AI Stocks Move Clean Energy, Oil and Food Markets in Surprising Asymmetric Ways

Artificial intelligence is often described as a single force reshaping the global economy, but new research suggests its financial fingerprints look very different depending on where you look — and on whether the technology is having a good day or a bad one. A study published in the journal Environmental Challenges examines how the performance of AI-related equities travels through three strategically distinct markets: clean energy, fossil fuels, and agribusiness. Using daily data spanning more than eight years, from 15 June 2018 to 29 June 2026, the authors find that the connections between AI stocks and these sectors are neither uniform nor symmetric, and that they change dramatically depending on the investment horizon and on market conditions.

The research team, led by Victoria Olushola Olanrewaju and including Seyed Alireza Athari, Mohamed Djafar Henni, Anar Eminov, and Emmanuel Oluwatosin Adewusi, measured AI through the S&P Kensho Artificial Intelligence Enablers Index. They then tracked its relationship with the S&P Global Clean Energy Transition Index, the S&P GSCI Energy Index, and the S&P Global Agribusiness Index. The choice of sectors is deliberate. AI is widely credited with improving renewable energy forecasting, smart-grid optimization, and storage management, yet it simultaneously boosts efficiency in oil and gas operations through predictive maintenance, reservoir characterization, and production optimization. In agriculture, AI underpins precision farming, crop monitoring, yield prediction, and supply-chain coordination. Any single average effect would therefore risk obscuring a far messier reality.

That messiness is precisely what the data reveal. Before running their main models, the authors checked the statistical character of each return series, and the results were striking. The fossil-energy index showed the widest swings of all, with daily returns ranging from a plunge of more than 30 percent to a gain of nearly 16 percent, and a standard deviation of 2.51 — far larger than the dispersion seen in AI, clean energy, or agribusiness returns. All four series were negatively skewed and heavily leptokurtic, meaning they exhibit fat tails and sudden extreme moves far more often than a normal distribution would predict. Jarque-Bera tests rejected normality at the one percent level for every series, while BDS diagnostics uncovered nonlinear dependence across multiple embedding dimensions. In plain terms: these markets do not behave like tidy textbook systems, and linear models that report a single average relationship would hide much of the action.

To confront this complexity, the study deployed a relatively novel econometric toolkit called Asymmetric Wavelet Quantile Regression, or AWQR. The method combines three ideas. First, quantile regression, introduced by Koenker and Bassett in 1978, estimates relationships across the entire conditional distribution of outcomes rather than just the mean, capturing what happens in calm markets and in turbulent tails alike. Second, a maximal overlap discrete wavelet transform using the LA8 filter decomposes each time series into frequency components, separating short-term fluctuations lasting up to 168 trading days, medium-term cycles running to 672 days, and long-horizon components beyond that. Third, and most importantly for the study’s headline finding, the AI return series is split into positive and negative components, allowing the researchers to ask whether good news for AI stocks moves markets differently than bad news.

The short-horizon results were, in a sense, a story of restraint. Across all three markets, the associations between AI-related equity performance and sector returns were generally small and mostly statistically insignificant over horizons of up to roughly 168 trading days. AI-driven market movements, in other words, do not appear to translate into immediate, systematic shifts in clean energy, fossil energy, or agribusiness performance. This finding challenges a popular narrative in which every AI headline instantly ripples through commodity and sector markets. Instead, the transmission appears to build over time, suggesting that investors and market participants need longer windows to reprice the economic implications of technological change.

At the medium horizon, the picture changed sharply. The symmetric specification — which uses the full AI return series — produced predominantly positive coefficients across the conditional distribution for all three markets, indicating that stronger AI-related equity performance tends to accompany stronger sector returns over these intermediate windows. But when the authors separated positive AI shocks from negative ones, a fascinating pattern emerged: the positive-shock coefficients were frequently negative, while the negative-shock coefficients were frequently positive. This is not a contradiction. The symmetric coefficient summarizes the net association over all AI movements, while the directional models condition separately on observations of opposite sign, revealing information that aggregation conceals. For the fossil-energy index, the negative association with positive AI shocks at the medium horizon is consistent with market expectations that AI advances support energy efficiency and substitution toward cleaner technologies, even as AI simultaneously improves drilling and production economics.

Crucially, the authors refused to declare asymmetry merely because two coefficients had different signs. They applied a formal bootstrap Wald test of the null hypothesis that the positive- and negative-shock coefficients are equal, using 499 pairs-bootstrap replications for each quantile-horizon cell. Only where this test rejected equality did they conclude that genuine asymmetry exists. The verdict: for the clean energy index, the fossil-energy index, and the agribusiness index alike, the null of coefficient equality was rejected throughout most of the medium horizon and across a substantial share of long-horizon quantiles, while evidence at short horizons was far more limited. The asymmetry, in other words, is statistically real where it matters most — at the horizons where medium- and long-term investors operate.

The long-horizon results added further texture. For the clean energy market, the symmetric association shifted from negative or weak values at lower quantiles toward positive values at the upper tail, hinting that AI and clean energy move together most strongly when clean energy markets are already performing well. For fossil energy, coefficients varied substantially across quantiles, with some positive long-horizon associations at upper quantiles plausibly reflecting the growing use of machine learning in petroleum exploration, predictive maintenance, and production forecasting. For agribusiness, the long-horizon symmetric coefficient was negative at lower quantiles and positive toward the upper tail, a profile consistent with AI’s role in crop monitoring, yield prediction, logistics, and resource allocation — benefits that may matter most when agricultural markets are strong.

The authors are careful about what these findings do and do not mean. The framework is deliberately bivariate, with no controls for broader stock-market conditions, interest rates, exchange rates, commodity prices, or geopolitical events, any of which could influence both AI equities and the sector indices. The indices may also share constituent companies or common factor exposures, meaning part of the estimated comovement could reflect overlapping exposures rather than a distinct AI transmission channel. And with hundreds of coefficients estimated across quantiles and horizons, isolated significant results could arise by chance, since no multiple-testing correction was applied. The researchers therefore place greater weight on patterns that persist across adjacent quantiles, survive lag-based robustness checks at lags one through four, and pass the formal Wald tests. The conclusions are framed as conditional associations, not causal effects.

Even with those caveats, the policy implications are tangible. For clean energy, the results reinforce the case for investment in AI-enabled forecasting, grid management, and storage applications. For fossil-energy activities, AI can be directed toward operational efficiency, emissions monitoring, safety, and environmental compliance — a reminder that the same technology driving the energy transition is also entrenching itself in the industry it may eventually displace. For agribusiness, digital infrastructure, precision agriculture, and climate-risk forecasting emerge as practical levers, though barriers such as high costs, weak infrastructure, and skill gaps can create uneven adoption. Perhaps the deepest lesson is for investors and policymakers alike: the market consequences of the AI revolution are not a single number but a landscape — one that shifts with the direction of the shock, the part of the distribution you occupy, and the clock you are watching.

Subject of Research: Asymmetric transmission of AI-related equity performance to clean energy, fossil energy, and agribusiness markets

Article Title: Digital intelligence, green transition, and fossil energy: Asymmetric market transmission from AI-related equity performance

Article References: Digital intelligence, green transition, and fossil energy: Asymmetric market transmission from AI-related equity performance. (n.d.). https://doi.org/10.1016/j.envc.2026.101673

Image Credits: AI Generated

DOI: 10.1016/j.envc.2026.101673

Keywords: artificial intelligence, clean energy, fossil fuels, agribusiness, quantile regression, wavelet analysis, financial markets, energy transition, asymmetric shocks, S&P indices, econometrics, market volatility