Counterexamples for FX Options Interpolations - Part II
Jherek Healy · 2025
Read Counterexamples for FX Options Interpolations - Part II by Jherek Healy (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-19625.txt.
Introduction
Following his catalogue of interpolation failures, Jherek Healy turns to what those choices do to the numbers traders actually quote. Different exact interpolations of the same broker quotes, he shows, can produce different implied volatilities at the standard benchmark points, the 10-delta and 25-delta puts and calls of vanilla FX options.
Because implied volatility is the common language of the FX options market, two desks fitting identical market data with different interpolation schemes can end up quoting different volatilities for the same contracts. The follow-up makes that spread concrete. Read with the first part, it argues that interpolation is not a neutral technical step but a modeling choice with visible pricing consequences.
Catalog record
| CREATOR | Jherek Healy |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-19625
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Counterexamples for FX Options Interpolations - Part II Jherek Healy Abstract This follow-up article analyzes the impact of foreign exchange option interpolation on the vanilla option implied volatilities. In particular different exact interpolations of broker quotes may lead to different implied volatilities at the 10 Δ \Delta and 25 Δ \Delta Puts and Calls. keywords fx options; volatility; arbitrage; interpolation † † firstpage: 1 † † issue: 1 † † articlenumber: 1 † † year: 2024 † † copyright-year: 2024 † † authornames: Jherek Healy † † address: [email protected] \history 1 Introduction Market prices of Foreign exchange (FX) options are typically quoted as a sparse set volatilities per option maturity. Those correspond to the volatilities of an at-the-money (ATM) option, 25 \Delta and 10 \Delta risk-reversals and butterflies. Depending on the currency pair and the maturity, there is much variation in what ATM exactly means: is it at the money straddle? with or without premium? spot or forward ? There are also two conventions for the risk-reversal (RR) and butterfly (BF): the simple smile convention where \displaystyle\sigma_{\textmd{call}}=\sigma_{\textmd{ATM}}+\sigma_{\textmd{BF}}+\frac{1}{2}\sigma_{\textmd{RR}}\,,\quad\sigma_{\textmd{put}}=\sigma_{\textmd{ATM}}+\sigma_{\textmd{BF}}-\frac{1}{2}\sigma_{\textmd{RR}}\,, (1) and the more involved broker convention which requires a numerical solver. In this article, we focus on the latter. In the broker convention, the ATM quote stays simple. Latin America currency pairs use the ATM forward convention where the vanilla ATM strike K is such that K=F(0,T) where F(0,T) is the forward to maturity (e.g. USD/BRL, USD/MXN). For other currency pairs, pips or percent \Delta may be used. The choice depends on the general convention (also applicable to the other option quotes beside ATM) of the currency pair and is based on the strike of a delta neutral straddle. If the premium currency is the same as the primary currency, the convention is to use a percentage 1 1 1 Also known as \Delta with premium. \Delta (e.g. EUR/CHF, EUR/TRY), and otherwise the pips \Delta (e.g. EUR/USD, AUD/USD). When the convention is different from the ATM forward convention, the percentage or pips spot \Delta may 2 2 2 This is not always true, since the 2008 crisis and the associated low levels of liquidity in short-term interest rate products, forward \Delta tends to be also used. be used for maturities up to and including one year when the currency pair only contains currencies from OECD economies (e.g. USD, EUR, JPY, GBP, AUD, NZD, CAD, CHF, NOK, SEK, DKK). For all other cases, the pips or percentage forward \Delta is used. Those read \Delta_{F,\textmd{pips}}=\eta\Phi(\eta d_{1}(K,\sigma))\,,\quad\Delta_{F,\textmd{pct}}=\frac{K}{F(0,T)}\eta\Phi(\eta d_{2}(K,\sigma))\,, (2) where d_{1}(K,\sigma)=\frac{1}{\sigma\sqrt{T}}\ln\frac{F(0,T)}{K}+\frac{1}{2}\sigma\sqrt{T} , d_{2}=d_{1}-\sigma\sqrt{T} and \eta=\pm 1 for respectively a call and a put option. We refer the reader to 2 ; 9 for exhaustive (but key) details on the various conventions. The delta neutral straddle strike is determined by the following equation: \Delta(1,K,T,\sigma_{\textmd{ATM}})+\Delta(-1,K,T,\sigma_{\textmd{ATM}})=0\,. (3) The strike is independent of the spot or forward delta convention as the discounting term will factor out. Equation 3 leads to explicit solutions for the ATM delta neutral strike price: K_{\textmd{DNS},\textmd{pips}}=F(0,T)e^{\frac{1}{2}\sigma_{\textmd{ATM}}^{2}T}\,,\quad K_{\textmd{DNS},\textmd{pct}}=F(0,T)e^{-\frac{1}{2}\sigma_{\textmd{ATM}}^{2}T}\,, (4) where \sigma_{\textmd{ATM}} is the given ATM market volatility. The 10- and 25- Put and Call \Delta are not directly given but implied from the quotes of markets strangles and risk-reversals. The price of a market strangle reads \displaystyle V_{\textmd{MS},x\Delta} \displaystyle=V(-1,K_{\textmd{MS},\textmd{P},x\Delta},T,\sigma_{\textmd{ATM}}+\bar{\sigma}_{\textmd{BF},x\Delta}) \displaystyle+V(1,K_{\textmd{MS},\textmd{C},x\Delta},T,\sigma_{\textmd{ATM}}+\bar{\sigma}_{\textmd{BF},x\Delta})\,, (5) where V(\pm 1,K,T,\sigma) is the price of a vanilla European call (resp. put) option of strike K and maturity T and volatilty \sigma and x\in\{10,25\} . The same volatility (our market quote \bar{\sigma}_{\textmd{BF},x\Delta}) ) is used for both call and put in the broker convention, at different strikes K_{\textmd{MS},\textmd{P},x\Delta} and K_{\textmd{MS},\textmd{C},x\Delta} . Those strikes are defined such that the following holds: \displaystyle\Delta(-1,K_{\textmd{MS},\textmd{P},x\Delta},T,\sigma_{\textmd{ATM}}+\bar{\sigma}_{\textmd{BF},x\Delta}) \displaystyle=-x\%\,, \displaystyle\Delta(1,K_{\textmd{MS},\textmd{C},x\Delta},T,\sigma_{\textmd{ATM}}+\bar{\sigma}_{\textmd{BF},x\Delta}) \displaystyle=+x\%\,. (6) While Equation 6 gives the strikes of the market strangle, we do not know the European option volatilities for those strikes, we only know the overall strangle strategy price through \sigma_{\textmd{ATM}}+\sigma_{\textmd{BF},x\Delta} . The risk-reversal quotes are defined such that \displaystyle\sigma_{\textmd{RR},x\Delta}=\sigma_{\textmd{C},x\Delta}-\sigma_{\textmd{P},x\Delta}\,. (7) In particular, the quote is negative if \sigma_{\textmd{C},x\Delta}<\sigma_{\textmd{P},x\Delta} . Finding the volatilities \sigma_{\textmd{C},x\Delta} and \sigma_{\textmd{P},x\Delta} (and the corresponding vanilla option strikes) necessitates a continuous representation of volatilities across strikes, as the vanilla option strikes do not correspond to the market strangle strikes. Let \sigma(K) be such a representation, we have then the following system to solve: \displaystyle\sigma(K_{\textmd{ATM}}) \displaystyle=\sigma_{\textmd{ATM}}\,, (8) \displaystyle\Delta(1,K_{\textmd{C},x\Delta},T,\sigma(K_{\textmd{C},x\Delta})) \displaystyle=+x\Delta\%\,, (9) \displaystyle\Delta(-1,K_{\textmd{P},x\Delta},T,\sigma(K_{\textmd{P},x\Delta})) \displaystyle=-x\Delta\%\,, (10) \displaystyle V(-1,K_{\textmd{MS},\textmd{P},x\Delta},T,\sigma(K_{\textmd{MS},\textmd{P},x\Delta}))+V(1,K_{\textmd{MS},\textmd{C},x\Delta},T,\sigma(K_{\textmd{MS},\textmd{C},x\Delta})) \displaystyle=V_{\textmd{MS},x\Delta}\,, (11) \displaystyle\sigma(K_{\textmd{C},x\Delta})-\sigma(K_{\textmd{P},x\Delta}) \displaystyle=\sigma_{\textmd{RR},x\Delta}\,. (12) Due