LP-Fx: Local Volatility in FX Derivatives

• LP-Fx is a framework that extends Dupire's method to a three-factor FX model with state-dependent local volatility and stochastic domestic and foreign rates.
• It employs calibration techniques, including Monte Carlo, PDE, and quantization, to accurately price vanilla and exotic FX derivatives like PRDCs and barriers.
• LP-Fx also informs operational decisions in FX hedge tenor optimization, liquidity management, and macro-financial resilience diagnostics.

LP-Fx is a label that, in the supplied arXiv literature, is used most centrally for local-volatility pricing of FX derivatives in models with stochastic domestic and foreign rates, and more loosely for adjacent FX quantitative problems in pricing, hedging, liquidity management, and macro-financial diagnostics. Its most developed meaning is the extension of Dupire-style local-volatility methodology to a three-factor FX setting driven by spot, domestic short rates, and foreign short rates, explicitly motivated by long-dated FX derivatives such as PRDCs, long-dated barriers, and other hybrid products (Deelstra et al., 2012).

1. Terminological scope

In the supplied corpus, LP-Fx does not denote a single universally standardized construction. Instead, it appears across several technically distinct FX-centered frameworks. This suggests that LP-Fx functions as a context-dependent shorthand whose meaning is fixed by the modeling problem under discussion rather than by a single canonical expansion.

Usage in the literature	Representative content	arXiv id
Local-volatility pricing for long-dated FX derivatives	Three-factor stochastic-rates generalization of Dupire	(Deelstra et al., 2012)
Fully parameterized local-volatility calibration	Grid-based FX local-vol surface for American and Asian options	(Wu et al., 2022)
Quantization-based Bermudan pricing in the FX world	Three-factor Bermudan and PRDC-type pricing	(Fayolle et al., 2019)
FX-forward hedge-tenor optimization	Carry maximization under bucket-level CFaR constraints	(Zhang et al., 2019)
FX market making with internal liquidity	HJBQVI control of quoting and internalization	(Barzykin et al., 4 Dec 2025)
FX resilience measurement	Marginal sensitivity of FX volatility to capital-flow volatility	(Chen et al., 2022)
Alternative FX models for barriers	Regime-switching Lévy models with memory	(Boyarchenko et al., 2024)

The dominant technical thread is nevertheless the local-volatility pricing program. In that strand, LP-Fx is anchored in a three-factor domestic risk-neutral FX model in which the spot process has state-dependent local volatility and both short rates follow Hull–White one-factor Gaussian dynamics (Deelstra et al., 2012).

2. Three-factor local-volatility framework

The core LP-Fx construction in the pricing literature starts from the domestic risk-neutral measure $Q_d$ and models the FX spot by

$$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$$

with domestic and foreign short rates

$$dr_d(t)=\bigl[\theta_d(t)-\alpha_d(t)r_d(t)\bigr]dt+\sigma_d(t)\,dW^{DRN}_d(t),$$

$$dr_f(t)=\bigl[\theta_f(t)-\alpha_f(t)r_f(t)-\rho_{fS}\sigma_f(t)\sigma(t,S(t))\bigr]dt+\sigma_f(t)\,dW^{DRN}_f(t).$$

The foreign-rate drift adjustment is the usual quanto correction induced by expressing the foreign short rate under the domestic measure. The Brownian motions under $Q_d$ have correlation matrix

$$\begin{pmatrix} 1 & \rho_{Sd} & \rho_{Sf}\ \rho_{Sd} & 1 & \rho_{df}\ \rho_{Sf} & \rho_{df} & 1 \end{pmatrix}.$$

This construction generalizes the standard one-factor Dupire setting with deterministic rates to a three-factor FX model suitable for long-dated derivatives (Deelstra et al., 2012).

The technical derivation proceeds under the $t$-forward measure $Q_t$ associated with the domestic zero-coupon bond $P_d(\cdot,t)$. With forward call price

$$\widetilde C(K,t)=\frac{C(K,t)}{P_d(0,t)} =\mathbb{E}^{Q_t}\bigl[(S(t)-K)^+\bigr],$$

the paper derives

$$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$$0

and then the generalized Dupire identity

$$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$$1

Equivalently, in terms of the undiscounted call price,

$$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$$2

The distinguishing feature relative to standard Dupire is the expectation term coupling spot and both short rates. When rates are deterministic, the formula collapses to the familiar one-factor expression

$$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$$3

A common misconception is therefore ruled out explicitly: in the stochastic-rate FX setting, local volatility cannot be extracted from vanilla prices alone through strike and maturity differentiation. The missing object is the nontrivial forward-measure expectation

$$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$$4

so stochastic-rate local volatility is not a purely vanilla-surface inversion problem (Deelstra et al., 2012).

A further interpretation compares the stochastic-rate local-volatility surface $$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$$5 with the deterministic-rate surface $$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$$6: $$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$$7 For long maturities, those covariance corrections need not be negligible.

