Tepe Quant Research
Back to research
Systematic / Quant

Cross-Currency Option Hedging and Replication: EUR/TRY as a Worked Example

· ~29 min read fx-optionscorrelationcross-currencycegafrtbeurtry

An option on a product or cross rate — EUR/TRY = EUR/USD × USD/TRY — is fundamentally an option on the covariance of its two factors. Its value lives in the second moment via σ₃² = σ₁² + σ₂² + 2ρσ₁σ₂, so the dominant non-standard exposure is correlation risk (“cega”), with dV/dρ = Vega₃·(σ₁σ₂/σ₃). You cannot statically replicate max(S₁S₂−K,0) with vanillas on S₁ and S₂: you can match local Greeks at spot, but never the global payoff, and correlation P&L leaks even under dynamic hedging.

The three vanilla vol markets pin implied correlation exactly, which enables the “vol triangle” trade. For a product cross, the long-correlation leg is long the cross variance, short both leg variances — the inverse of the ratio rule. Because an FX cross is an exact algebraic product, the covariance replication is exact absent jumps and given replicable variance swaps — unlike multi-name equity dispersion, which is only an average-correlation approximation.

For fragile EM currencies like TRY the picture degrades: the cross-leg vanilla market is illiquid, the joint tail is fat and jump-driven (co-skewness), continuous-hedging assumptions break, and even the “clean” delta hedge is exposed to offshore funding squeezes. So residual correlation + co-skew + cross-gamma must largely be warehoused — and under FRTB that warehousing carries punitive NMRF/RRAO capital, which is the real reason desks quote wide. The risk is priced, not hedged.

Key findings

  1. Log-returns add, so the cross is a covariance instrument. For S₃ = S₁S₂, ln S₃ = ln S₁ + ln S₂, hence σ₃² = σ₁² + σ₂² + 2ρσ₁σ₂ (the law of cosines / “currency triangle”). The cross option’s price is monotonic in this combined variance and therefore directly exposed to ρ. This is the master identity from which everything follows.

  2. Correlation sensitivity is a rescaled vega. dV/dρ = (∂V/∂σ₃)(∂σ₃/∂ρ) = Vega₃·(σ₁σ₂/σ₃). The vega decomposition onto the legs is correlation-dependent: ∂σ₃/∂σ₁ = (σ₁ + ρσ₂)/σ₃. So how much leg-1 vanilla you need to hedge the cross’s vega itself moves with ρ.

  3. Static replication is impossible; gamma and cross-gamma are the same convexity in two coordinate systems. A book of vanillas on S₁ and S₂ separately spans only f(S₁)+g(S₂), never the product S₁S₂, because single-name options carry zero cross-gamma ∂²V/∂S₁∂S₂. And by the chain rule Γ₁₂ = Γ₃S₃ + Δ₃, so once the cross is delta- and gamma-hedged with cross options, cross-gamma vanishes too. The corollary is unforgiving: the only clean hedge for either is the cross (EUR/TRY) option itself, which is illiquid — so both are warehoused.

  4. The triangle pins implied correlation exactly — and the trade direction depends on product vs. ratio. ρ = (σ₃² − σ₁² − σ₂²)/(2σ₁σ₂) for a product cross. To go long correlation on a product cross you go long the cross variance swap and short both leg variance swaps; for a ratio cross the sign flips. Getting this backwards leaves a desk short the correlation it meant to be long.

  5. The joint smile ≠ sum of marginal smiles; co-skewness and jumps dominate the tail. The cross’s smile is generated by the joint risk-neutral density, carrying co-skewness/co-kurtosis that two same-delta marginal legs cannot reproduce. For managed / devaluation-prone EM currencies, jumps create fat joint tails and Black–Scholes continuous-hedging fails because a gap happens between rebalances.

Part A — Rigorous quantitative treatment

A.1 Options on a product vs. options on the factors — with measure consistency

Take three FX rates linked by triangulation: EUR/TRY = EUR/USD × USD/TRY. Write S₃ = S₁S₂ and price everything under a single common pricing measure — the TRY (domestic) risk-neutral measure Q^TRY. This matters: only two of the three rates are “natural” tradables under Q^TRY.

drift(ln S₁) = drift(ln S₃) − drift(ln S₂)
             = [(r_TRY − r_EUR) − ½σ₃²] − [(r_TRY − r_USD) − ½σ₂²]
             = (r_USD − r_EUR) − ½σ₁² − ρσ₁σ₂
⇒ μ₁ = r_USD − r_EUR − ρσ₁σ₂        (add back ½σ₁² for the arithmetic drift)

So EUR/USD picks up a −ρσ₁σ₂ correction when expressed under the TRY measure. This is the correct home of the −ρσσ term — a measure change for a non-domestic pair — and it is the same object that reappears in the third-currency quanto of §B.7. It is not what a natively quote-settled EUR/TRY vanilla carries.