to the difference between K_{\textmd{MS},\textmd{P},x\Delta} and K_{\textmd{P},x\Delta} , there are nine different strikes involved, not just five. The strikes K_{\textmd{C},x\Delta},K_{\textmd{P},x\Delta} may be solved exactly using Equations 9 and 10 for given volatilities. We then have 5 remaining equations to fit the volatility representation \sigma(K) . There are several ways to perform the fit. 2 describes how to solve the system assuming a single unknown, the smile butterfly volatility \sigma_{\textmd{BF},25\Delta} , and does not try to fit to 10\Delta quotes. We adapt the algorithm to handle 10\Delta and 25\Delta quotes together: 1. Choose an initial guess for the smile butterfly volatilities \{\sigma_{\textmd{BF},25\Delta},\sigma_{\textmd{BF},10\Delta}\} , typically the market butterfly quotes \{\bar{\sigma}_{\textmd{BF},25\Delta},\bar{\sigma}_{\textmd{BF},10\Delta}\} (which corresponds to slightly different strikes). 2. Use Equation 1 to obtain \sigma_{\textmd{C},25\Delta},\sigma_{\textmd{P},25\Delta},\sigma_{\textmd{C},10\Delta},\sigma_{\textmd{P},10\Delta} from the risk reversal quotes. Equation 1 is exact when the vanilla volatilities correspond to the smile implied \Delta s, which is the case here. 3. Calculate the vanilla option strikes by solving Equations 9 and 10 . 4. Find parameters of \sigma(K) using a least-squares-optimizer on the volatilities \sigma_{\textmd{ATM}},\sigma_{\textmd{C},25\Delta},\sigma_{\textmd{P},25\Delta},\sigma_{\textmd{C},10\Delta},\sigma_{\textmd{P},10\Delta} . 5. Price up the market strangle using \sigma(K) . The least-squares error in model price against the market price (Equation 11 ) defines the objective. A Gauss-Newton optimizer 6 may be used to find the least-squares solution in option prices. The unknown \sigma_{\textmd{BF},25\Delta} may be negative, it is perfectly acceptable ( 12 ) . The bound constraint is for the vanilla volatilities given by Equation 1 to be positive. This presupposes that the representation \sigma(K) fits well to the ATM, 10 \Delta and 25 \Delta strikes: the smile fitting procedure to vanillas is somewhat decoupled from the fitting to market strangles. For non-exact interpolations, it may be particularly relevant to add the error in ATM and RR volatilities, corresponding to Equations 8 and 12 , to the objective in step (5). The output has then a dimension of five. The intent is to recouple the vanilla fit with the market strangle fit. There is however a caveat: the market strangle error is minimized in terms of price, while the ATM and RR errors are minimized in terms of volatility. In order to obtain a similar error scale, an inverse vega weight w_{\textmd{MS},x\Delta} must be used in the market strangle model vs. market price error: w_{\textmd{MS},x\Delta}=\frac{1}{\left[\phi\left(d_{1}(K_{\textmd{MS},\textmd{P}},\sigma_{\textmd{ATM}}+\bar{\sigma}_{\textmd{BF},x\Delta})\right)+\phi\left(d_{1}(K_{\textmd{MS},\textmd{C}},\sigma_{\textmd{ATM}}+\bar{\sigma}_{\textmd{BF},x\Delta})\right)\right]\sqrt{T}F(0,T)}\,. (13) An additional subtlety is how and when to solve for the strikes of the \Delta based quotes. Indeed, the two risk-reversals define vanilla \Delta -based quotes according to the currency pair and maturity convention. In Step 3, 2 suggests that the smile parameterization should be used, and thus Step 4 should actually be merged with Step 3, such that we solve for the strike, using the smile. This makes a difference if the objective of dimension 2, and we will adopt this approach in our numerical examples, but it should not matter much if the objective is of dimension 5 with an error in RR volatilities computed