3. Calibration routes, Markovian projection, and hybrid volatility

The three-factor LP-Fx framework proposes a layered calibration procedure. First, calibrate the domestic and foreign Hull–White parameters $$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$$8 to their respective cap/floor and swaption markets. Second, estimate $$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$$9, typically from historical data. Third, calibrate the FX local-volatility surface $$dr_d(t)=\bigl[\theta_d(t)-\alpha_d(t)r_d(t)\bigr]dt+\sigma_d(t)\,dW^{DRN}_d(t),$$0 to the FX implied-volatility surface. Because the generalized Dupire formula contains the expectation coupling rates and spot, the paper discusses four routes: Monte Carlo under the $$dr_d(t)=\bigl[\theta_d(t)-\alpha_d(t)r_d(t)\bigr]dt+\sigma_d(t)\,dW^{DRN}_d(t),$$1-forward measure, a forward-PDE approach, a correction from deterministic-rate local volatility, and a mimicking route from a stochastic-volatility model (Deelstra et al., 2012).

The mimicking result is central. For

$$dr_d(t)=\bigl[\theta_d(t)-\alpha_d(t)r_d(t)\bigr]dt+\sigma_d(t)\,dW^{DRN}_d(t),$$2

the local volatility reproducing the same vanilla option prices satisfies

$$dr_d(t)=\bigl[\theta_d(t)-\alpha_d(t)r_d(t)\bigr]dt+\sigma_d(t)\,dW^{DRN}_d(t),$$3

or, under the $$dr_d(t)=\bigl[\theta_d(t)-\alpha_d(t)r_d(t)\bigr]dt+\sigma_d(t)\,dW^{DRN}_d(t),$$4-forward measure,

$$dr_d(t)=\bigl[\theta_d(t)-\alpha_d(t)r_d(t)\bigr]dt+\sigma_d(t)\,dW^{DRN}_d(t),$$5

This is the stochastic-rate analogue of the classical Dupire–Derman–Kani Markovian projection identity. If $$dr_d(t)=\bigl[\theta_d(t)-\alpha_d(t)r_d(t)\bigr]dt+\sigma_d(t)\,dW^{DRN}_d(t),$$6 and $$dr_d(t)=\bigl[\theta_d(t)-\alpha_d(t)r_d(t)\bigr]dt+\sigma_d(t)\,dW^{DRN}_d(t),$$7 are independent, the conditional expectation simplifies to

$$dr_d(t)=\bigl[\theta_d(t)-\alpha_d(t)r_d(t)\bigr]dt+\sigma_d(t)\,dW^{DRN}_d(t),$$8

The same paper extends the construction to a hybrid model

$$dr_d(t)=\bigl[\theta_d(t)-\alpha_d(t)r_d(t)\bigr]dt+\sigma_d(t)\,dW^{DRN}_d(t),$$9

with diffusion for $$dr_f(t)=\bigl[\theta_f(t)-\alpha_f(t)r_f(t)-\rho_{fS}\sigma_f(t)\sigma(t,S(t))\bigr]dt+\sigma_f(t)\,dW^{DRN}_f(t).$$0,

$$dr_f(t)=\bigl[\theta_f(t)-\alpha_f(t)r_f(t)-\rho_{fS}\sigma_f(t)\sigma(t,S(t))\bigr]dt+\sigma_f(t)\,dW^{DRN}_f(t).$$1

The relation between the pure-local surface $$dr_f(t)=\bigl[\theta_f(t)-\alpha_f(t)r_f(t)-\rho_{fS}\sigma_f(t)\sigma(t,S(t))\bigr]dt+\sigma_f(t)\,dW^{DRN}_f(t).$$2 and the hybrid local factor $$dr_f(t)=\bigl[\theta_f(t)-\alpha_f(t)r_f(t)-\rho_{fS}\sigma_f(t)\sigma(t,S(t))\bigr]dt+\sigma_f(t)\,dW^{DRN}_f(t).$$3 is

$$dr_f(t)=\bigl[\theta_f(t)-\alpha_f(t)r_f(t)-\rho_{fS}\sigma_f(t)\sigma(t,S(t))\bigr]dt+\sigma_f(t)\,dW^{DRN}_f(t).$$4

hence

$$dr_f(t)=\bigl[\theta_f(t)-\alpha_f(t)r_f(t)-\rho_{fS}\sigma_f(t)\sigma(t,S(t))\bigr]dt+\sigma_f(t)\,dW^{DRN}_f(t).$$5

The motivation for hybridization is stated explicitly. Pure local volatility reproduces the entire vanilla smile exactly but often gives unrealistic smile dynamics, whereas pure stochastic volatility captures more realistic dynamics but may hedge vanillas less consistently. The hybrid construction is proposed as a way to preserve vanilla consistency while introducing stochastic-volatility behavior. The associated caution is equally explicit: matching vanilla prices alone does not guarantee correct exotic pricing (Deelstra et al., 2012).

