SABR was designed to fit equity vol smiles. Hagan, Kumar, Lesniewski, and Woodward published the model in 2002 specifically for interest rate swaptions. It became the dominant smile model in rates markets within three years. Then practitioners started applying it to commodity options — and that is where the problems begin.
What SABR Assumes That Commodity Markets Violate
The original SABR formulation rests on three implicit assumptions that hold reasonably well for equity and rate derivatives. First, the forward price follows a continuous diffusion. Second, the vol smile is approximately symmetric around the at-the-money strike. Third, the correlation between forward price and stochastic vol parameter beta does not exhibit strong term-structure effects.
Commodity markets break all three. Natural gas forward prices gap at expiry when storage limits bind. Crude oil smiles skew strongly toward downside puts during supply gluts and toward upside calls during inventory squeezes. Agricultural options show smile shapes that change character completely across seasons — corn options before harvest look nothing like corn options in February.
When you feed a SABR calibration routine crude oil data from Q4 2018 — a quarter where Brent dropped from $86 to $53 — the fitted alpha, beta, rho, and nu parameters either diverge or produce negative densities at the wings. The model produces numbers, but those numbers are meaningless.
The Specific Problem with Beta in Energy Markets
Beta in SABR controls the backbone — the relationship between the at-the-money vol level and the forward price as the forward moves. For equities, beta around 0.5 (log-normal backbone) fits reasonably. For energy commodities, the correct beta depends on the commodity and the contract tenor.
Front-month WTI crude options exhibit near-normal backbone dynamics when prices are above $60. Below $30 — as happened in 2020 — the backbone flips toward log-normal because absolute dollar moves dominate. A single beta parameter cannot capture both regimes. If your risk system uses SABR with a fixed beta for crude options and crude falls below $40, your delta calculations will be systematically wrong by several percentage points.
In our testing on CME WTI options data from 2015 to 2024, a fixed-beta SABR model produced delta estimates that differed from realized delta (computed from next-day P&L attribution) by an average of 4.2% for front-month options when prices were below $40 per barrel. That is not a rounding error on a book with 500,000 barrels of delta.
Agricultural Options: The Seasonal Smile Problem
Corn and soybean options have smile shapes that rotate with crop cycles. Pre-harvest options — July corn, November soybean — carry put skew because traders hedge against poor yields. Post-harvest options carry call skew because storage costs and export demand can push prices higher unexpectedly.
SABR's rho parameter captures the correlation between forward price and vol, which drives smile asymmetry. But a single rho calibrated to pre-harvest data will misfit post-harvest data by design. The seasonal rotation in rho is not small — it can shift from -0.35 pre-harvest to +0.15 post-harvest for nearby soybean options.
Practitioners who use a static SABR calibration updated weekly will consistently misprice seasonal roll risk. They may not notice in calm markets where the misfits are small relative to bid-ask spreads. They will notice during weather events, when implied vol moves quickly and accurate smile shape matters for delta and gamma calculations.
What We Built Instead: The Allasso Seasonal Surface Model
Allasso uses a two-layer approach for commodity smile construction. The first layer is a modified stochastic local vol (SLV) model where the local vol component is parameterized by commodity-specific seasonality functions. We fit separate seasonal curves for each major commodity: energy contracts use a four-parameter curve tied to Northern Hemisphere demand seasonality, agricultural contracts use an eight-parameter curve tied to planting and harvest calendar dates.
The second layer is an arbitrage-free interpolation scheme that enforces no-butterfly-spread arbitrage across strikes and no-calendar-spread arbitrage across tenors simultaneously. Standard SABR interpolation handles these separately, which allows inconsistent surfaces when liquid strikes are sparse — a common situation in base metals and softs markets where only a few strikes trade actively on any given day.
The result is a surface that calibrates cleanly to thin market data without producing negative densities or butterfly arbitrage at the wings. For natural gas options — arguably the most difficult surface to fit due to delivery risk spikes at month-end — our model produces surfaces that price liquid strikes within 0.3 vol points of mid-market on average, compared to 1.1 vol points for standard SABR.
When SABR Still Works Fine
SABR is not useless for commodity options. For liquid, near-dated options on major exchange-traded commodities — front three months on WTI, Brent, gold, and copper — SABR with careful beta management and frequent recalibration produces reasonable results. The problems emerge at the wings, in thin markets, over long tenors, and during stress periods.
If your book is primarily at-the-money vanilla options on a handful of liquid underlyings, and you recalibrate daily, SABR will serve you adequately. If your book includes exotic payoffs, wide-strike ranges, illiquid markets, or multi-year tenor options, the model's limitations will cost you money in delta hedging and will show up as unexplained P&L.
The test we recommend: take your SABR surface for any commodity, compute the risk-neutral density implied by the surface, and check whether it integrates to one. If the density has mass outside your strike range that is being ignored, or if it goes negative anywhere, your surface is arbitrageable. Fix the model, not the parameters.
Practical Implications for Risk Management
Delta calculations from a mispecified smile model translate directly into P&L attribution errors. When your delta is wrong by 4%, and you are hedging 10,000 barrels of WTI daily, you are leaving $4,000–$8,000 of unhedged delta on the table every day depending on crude price moves. Over a quarter, that is $360,000–$720,000 in uncompensated risk exposure — which will look like vol-of-vol risk in your P&L but is actually a model error.
Gamma is even more sensitive to smile shape. The gamma surface is the second derivative of option price with respect to spot, and it depends heavily on how the smile interpolates between strikes. A SABR model that fits three liquid strikes cleanly may interpolate the smile incorrectly between those strikes, producing gamma spikes or troughs at illiquid levels. Those gamma errors appear when spot moves through those levels — exactly when you need accurate risk numbers most.
How Allasso Lets You Compare Models Live
Rather than forcing a single model choice, Allasso lets traders price any position under Black-76, standard SABR, and the Allasso seasonal model simultaneously. The platform displays the Greeks from each model side by side, flags divergences above a configurable threshold, and logs every calibration in an audit trail.
The typical workflow on an energy desk: run the Allasso seasonal model as the primary pricing engine, use Black-76 as the internal benchmark, and inspect divergences above 5 vol points or 3% delta difference. Positions where the models disagree significantly are flagged for human review before hedging. This model risk overlay costs almost nothing in setup and has prevented several significant hedging errors in early customer deployments.
For desks moving from SABR to a more complete surface model, the transition does not need to happen overnight. Allasso runs both models in parallel, allowing you to validate the new model against your existing SABR output on historical data before switching over live pricing. Contact us if you want to run a comparison on your own book.