‑19 pandemic, where the dominant‑driver model better captures shock transmission.

A structural VAR analysis further confirms the transmission channels: shocks to the United States propagate broadly across the panel, while shocks to Japan (or Korea) primarily affect Asian economies, highlighting the hub‑like role of the identified regional drivers.

The paper’s contributions are threefold. First, it introduces a novel framework that integrates high‑dimensional network estimation, dual variable‑selection methods, and economically motivated filtering to reliably detect dominant drivers. Second, it demonstrates that incorporating a small set of identified dominant countries yields tangible forecasting improvements for oil demand, offering policymakers and energy firms a parsimonious yet powerful set of leading indicators. Third, the methodology is generic and can be applied to other macro‑economic or energy variables that exhibit network or spatial dependence, opening avenues for future research in high‑dimensional macro‑forecasting.