Applying Itô to the additive logs yields the central variance identity:

σ₃² = σ₁² + σ₂² + 2ρσ₁σ₂.

For a ratio (S₃ = S₁/S₂, e.g. EUR/GBP from two USD pairs) the sign flips: σ₃² = σ₁² + σ₂² − 2ρσ₁σ₂. This is exactly the law of cosines, which is why a currency trio maps to a triangle: side lengths are vols, the cosine of each angle is a pairwise correlation (Walter & Lopez, FRBSF, “The Shape of Things in a Currency Trio”; Zerolis, “Triangulating Risk,” Risk 1996). A historical illustration: the USD/DEM/ITL triangle visibly deformed when the lira was ejected from the EMS in October 1992 — the same geometry that governs TRY today.

The structural consequence: an option on S₃ depends on ρ, but options on S₁ and S₂ individually do not. The product option’s value is a function of the combined variance σ₃², so it sits one moment higher than the marginals — a covariance/correlation instrument wearing the clothes of a vanilla.

The cross-asset generalization. This is one instance of the general “option on a function of several underlyings” class:

All share the same DNA: the payoff is non-linear in a combination of underlyings, so the second cross-moment enters the price, and that covariance is not spanned by single-name vanillas.

A.2 Full Greek exposure of a cross-currency option

Treat the EUR/TRY option as a vanilla on S₃ with vol σ₃, remembering σ₃ = σ₃(σ₁, σ₂, ρ). Standard (Garman–Kohlhagen) first-order Greeks apply in the S₃ representation:

CBRT policy rate 37% held at the 11 Jun 2026 meeting (third straight hold)
O/N lending 40% borrowing 35.5% — carry dominated by forward points

The interesting structure is in the decomposition and cross-Greeks, via σ₃² = σ₁²+σ₂²+2ρσ₁σ₂:

Where the value lives. Because V is, to leading order, a function of σ₃² — itself a quadratic form with ρ in the off-diagonal — the cross option’s content is a second-moment object. First-moment (delta/forward) hedging removes direction; the residual P&L is governed by realized vs implied covariance.

A.3 Why static replication of max(S₁S₂−K,0) with leg vanillas is impossible

The spanning argument. Single-underlying European payoffs are spanned by a vanilla continuum (Breeden–Litzenberger / Carr–Madan):

f(S₁) = f(κ) + f'(κ)(S₁−κ) + ∫ f''(K)(S₁−K)₊ dK.

A book of vanillas on S₁ plus a book on S₂ therefore spans exactly the additively separable payoffs f(S₁) + g(S₂). The product payoff max(S₁S₂−K,0) is not in this span: it contains the irreducible bilinear term S₁S₂, with ∂²/∂S₁∂S₂ ≠ 0, while every single-name option has zero cross-gamma. This is a statement about function spaces, not about how many options you buy. (Exact replication of the product needs two-dimensional “quadrant” payoffs (Kᵢ−Sⁱ)₊(Kⱼ−Sʲ)₊ over a 2-D strike grid — i.e. genuine two-asset options.)

What can and cannot be matched:

Why correlation P&L leaks even under dynamic hedging. Dynamically delta-hedging the cross, the hedged P&L over dt is, to second order:

dP&L ≈ ½[ Γ₁ S₁²(σ₁,r² − σ₁,i²)
        + Γ₂ S₂²(σ₂,r² − σ₂,i²)
        + 2 Γ₁₂ S₁S₂ (ρ_r σ₁,r σ₂,r − ρ_i σ₁,i σ₂,i) ] dt

The cross-gamma term multiplies realized minus implied covariance. Delta-hedging neutralizes the drift but not this term: if realized correlation differs from the ρ you priced, cross-gamma bleeds P&L every rebalance. You can only remove it by holding offsetting cross-gamma — another product/cross option or a covariance swap — never with leg vanillas or forwards. Hence correlation is the residual that dynamic hedging cannot kill. The desk reading: such two-underlying structures are toxic at extreme correlation, their vega and cross-gamma reverse sign as the trade evolves (forcing costly rebalancing), and the correlation leg is hard to offset because correlation-trading markets are illiquid.

