Small Group Tutorials

Here to help students catch up, keep up, and move ahead. Book a consultation here.

How Breeden–Litzenberger Algorithms Extract Risk-Neutral Densities from Option Prices: Strike Derivatives, Butterfly Spreads, Smoothing, Arbitrage Checks and Tail Failure

Reader question: An option chain gives prices at many strikes. Can those prices be turned into a complete probability distribution for the underlying price at expiry?

Under no-arbitrage conditions, European option prices contain enough cross-sectional information to recover a risk-neutral state-price density. The Breeden–Litzenberger result makes the connection unusually direct: the second derivative of a call price with respect to strike is proportional to the risk-neutral density at that strike.

This article owns the option-price-to-risk-neutral-density extraction problem: same-expiry option quotes + discounting/forward information + a smooth arbitrage-consistent strike curve → digital probabilities, state-price density, implied distributional moments and diagnostics. It does not own implied-volatility solving, SVI calibration, local-volatility dynamics or physical-probability forecasting.

This is public mathematical and computational education. It is not personalised financial advice, an options-trading recommendation or a claim that risk-neutral probabilities are literal forecasts of real-world outcomes.

1. Start from the European call payoff

For deterministic discounting, a European call with strike K and maturity T has price:

C(K,T) = D(0,T) EQ[(ST − K)+],

where:

  • D(0,T) is the discount factor;
  • Q is the risk-neutral pricing measure;
  • ST is the terminal underlying price.

If ST has risk-neutral density fQ(s), then:

C(K,T) = D(0,T) ∫K∞ (s−K) fQ(s) ds.

The strike K appears both in the payoff and in the lower integration boundary. Differentiating with respect to K reveals the distribution.

2. First strike derivative: a digital tail probability

Differentiate once:

∂C/∂K = −D(0,T) Q(ST > K),

under the usual smoothness/continuity conditions.

Therefore:

−(1/D) ∂C/∂K

is the risk-neutral probability that the terminal price exceeds K.

This has a simple trading interpretation. A narrow call spread approximates a digital option: long a call at K and short a nearby call at K+ΔK. As ΔK becomes small, the payoff approaches an indicator of finishing above the strike.

3. Second strike derivative: the density

Differentiate again:

∂²C/∂K² = D(0,T) fQ(K).

So:

fQ(K) = [1/D(0,T)] ∂²C/∂K².

If D = e−rT under a constant continuously compounded rate r, this is commonly written:

fQ(K) = erT ∂²C/∂K².

This is the Breeden–Litzenberger extraction formula.

4. Butterfly spreads make the second derivative tangible

Consider equally spaced strikes K−h, K and K+h. The butterfly combination:

C(K−h) − 2C(K) + C(K+h)

divided by h² is a central finite-difference approximation to:

∂²C/∂K².

The terminal butterfly payoff is concentrated near K. As h shrinks, it behaves like a narrow state-contingent claim.

This is why call-price convexity in strike has a probability interpretation: an arbitrage-free butterfly cannot have negative value, so an arbitrage-free smooth density cannot be negative.

5. No-arbitrage shape restrictions come before differentiation

For a fixed maturity, a well-behaved European call-price curve should satisfy core static-arbitrage properties such as:

  • non-negativity;
  • decreasing in strike;
  • convex in strike;
  • appropriate lower/upper bounds;
  • put–call parity consistency when calls and puts are combined.

The density formula magnifies violations. A tiny local concavity in fitted call prices can become a negative density spike after taking a second derivative.

6. Why raw option quotes cannot simply be differentiated

Listed options are available only at discrete strikes. Quotes contain:

  • bid–ask spreads;
  • stale prices;
  • rounding;
  • different liquidity across strikes;
  • occasional data errors;
  • microstructure noise.

A second derivative is numerically sensitive to noise. Differentiating raw mid prices can therefore create alternating positive and negative spikes that are artifacts of the quote grid rather than genuine market-implied structure.

The practical algorithm needs a smoothing/interpolation layer.

7. A robust extraction pipeline

  1. Choose one underlying and one expiry.
  2. Collect calls/puts with reliable bid, ask, strike and contract metadata.
  3. Infer or obtain the forward and discount factor consistently.
  4. Use put–call parity to improve coverage or cross-check quotes.
  5. Remove or flag obvious arbitrage violations/stale quotes.
  6. Convert prices to implied volatility where appropriate.
  7. Fit a smooth, preferably static-arbitrage-aware strike/volatility curve.
  8. Convert the smooth curve back into call prices on a dense strike grid.
  9. Differentiate once for tail probabilities and twice for density.
  10. Check positivity, normalisation, forward consistency and quote repricing.
  11. Stress smoothing, grid spacing and tail extrapolation.

