Oil is not a black swan: how a mathematical model predicted the volatility of the Iran-Israel conflict

What if major geopolitical conflicts aren’t as unpredictable as we think? A new study shows that the March 2026 crisis was already “written” into the market’s risk structure.

The insight that changes the game

In March 2026, when tensions between Iran, the United States, and Israel escalated into open conflict, energy markets reacted with the violence one would expect. Brent Crude Oil prices spiked, volatility surged, and headlines repeated the familiar refrain: “An unpredictable event shakes the markets.”

But what if that’s not quite right? What if, instead of being a black swan — that impossible-to-anticipate event — the geopolitical crisis was simply the manifestation of a risk dynamic already there, latent, waiting to be activated?

A recent academic study makes exactly that case. And it does so with an approach as provocative as it is rigorous: it uses a mathematical model — an ARMA-GARCH, for those familiar — calibrated exclusively on previous geopolitical tensions (2023–2025) and applies it to the critical period from September 2023 to March 2026. The result is striking: the model “predicts” the volatility of the crisis with 93% accuracy.

Beyond price: the physics of uncertainty

The author isn’t interested in predicting whether the price of oil goes up or down. That would be like trying to predict the exact trajectory of a particle in turbulent fluid: impossible.

What matters is the energy of the system, the market’s “temperature”: volatility.

From this perspective, a market isn’t a pricing machine. It’s a complex system with memory. When a geopolitical conflict occurs, the system doesn’t “invent” a new way to react. It simply activates a risk structure that already existed — a kind of “dynamic attractor” that determines how uncertainty behaves.

The key question, then, is: did the Iran-Israel conflict activate the same attractor that operated during previous tensions (2023–2025), or was it a complete rupture?

The study’s answer is clear: the 2026 conflict did not create a new risk dynamic. It simply activated the one that already existed.

The evidence: numbers that speak

The study, which analyzes nine distinct historical periods, offers compelling data:

• The model calibrated on the 2023–2025 period (an eGARCH(2,1)-ARMA(1,1) with heavy tails) achieves an AIC of -5.2587, an indicator of excellent fit.
• Applied to the crisis period (2023–2026), the same model maintains an AIC of -5.1987, a marginal degradation of just 1.14%.
• The hit rate — the percentage of actual observations falling within the model’s confidence intervals — is 93.1%.
• Volatility persistence — how long it takes for a shock to dissipate — remains virtually identical across both periods: about 19 days.

But perhaps the most fascinating finding is the contrast with the COVID-19 crisis of 2020.

Two crises, two radically different behaviors

When the 2023–2025 model is applied to the pandemic period, the result is catastrophic:

• The hit rate collapses to 33.3%. Only one in three predictions is accurate.
• The error (MAE) increases by 39% relative to the reference period.

The conclusion is revealing: not all crises are alike. The COVID-19 crisis was a structural break: demand collapse, negative oil prices, global paralysis. It was an event that changed the rules of the game.

In contrast, the 2026 Iran-Israel conflict was an event within the existing regime. It moved prices, yes, but it did not alter the underlying structure of volatility. The market’s temperature rose, but the “thermostat” remained the same.

The final surprise: in crisis, simpler is better

A complementary analysis in the study yields a counterintuitive but valuable lesson. When the researchers allow the crisis data to choose its own optimal model, the result isn’t a more complex model — quite the opposite.

The model that best fits the critical period is sGARCH(1,1)-ARMA(1,1), a simpler version that eliminates the “leverage effect” (the asymmetry in how markets react to bad versus good news).

Why? Because in an acute crisis, all news is bad. Positive and negative shocks generate uncertainty equally. Asymmetry fades. And the model simplifies.

This finding has a powerful implication for risk managers: in times of maximum uncertainty, the simplest models may be the most robust.

Why this matters (and not just for economists)

For those working in risk management — at banks, investment funds, energy companies — this study is no academic curiosity. It’s a tool with direct practical implications:

1. It validates that GARCH models, calibrated on periods of prior tension, can continue to perform in new crises of the same type. This is critical for VaR (Value at Risk) backtesting and capital allocation.
2. It offers a criterion to distinguish between “normal” crises and “ruptures.” If a model calibrated on the past maintains its accuracy, the crisis is endogenous to the regime. If it collapses, we’re facing a structural change that demands a full recalibration.
3. It demonstrates that, under stress, parsimony is a virtue. The transition from eGARCH to sGARCH is an invitation not to overcomplicate models when robustness is most needed.

Beyond neural networks: why GARCH still matters

At a time when neural networks and deep learning dominate the conversation about prediction, the author makes a strong case for the continued relevance of GARCH models:

• Interpretability: In a GARCH model, every parameter has a clear meaning. Risk committees, auditors, and regulators (Basel III/IV, Solvency II) need to understand why a model predicts what it predicts. A black-box neural network doesn’t allow that.
• Stability: The study shows that the GARCH model remains stable across 9 distinct periods. Neural networks, with their tendency to overfit, can exhibit instability across short time windows.
• Modest data requirements: GARCH models converge with 500–1,000 observations. Neural networks often require thousands or tens of thousands of data points to be reliable. In energy markets, where high-frequency series can be limited, this is a distinct advantage.
• Theoretical grounding: GARCH models have foundations in financial theory (connections to stochastic variance processes, option pricing models). Neural networks, for now, lack that scaffolding.

The black swan that wasn’t

The closing line of the study captures its core insight:

“The conflict did not create a new risk dynamic; it simply activated the one that already existed, adapting its expression to the nature of the crisis regime.”

What appears to be a black swan — an unpredictable event, outside any known structure — may actually be the manifestation of a dynamic attractor that was already there, waiting. Volatility is not a chaotic, unknowable phenomenon. It is a structural property of the system, one that can be modeled, anticipated, and — above all — understood.

For investors, for risk managers, for energy policy makers, this is hopeful news: uncertainty has structure, and that structure can be learned.

About the study

”Analysis of Brent Crude Oil volatility structure: Can an ARMA-GARCH model calibrated in previous conflicts explain the 2026 Iran-USA-Israel crisis?* Author: Luis Adrián Martínez Pérez Date: March 2026 Full article: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6462041