# SABR Model: Alpha, Beta, Rho, Nu, Calibration, and Model Risk

Understand the SABR stochastic-volatility model, how alpha, beta, rho, and nu shape an implied-volatility smile, and how to calibrate and validate Hagan approximations.

Canonical: https://wiki.fcontext.com/options/sabr-model/
Fact checked: 2026-07-22

> For educational purposes only; not investment advice.

<a id="answer"></a>

## Direct answer

The **SABR model** is a stochastic-volatility model in which a forward `F_t` and its volatility level `α_t` move randomly and their shocks are correlated. It is widely used to parameterize an implied-volatility smile with a small, interpretable parameter set and to generate smile-consistent prices and sensitivities.

SABR is not one universal volatility surface. It is commonly calibrated separately by expiration or rate-option expiry/tenor, using a Black, normal, or shifted quotation convention. The familiar Hagan implied-volatility formula is an asymptotic approximation to the model, not an exact price and not automatically arbitrage-free at all strikes and maturities.

<a id="mechanism"></a>

## Dynamics and parameter roles

Under a pricing measure, the classic forward SABR dynamics are:

`dF_t = α_t F_t^β dW_t¹`

`dα_t = ν α_t dW_t²`

`d⟨W¹,W²⟩_t = ρ dt`

with current `α > 0`, commonly `0 ≤ β ≤ 1`, `ν ≥ 0`, and `−1 < ρ < 1`.

- **Alpha (`α`)** sets the current volatility scale, but is not generally equal to ATM implied volatility.
- **Beta (`β`)** controls elasticity: `β = 1` is lognormal-like, `β = 0` is normal-like, and intermediate values create CEV scaling.
- **Rho (`ρ`)** correlates forward and volatility shocks and strongly influences smile asymmetry.
- **Nu (`ν`)** is volatility of volatility and commonly increases smile curvature as it rises.

These interpretations are not independent. Alpha and Beta can compensate around ATM; Rho and Nu can compensate across a limited strike range. Many implementations fix or tightly constrain Beta, then calibrate Alpha, Rho, and Nu. A low fitting error does not prove parameter identification or realistic future smile dynamics.

For Black-style SABR, a leading ATM intuition is:

`ATM Black IV ≈ α / F^(1−β)`

with maturity corrections involving all parameters. Normal SABR is often used where rates or forwards can be near or below zero. Shifted SABR applies a displacement to `F` and `K`; that shift changes dynamics and tails and must be governed as a model parameter.

<a id="example"></a>

## Parameter-scale example

Let forward `F = 100`, maturity `T = 1 year`, fixed `β = 0.50`, and `α = 2.00`. The leading ATM approximation is:

`ATM IV ≈ 2.00 / 100^(1−0.50) = 2.00 / 10 = 20.00%`

If the forward falls to `81` while Alpha is held at `2.00`, the same approximation becomes:

`2.00 / 81^0.50 = 2.00 / 9 = 22.22%`

Thus unchanged Alpha does not imply unchanged ATM IV when `β < 1`. Conversely, comparing Alpha across assets with different units or Beta values is generally meaningless.

Now compare two illustrative configurations with the same `F`, `T`, `β`, and `α`:

| Set | `ρ` | `ν` | Typical qualitative effect |
| --- | ---: | ---: | --- |
| A | `−0.20` | `0.30` | milder negative skew and curvature |
| B | `−0.70` | `0.80` | steeper downside skew and stronger curvature |

That description is conditional, not a numerical price guarantee. A valid comparison must evaluate the selected Black/normal formula at actual strikes, convert IV errors into price errors, and inspect extrapolation. Parameters B can fit observed nodes yet produce unstable or implausible far wings.

<a id="risks"></a>

## Calibration and validation checklist

- Fix underlying, forward construction, discounting, expiry, tenor, settlement, and quote convention.
- Clean same-time Bid/Ask quotes and retain liquidity weights and exclusion reasons.
- Decide explicitly among Black, normal, and shifted SABR; never mix their volatility units.
- Fix or bound Beta with documented rationale and repeat valuation under alternative Beta values.
- Enforce `α > 0`, `ν ≥ 0`, and a safe interior bound for Rho.
- Use multiple initial guesses and inspect local minima, parameter covariance, and day-to-day jumps.
- Minimize price or Vega-weighted errors as appropriate, not unweighted IV error by habit.
- Report fit separately for ATM, downside, upside, and Bid/Ask coverage.
- Perform leave-one-out or out-of-sample checks instead of relying only on in-sample residuals.
- Convert the fitted smile back to option prices and check strike monotonicity and convexity.
- Inspect calendar consistency across separately calibrated expirations.
- Compare Hagan approximation prices and Greeks with a more accurate numerical method at long maturities, high Nu, extreme Rho, and distant strikes.
- Stress Alpha, Beta, Rho, Nu, forward, shift, rates, and correlations jointly.
- Preserve market data, versioned formula, objective function, bounds, solver, initial values, and calibration timestamp.

<a id="misconceptions"></a>

## Common misconceptions

- “SABR is an exact closed-form pricing model.” The popular formula is an asymptotic implied-volatility approximation.
- “Alpha is ATM IV.” It also depends on forward scale, Beta, maturity, Rho, and Nu corrections.
- “Rho is historical price/volatility correlation.” It is a risk-neutral model parameter calibrated to option prices.
- “Nu controls only the wings.” Parameters interact, and changing Nu can affect ATM and hedge sensitivities.
- “Beta can always be estimated freely.” Alpha-Beta collinearity often makes the fit unstable.
- “A perfect smile fit proves good hedges.” Different dynamics can fit the same vanilla snapshot and hedge exotics differently.
- “Normal and Black volatilities are interchangeable percentages.” They use different units and pricing conventions.
- “Shifted SABR solves negative rates without cost.” The displacement is an additional structural assumption affecting tails.
- “Positive fitted IV means no arbitrage.” The resulting price curve can still violate convexity or calendar conditions.
- “One SABR fit describes the entire surface.” Independent slices can be inconsistent across expirations and tenors.

<a id="related"></a>

## Related topics

- [Dupire local volatility](/options/dupire-local-volatility/)
- [Local volatility model](/options/local-volatility-model/)
- [Implied volatility](/options/implied-volatility/)
- [Option model risk](/options/option-model-risk/)

<a id="sources"></a>

## Sources

- [Hagan et al.: Managing Smile Risk](https://www.next-finance.net/IMG/pdf/pdf_SABR.pdf)
- [Obloj: Fine-Tune Your Smile—Correction to Hagan et al.](https://arxiv.org/abs/0708.0998)
- [Fukasawa: On Asymptotically Arbitrage-Free Approximations of Implied Volatility](https://arxiv.org/abs/2201.02752)
- [Choi and Seo: Option Pricing under the Normal SABR Model with Gaussian Quadratures](https://arxiv.org/abs/2301.02797)