8. Why many implementations smooth implied volatility first

Direct call prices across strike have strong deterministic shape. Implied volatility often changes more gradually across moneyness, which can make interpolation numerically easier.

A common workflow is:

option quotes → implied vols → smooth smile → dense implied vols → option prices → strike derivatives.

But a smooth implied-volatility curve is not automatically arbitrage-free. The resulting call-price function must still be convex in strike.

See SVI volatility-surface algorithms for one important parameterisation and its arbitrage diagnostics.

9. Finite-difference density on an equally spaced grid

If a smooth call curve is sampled on strikes K−h, K, K+h:

fQ(K) ≈ [C(K−h) − 2C(K) + C(K+h)] / [D h²].

The approximation error decreases with a finer grid only if the underlying smooth curve is itself reliable.

Making h extremely small against a noisy/interpolated surface can make the estimate worse because differentiation amplifies fitting error.

10. Unequally spaced strikes require the correct formula

Real option strikes are not always equally spaced, particularly in wings or across contract changes.

Applying the equal-spacing formula to unequal strikes is a mechanical error. A correct nonuniform finite-difference formula, local polynomial derivative or analytic derivative of the fitted surface should be used.

11. Normalisation diagnostic

A valid terminal probability density should satisfy:

∫0∞ fQ(s) ds = 1

for a nonnegative underlying with no probability mass omitted.

In practice, the observed strike range is finite. A fitted density may integrate to 0.94 over quoted strikes because 6% of fitted probability lies in extrapolated tails.

The missing mass is not automatically an error—but it must be accounted for explicitly.

12. Mean/forward diagnostic

Under the appropriate risk-neutral setup, the discounted expected terminal underlying is linked to the forward price.

A density estimate should therefore reproduce the forward consistently:

EQ[ST] = ∫ s fQ(s) ds

with the corresponding carry/dividend conventions.

If the density integrates to one but its mean disagrees materially with the observed forward, the surface, discounting, dividend/carry treatment or tail extrapolation is inconsistent.

13. Inputs and outputs

Inputs can include:

  • option bid/ask/mid prices;
  • strikes and expiry;
  • underlying spot and forward;
  • discount factor/rate curve;
  • dividend or carry assumptions;
  • contract multipliers and settlement type;
  • quote-quality filters;
  • implied-volatility solver;
  • smoothing/interpolation method;
  • tail extrapolation rule;
  • differentiation grid.

Outputs can include:

  • smoothed call-price curve;
  • digital/tail probability curve;
  • risk-neutral density;
  • risk-neutral CDF;
  • implied mean, variance, skewness and kurtosis;
  • probabilities of user-defined terminal regions;
  • normalisation error;
  • forward-consistency error;
  • negative-density diagnostics;
  • quote-repricing residuals;
  • smoothing/tail sensitivity.

14. Risk-neutral probability is not physical probability

This is one of the most important interpretation boundaries.

Risk-neutral probabilities are the probabilities that make discounted traded asset prices martingales under the pricing measure. They combine market beliefs with risk prices.

States investors particularly dislike can receive high state prices even if their real-world probability is lower.

Therefore a statement such as “the option-implied distribution assigns 20% risk-neutral probability below X” does not mean investors literally believe there is a 20% real-world chance of that outcome.

The Bank of England explicitly distinguishes physical (P) distributions from risk-neutral (Q) distributions when interpreting option-implied densities.

15. Evidence polarity

Evidence for confidence includes:

  • fitted option prices lie within or close to credible bid–ask ranges;
  • call prices are monotone and convex in strike;
  • density is nonnegative;
  • density mass is approximately one after tails are included;
  • density mean matches the forward/carry relationship;
  • small changes in smoothing do not radically change main features;
  • features persist across nearby expiries or time snapshots where economically plausible;
  • butterfly prices implied by the density reprice observed butterflies sensibly.

Evidence against confidence includes:

  • negative density regions;
  • large quote repricing errors;
  • density mass materially different from one;
  • mean inconsistent with the forward;
  • spikes that move when one quote is removed;
  • multimodality appearing only under one spline tension or grid;
  • wing probabilities dominated by arbitrary extrapolation.

16. Counterexample: noisy quotes create negative probabilities

Suppose three adjacent call mids are 12.0, 9.8 and 7.5. A small quote error in the middle price can make the finite-difference butterfly negative.

The raw second derivative then implies negative density.

