Derive the BAW American-option shortcut, its quadratic exponent and exercise boundary, then compare a checked put example with refined binomial trees.
Barone-Adesi–Whaley approximation
The Barone-Adesi–Whaley approximation (BAW) estimates a vanilla American option price by adding an approximate early-exercise premium to the corresponding European option price. It replaces a time-dependent pricing problem with a power function and a numerical solve for the exercise boundary. That makes it useful for fast model comparisons and repeated valuations when a full tree or differential-equation solver would be more expensive.
The method is often called a quadratic approximation. The quadratic is an equation for the power exponent; the option price is not approximated by a quadratic polynomial in spot. Nor does “analytic approximation” mean a globally exact formula: the critical price still has to be found iteratively. Barone-Adesi and Whaley's original paper, published in 1987, develops the method for constant proportional carry.
Start with the premium for being able to exercise
Consider a put with positive spot S, strikeK, remaining time τ>0, constant annual volatility σ>0, positive continuously compounded rate r, and constant continuous dividend yieldq≥0. The stock follows the Black–Scholes model, with risk-neutral drift b=r−q. Exercise pays immediately; there are no transaction costs or discrete cash dividends.
Write the American value as
PA(S,τ)=pE(S,τ)+ϵ(S,τ).
The European component pE is known. The unknown ϵ measures the value of choosing when to exercise. Inside the continuation region, subtraction of the two pricing equations gives
∂τ∂ϵ=21σ2S2ϵSS+bSϵS−rϵ.
For example, selling a put for its market value and exercising it for its intrinsic value are different actions. This equation concerns the model value of the contractual exercise choice.
The approximation happens after changing the time variable
Set
h=1−e−rτ,M=σ22r,N=σ22b,ϵ=hf(S,h).
Substituting and collecting terms gives
S2fSS+NSfS−hMf−M(1−h)fh=0.
BAW drops the final term, which contains the derivative of the transformed premium with respect to transformed time. It does not simply set the entire original time derivative ϵτ to zero. Some maturity dependence survives through h. This is the approximation identified in Section I.B.1 of the original derivation.
At a fixed maturity, substitute a trial solution proportional to Sλ into the reduced equation:
λ2+(N−1)λ−hM=0.
For a put, the premium must decay at large spot, so retain the negative root
λ−=2−(N−1)−(N−1)2+4M/h<0.
This choice determines a shape, but leaves its scale and the critical spot unknown. The remaining conditions come from the economics of exercise.
Match both the value and the slope
At the approximate critical spot S∗, the continuation expression must meet the exercise proceeds and have the same slope:
P(S∗,τ)=K−S∗,PS(S∗,τ)=−1.
These are value matching and smooth pasting. The slope condition determines the premium coefficient
where Φ is the standard normal cumulative distribution function. Value matching then gives the scalar equation to solve:
K−S∗=pE(S∗,τ)+A.
After solving it, use the correct side of the boundary:
P(S,τ)={K−S,pE(S,τ)+A(S/S∗)λ−,S≤S∗,S>S∗.
Below the boundary, exercise is the approximation's decision. Extending the continuation power function there would price a different expression. Above it, waiting preserves enough optionality to outweigh immediate exercise. The QuantLib implementation uses this same negative-root, coefficient and piecewise structure. Root labels vary across references; here λ− always means the negative put exponent.
A one-year put, with a separate numerical comparison
Use fictional USD-per-share inputs: S=K=100, r=5%, q=0, σ=20%, and τ=1 year. All rates are annual and continuously compounded.
Solving the boundary equation gives
λ−≈−7.948814,S∗≈81.695434,A≈2.614140.
At spot 100 the put is in the continuation region:
Thus the approximate early-exercise premium is **0.524089pershare∗∗.ThecoefficientAisthepremiumattheboundary,notthepremiumatspot100.Atspot80,thissameapproximationinsteadreturnsimmediateexercisevalue100-80=20$.
Synthetic one-year put price against spot in USD per share. The BAW curve meets exercise value at spot 81.6954; the exercise region lies below this boundary. At spot 100 BAW is 6.097615 and the European put is 5.573526. A 4000-step CRR tree gives 6.090181.
Original constant-parameter model illustration. BAW imposes tangency at its approximate boundary. The sidebar compares separately calculated CRR tree values; none is labelled the exact continuous-time American price.
A Cox–Ross–Rubinstein tree, checking exercise at each node, gives the following prices for the same contract:
Tree steps
American put value, USD per share
500
6.088810
1,000
6.089595
2,000
6.089990
4,000
6.090181
BAW is about $0.007435, or 0.122%, above the 4,000-step result. This is a difference from a numerical comparator, not a certified error bound against an exact answer. Refining a tree tests its discretisation; tightening the BAW boundary solve only solves the approximate equation more accurately. It cannot recover the transformed-time term that was dropped.
Where the shortcut needs care
Value matching and smooth pasting are local boundary conditions. Satisfying both does not prove that the approximate surface solves the full American pricing problem. Check price bounds, exercise-region behaviour and sensitivity to inputs, then compare with a suitably refined tree or finite-difference calculation over the actual parameter range. A small price discrepancy also does not establish accurate Greeks, which involve derivatives.
The formulas above deliberately assume r>0. At r=0, M/h appears as 0/0; its fixed-maturity limit is 2/(σ2τ). That algebraic limit alone does not establish a well-posed boundary solve in every zero-rate regime. QuantLib includes a near-zero-rate limiting coefficient, and its critical-price routine explicitly rejects risk-free discount factors above one, corresponding to negative effective rates. A negative-rate exercise geometry needs a separately validated method. Expiry, zero volatility, zero spot or zero strike also need their own limiting or deterministic treatment rather than direct substitution into these equations. Current implementation, checked 29 October 2026.
Continuous yield is a specific cash-flow model. Replacing scheduled cash dividends with a single yield changes both stock dynamics and exercise incentives; the displayed calculation does not price that dividend schedule. For calls under the stated positive-rate assumptions, a non-dividend stock has no early-exercise premium; dividend-paying calls require the positive-root version and a boundary on the high-spot side. The put formula cannot simply be reused unchanged.
For the underlying exercise logic, American Options and Early Exercise develops stopping, continuation and free-boundary conditions; its preview is public and the full lesson requires membership. The public Options Pricing tool can reproduce the European baseline: select a put and enter spot and strike 100, 365 days, volatility 20%, rate 5% and dividend yield 0%. It calculates European Black–Scholes values and does not calculate the BAW boundary or American premium.