A.4 Covariance/correlation swaps, variance swaps, and the volatility triangle

Variance swap building block. A variance swap pays realized minus strike variance. Under a continuous diffusion it is model-independently replicated by a static 1/K²-weighted strip of OTM options plus a dynamic stock position (the log contract; Demeterfi–Derman–Kamal–Zou, Goldman 1999 / Neuberger / Carr–Madan):

−ln(S_T/S₀) = −(S_T−S₀)/S₀
            + ∫ over [0, S₀]  (K−S_T)₊ / K² dK
            + ∫ over [S₀, ∞]  (S_T−K)₊ / K² dK

As Bossu notes, only variance has a cost-effective static replication; volatility swaps and VIX futures are model-dependent. Crucial caveat: this needs no jumps and a full strike continuum — both fail for TRY (jumps + truncated illiquid wings), so the TRY variance swap is only approximately replicable and degrades into a corridor variance swap bounded by available strikes.

Covariance swap via polarization — get the direction right. Realized covariance of two log-prices satisfies the polarization identity. The form — and therefore the replicating trade — depends on whether the cross is a product or a ratio:

This is the rigorous backbone of the “vol triangle” trade — net the directional vega across the three legs and what remains is a pure covariance/correlation position. The sign is not cosmetic: applying the ratio rule to EUR/TRY (a product) leaves a desk short the correlation it intended to be long.

Because an FX cross is an exact algebraic product (or ratio), the two-asset covariance replication is exact, absent jumps and given replicable variance swaps — in contrast to a multi-name equity index, whose basket variance involves N(N−1)/2 pairwise covariances and where dispersion delivers only an average-correlation approximation (σ_basket ≈ σ_avg·√ρ̄).

Correlation swaps and dispersion. A correlation swap pays realized average correlation minus strike. Bossu (2005/07; Bossu–Gu 2004) showed it can be quasi-replicated by a zero-cost variance-dispersion trade, with P&L ≈ (ρ_realized − ρ_implied)·Σwᵢσᵢ². Implied correlation persistently trades at a premium to subsequent realized (exotic desks are structurally short correlation from selling structured products), generating the dispersion alpha. Caveats Bossu and Jacquier stress: variance-swap liquidity is uneven across legs, the proxy ignores vol-of-vol, and a fully rigorous dynamic replication of the correlation-swap payoff is still open.

FX implied correlation triangle in formula. Backing correlation out of the three vanilla vol markets:

The sign subtlety is real: the correlation between two pairs with the same base carries the opposite sign to the correlation where one’s base is the other’s quote. Only one of the three correlations in a trio can be negative (one obtuse angle), and arccos ρ₁ + arccos ρ₂ + arccos ρ₃ = π. The classic liquid example is EUR/USD, USD/JPY, EUR/JPY.

A.5 Co-skewness, the joint smile, and fat joint tails

The cross’s smile is not the sum of the leg smiles. The implied vol of an EUR/TRY option of a given delta is set by the joint risk-neutral density of (EUR/USD, USD/TRY), which carries:

Two same-delta vanillas reproduce the two marginal smiles by construction but encode nothing about dependence asymmetry. The market signals this with a correlation skew / “correlation frown”: the ρ implied by basket/spread options depends on strike (Alexander’s bivariate normal-mixture spread-option model; Austing, “Repricing the cross smile,” Risk 2011; De Col–Gnoatto–Grasselli multi-Heston; Linders et al. on Lévy implied correlation). Jumps are the physical source: a devaluation is a joint jump, injecting co-skewness/co-kurtosis no single Gaussian ρ can carry. Same-delta legs therefore cannot reproduce the joint tail.

Part B — Practitioner / market view (FX-specific)

B.1 How EM/FX cross-vol desks quote and warehouse the risk

FX quoting conventions. FX quotes vol against delta: per tenor the liquid points are ATM, 25Δ and 10Δ risk reversals (RR) and butterflies (BF):

The surface is interpolated (Vanna–Volga is FX-standard — Castagna–Mercurio 2007, Lipton–McGhee 2002 — or SABR/SSVI), with “sticky delta” the norm and bid/ask generated by a vol spread.

How cross vol is quoted and warehoused. The cross (EUR/TRY) has its own ATM/RR/BF surface. A desk decomposes its risk into directional (delta/forward), leg vegas (EUR/USD and USD/TRY vanillas), and the residual correlation/cross-gamma; it quotes the cross consistently with the legs and the triangle-implied correlation, hedges what it can in the liquid legs, and runs the rest. RBS’s structuring desk (Campbell-Smith & Hamdani, “The rise of multi-currency options,” Risk.net, 2010) describe visualizing this as an implied-volatility triangle, with multi-currency structures built precisely to monetize the implied-vs-realized correlation gap and netted vega. Castagna’s FX Options and Smile Risk (Wiley, 2010), Ch. 9, is the standard market-maker reference.