Negative probability is not an exotic market belief. It is evidence of inconsistent quotes, bid–ask noise or a bad interpolation.

Falsifier: check whether an arbitrage-aware fit removes the negative region while staying inside plausible quote intervals.

17. Counterexample: overfitting invents a second mode

A flexible spline can chase local quote noise and create a wavy second derivative. The resulting density can display two or three peaks that look like economically meaningful “market scenarios.”

Research on option-implied distributions warns that apparent multimodality can be an overfitting artifact.

Falsifier: vary smoothing strength and use bootstrap/quote perturbations. A mode that disappears under tiny defensible changes is not robust evidence.

18. Counterexample: sparse wings dominate crash probability

Deep out-of-the-money options can be illiquid or absent. Yet extreme left-tail probability is exactly what analysts often want.

If the lowest quoted strike is still far above the relevant crash region, the density beyond it comes from extrapolation assumptions rather than direct market prices.

Falsifier: report observed-strike support separately from extrapolated tail mass and compare multiple defensible tail rules.

19. Counterexample: stale puts and calls break parity

If a call and put with the same strike are stamped at different times during a fast market, put–call parity can appear violated.

Combining them without time alignment can corrupt the inferred forward and therefore the entire density.

Falsifier: reconcile timestamps, forward inputs and executable bid–ask ranges before treating parity residuals as structural.

20. Counterexample: wrong discounting or dividends shift the density

Even if option prices are clean, using the wrong discount factor or dividend/carry input can move the implied forward and alter density interpretation.

Falsifier: independently infer the forward from put–call parity and compare it with the external forward/carry curve.

21. Smoothing alternatives

Parametric implied-volatility fits: SVI and related forms can provide smooth derivatives and explicit static-arbitrage checks.

Cubic or smoothing splines: flexible, but unconstrained splines can violate convexity.

Local-polynomial/kernel methods: estimate derivatives non-parametrically but require bandwidth choices.

Constrained optimisation: fit state prices or call prices directly under positivity, monotonicity and convexity constraints.

Mixture distributions: impose a parametric density and fit option prices, trading flexibility for stability.

22. The density can reveal moments without claiming a forecast

Once fQ is available, risk-neutral moments can be calculated:

EQ[STm] = ∫ sm fQ(s) ds.

These moments summarise market pricing of dispersion, skew and tails. They remain risk-neutral moments, not physical expectations.

23. How this connects to existing Bukit Timah Tutor pages

Implied-volatility solver algorithms own the numerical inversion from option price to volatility.

SVI algorithms own smooth smile fitting and static-arbitrage diagnostics.

Dupire local-volatility algorithms use strike/maturity derivatives to infer a local volatility dynamic; the present page stops earlier at the one-expiry state-price density.

Black–Scholes supplies one parametric benchmark density, whereas Breeden–Litzenberger lets the cross-section of option prices reveal a non-lognormal risk-neutral shape.

24. Weak links

  • stale or crossed quotes;
  • inconsistent timestamps;
  • poor forward/dividend estimates;
  • wrong strike spacing in finite differences;
  • unconstrained smoothing creating butterfly arbitrage;
  • insufficient wing data;
  • overfitting creating false modes;
  • confusing density value with probability mass;
  • confusing risk-neutral with physical probability;
  • mixing expiries or settlement conventions.

25. What would falsify confidence?

Confidence should be withdrawn if the fitted call curve violates monotonicity or convexity; if the density is materially negative; if probability mass fails to normalise; if the density mean conflicts with the forward; if main features vanish under modest smoothing changes; if quote repricing is poor; or if tail conclusions are almost entirely extrapolation beyond traded strikes.

26. Verification and update triggers

Preserve the raw option snapshot, timestamps, bid/ask spreads, parity-implied forward, discount curve, dividend/carry assumptions, quote filters, smoothing model, fitted parameters, dense strike grid and derivative scheme.

Re-run or review when:

  • market liquidity changes materially;
  • new strikes become available;
  • the forward/carry curve changes;
  • surface-arbitrage diagnostics fail;
  • tail extrapolation contributes material mass;
  • the smoothing method or implementation changes;
  • an event causes the smile shape to change abruptly.

27. Primary and high-quality references

Educational boundary: Breeden–Litzenberger extraction reveals the distribution used by market prices under a pricing measure. It does not reveal an objective forecast free of risk premia, and a beautifully smooth density can still be wrong if the option data or interpolation assumptions are wrong.

Discover more from Bukit Timah Tutor

Subscribe now to keep reading and get access to the full archive.

Continue reading