B.2 The delta-hedging fallacy — offshore funding squeezes

Textbook theory treats delta-hedging via forwards/NDFs as frictionless. In a managed EM regime the forward/swap market is itself a primary source of tail risk. Turkish authorities (BRSA/BDDK) have repeatedly restricted local banks from lending lira offshore to choke short-sellers, and the offshore overnight swap rate then erupts:

Mar 2019 O/N swap >1,300% offshore overnight; from ~24% policy level over three days
Aug 2020 O/N swap ~1,050% same mechanism; spot barely moved
BRSA offshore cap 25% of bank equity — lira lendable offshore (2018 rout)

The implication for an options book: a continuously-rolled short-delta hedge can become prohibitively expensive or structurally impossible exactly when you need it, and the cross-asset spillover (forced selling of TRY bonds/equities) is itself a contagion channel. The delta hedge is conditionally clean, not unconditionally.

B.3 The implied correlation triangle in practice

Desks back out implied correlation from the three vanilla markets and trade it against realized. The canonical liquid trio is EUR/USD, USD/JPY, EUR/JPY, where all three pairs have deep vanilla markets so all three correlations are cleanly extractable and tradable:

The FRBSF and ECB studies (Walter–Lopez; ECB WP 447) find FX option-implied correlations are informative forecasters of realized correlation, sometimes beating GARCH/historical models — genuine information content, part of why implied correlation tends to trade at a premium.

B.4 What is hedgeable with what — the instrument map

Hedgeable vs warehoused. Direction (conditionally) and leg-vega/skew are hedgeable in the liquid legs. Gamma, cross-gamma and correlation are hedgeable only via the cross option / correlation swap — true for liquid trios, largely false for EUR/TRY. Co-skewness and the jump tail are warehoused, reserved, and priced.

B.5 The jump/gap/devaluation character of fragile EM currencies (TRY)

TRY stresses every assumption of the rigorous theory.

USD/TRY 1m vol ~63% record high reached Dec 2021; vs EUR/USD ~7–11%
EUR/TRY 25Δ RR +11.6 vols 10Δ-RR +27.1 vols (1y, 2022-11-29 dataset)
Aug 2018 gap −17 to −20% intraday vs USD after tariff doubling
Mar 2025 gap −12% record low ~42/USD on İmamoğlu arrest

B.6 Regulatory capital — why the spread is wide (FRTB)

For EM cross-vol, the binding constraint is frequently capital, not hedging cost. Under the Basel III market-risk framework (FRTB):

RRAO — exotic underlying 1.0% gross-notional add-on (Standardised Approach)
RRAO — other residual 0.1% correlation & gap-risk bucket (baskets, spreads, digitals)
NMRF share of IMA 30–60% of total internal-models capital, in practice
NMRF eligibility (RFET) 24 prices/yr and ≤1 month between observations

B.7 Settlement numeraires and the “quanto” question

Handling the settlement currency dictates whether any adjustment applies. The common error is to attach a quanto drift to a natively-settled cross; it does not.

Operational layer. Hedging the cross via two legs crosses two spreads (EUR/USD tight; USD/TRY wide and blowing out in stress), so the synthetic hedge is materially more expensive than notional suggests and worst on exit. TRY rates near 37–40% mean forward points/funding dominate carry; onshore–offshore (NDF) basis, ECB-fixing settlement risk, and time-zone liquidity gaps add operational and pin risk.

Recommendations

A staged decision framework for warehousing/hedging an EUR/TRY (or general EM cross) option book:

  1. Strip out direction first — but respect the funding squeeze. Delta-hedge with EUR/TRY forwards/NDFs (or the EUR/USD + USD/TRY synthetic), continuously within liquidity windows. Actively monitor onshore vs offshore implied yields. If BRSA-style limits tighten or offshore overnight swaps spike (the 2019/2020 precedent: 300% → more than 1,300%), widen rehedge bands and re-price the funding gap rather than chase deltas into a squeezed, decoupled curve.
  2. Neutralize leg vega and skew in the liquid legs. Hold EUR/USD vanillas/RR/BF to fully offset the EUR/USD-leg vega, vanna, volga (deep, cheap). Hold USD/TRY vanillas/RR/BF to the extent the market bears it; expect to pay the large one-sided USD-call/TRY-put skew (~11 vols on the RR). Threshold: if USD/TRY 1m vol pushes toward its ~60% record or the 25Δ RR widens beyond ~10–12 vols, cut gross vega and pre-buy tail protection rather than hedge into a gapping market.
  3. Treat gamma, cross-gamma and correlation as one warehoused cluster. They are linked (Γ₁₂ = Γ₃S₃ + Δ₃) and all require the illiquid EUR/TRY option to neutralize cleanly. If a usable cross vanilla / correlation-swap market exists at acceptable spread, lay off cega via the cross itself or a correlation swap; otherwise warehouse deliberately, to a hard cega/cross-gamma limit. Do not attempt vol-triangle dispersion on TRY — the cross market is too thin and vanna/volga path-dependency ruins the vanilla proxy. (For liquid trios like EUR/USD/JPY, the dispersion route is the right tool.) If you do run a dispersion or covariance swap, get the sign right for a product cross: long correlation = long the cross variance, short both legs.
  4. Price capital into the spread, not just hedging cost. Size the warehoused correlation/cross-gamma to FRTB limits and load the NMRF/RRAO capital (correlation and gap risk are named residual-risk buckets; NMRF can be 30–60% of IMA capital) into the client execution spread. This is usually the binding economic constraint.
  5. Fund dedicated jump reserves; treat co-skew as unhedgeable. Buy cheap convexity (OTM cross wings, short-dated USD/TRY USD-calls) as partial insurance, but accept the joint devaluation tail is not replicable. Stress the book to a 15–20% one-day lira gap (the 2018/2021/2025 precedents) with correlation snapping toward 1 and cross-gamma at maximum; hold physical reserves, because continuous delta-hedging fails in a gap.
  6. Manage the settlement/quanto layer correctly. TRY-settled = plain vanilla, no adjustment. EUR-settled = self-quanto (own-variance / Siegel adjustment). USD-settled = quanto (−ρσ₃σ_Q drift, fixed rate) or compo (floating). Budget double bid/ask on synthetic hedges and watch fixing/pin risk near ECB cross fixings and barriers.

Triggers that flip the strategy: EUR/TRY vol-market liquidity improving so the cross vanilla bid/ask falls below the cost of warehousing (incl. capital) → shift from warehouse to hedge on correlation/gamma. USD/TRY realized correlation with EUR/USD destabilizing (regime change, capital controls) → cut gross cross-gamma and cega regardless of carry. Offshore swap rates spiking → freeze/widen delta rehedging. Vol-triangle dispersion only where all three legs are liquid — never in TRY.

Caveats

Reference table — risk dimension → hedge instrument → status (EUR/TRY)

Risk dimensionMath objectHedge instrumentStatus for EUR/TRY
DirectionDelta ∂V/∂S₃EUR/TRY forward/NDF; or EUR/USD + USD/TRY forwardsConditionally hedgeable — exposed to offshore swap squeezes (O/N rates >1,300% in 2019, ~1,050% in 2020)
Leg vega∂V/∂σ₁, ∂V/∂σ₂EUR/USD vanillas (deep); USD/TRY vanillas (wide)Hedgeable, asymmetric cost
Skew (vanna)∂Vega/∂S25Δ/10Δ risk reversals on each legHedgeable; TRY RR large & one-sided
Convexity (volga)∂Vega/∂σButterflies on each legHedgeable
Gamma (cross vol)∂²V/∂S₃²Cross (EUR/TRY) vanillas onlyLargely warehoused — needs illiquid cross options; tied to cross-gamma by Γ₁₂ = Γ₃S₃ + Δ₃
Cross-gamma∂²V/∂S₁∂S₂Cross option / covariance swapLargely warehoused — same convexity as gamma in leg coordinates; leg vanillas carry zero cross-gamma
Correlation (cega)dV/dρ = Vega₃·σ₁σ₂/σ₃Cross option; correlation/covariance swap; vol-triangle dispersion (product cross: long cross var / short legs)Warehoused for TRY (cross & corr-swap markets thin); hedgeable for EUR/USD/JPY-type trios
Co-skew / joint tailco-skew, co-kurtosisOTM cross wings; worst-of/exotic structuresWarehoused / reserved; partially expressible, not replicable
Ratesρ_EUR, ρ_TRYRates/FX-forward instrumentsHedgeable (TRY rate ~37–40% dominates carry)
Settlement / quantoTRY: none · EUR: own-variance (self-quanto) · USD: −ρσ₃σ_Q (quanto) or compoMatch settlement ccy; quanto/compo adjustManageable — native TRY settlement needs no adjustment; only non-natural settlement does
Regulatory capitalNMRF (RFET) + RRAO (correlation/gap)Not hedgeable — priced into spreadCapitalised — NMRF can be 30–60% of IMA capital; RRAO on notional; the binding RoC constraint

This note is methodology-level. The concrete EUR/TRY vol and rate numbers are illustrative and convention-dependent, not a live quote.