4. Numerical implementations and exotic-option engines

A separate strand of LP-Fx work replaces Dupire inversion by direct parameterization of the local-volatility surface. In "Efficient and Accurate Calibration to FX Market Skew with Fully Parameterized Local Volatility Model" (Wu et al., 2022), the spot is modeled by

$$dr_f(t)=\bigl[\theta_f(t)-\alpha_f(t)r_f(t)-\rho_{fS}\sigma_f(t)\sigma(t,S(t))\bigr]dt+\sigma_f(t)\,dW^{DRN}_f(t).$$6

and the local-volatility surface is parameterized directly on the transformed domain

$$dr_f(t)=\bigl[\theta_f(t)-\alpha_f(t)r_f(t)-\rho_{fS}\sigma_f(t)\sigma(t,S(t))\bigr]dt+\sigma_f(t)\,dW^{DRN}_f(t).$$7

In the reported example the surface is represented on an $$dr_f(t)=\bigl[\theta_f(t)-\alpha_f(t)r_f(t)-\rho_{fS}\sigma_f(t)\sigma(t,S(t))\bigr]dt+\sigma_f(t)\,dW^{DRN}_f(t).$$8 grid, giving $$dr_f(t)=\bigl[\theta_f(t)-\alpha_f(t)r_f(t)-\rho_{fS}\sigma_f(t)\sigma(t,S(t))\bigr]dt+\sigma_f(t)\,dW^{DRN}_f(t).$$9 free parameters, with bilinear interpolation inside each rectangle. Calibration uses European FX vanilla options, relative price errors

$Q_d$0

and Levenberg–Marquardt with finite-difference Jacobians. The benchmark example has $Q_d$1 local-vol parameters and $Q_d$2 market instruments, with reported

$Q_d$3

The same calibrated surface is then used for American options through a grid method and for Asian options through Monte Carlo. The paper reports that American option values are stable within about $Q_d$4 basis point when increasing grid resolution, and Asian option values are stable within about $Q_d$5–$Q_d$6 basis points as path count rises from $Q_d$7 to $Q_d$8. It also imposes practical admissibility conditions on market inputs, notably increasing ATM total variance with maturity and $Q_d$9 at each tenor.

Long-dated exercisable FX products motivate a different numerical LP-Fx program based on Product Optimal Quantization. In "Quantization-based Bermudan option pricing in the $$\begin{pmatrix} 1 & \rho_{Sd} & \rho_{Sf}\ \rho_{Sd} & 1 & \rho_{df}\ \rho_{Sf} & \rho_{df} & 1 \end{pmatrix}.$$0 world" (Fayolle et al., 2019), the model is again three-factor, with stochastic domestic and foreign rates on top of stochastic FX, and the focus is Bermudan FX-linked products such as Bermudan PRDCs. The paper develops a 4D quantization tree for the exact Markov state $$\begin{pmatrix} 1 & \rho_{Sd} & \rho_{Sf}\ \rho_{Sd} & 1 & \rho_{df}\ \rho_{Sf} & \rho_{df} & 1 \end{pmatrix}.$$1 and a faster 2D non-Markov tree for $$\begin{pmatrix} 1 & \rho_{Sd} & \rho_{Sf}\ \rho_{Sd} & 1 & \rho_{df}\ \rho_{Sf} & \rho_{df} & 1 \end{pmatrix}.$$2. The theoretical trade-off is explicit: the 4D method is asymptotically exact, while the 2D method converges to a slightly biased limit. Numerically, for zero correlations, $$\begin{pmatrix} 1 & \rho_{Sd} & \rho_{Sf}\ \rho_{Sd} & 1 & \rho_{df}\ \rho_{Sf} & \rho_{df} & 1 \end{pmatrix}.$$3 bp relative error is reached very quickly for European benchmarks: non-Markov 2D up to about $$\begin{pmatrix} 1 & \rho_{Sd} & \rho_{Sf}\ \rho_{Sd} & 1 & \rho_{df}\ \rho_{Sf} & \rho_{df} & 1 \end{pmatrix}.$$4 ms and Markov 4D up to about $$\begin{pmatrix} 1 & \rho_{Sd} & \rho_{Sf}\ \rho_{Sd} & 1 & \rho_{df}\ \rho_{Sf} & \rho_{df} & 1 \end{pmatrix}.$$5 ms. For yearly Bermudans with $$\begin{pmatrix} 1 & \rho_{Sd} & \rho_{Sf}\ \rho_{Sd} & 1 & \rho_{df}\ \rho_{Sf} & \rho_{df} & 1 \end{pmatrix}.$$6 bp and zero correlations, the relative difference to the 4D method after convergence is at most about $$\begin{pmatrix} 1 & \rho_{Sd} & \rho_{Sf}\ \rho_{Sd} & 1 & \rho_{df}\ \rho_{Sf} & \rho_{df} & 1 \end{pmatrix}.$$7 bp even for a $$\begin{pmatrix} 1 & \rho_{Sd} & \rho_{Sf}\ \rho_{Sd} & 1 & \rho_{df}\ \rho_{Sf} & \rho_{df} & 1 \end{pmatrix}.$$8 product. To achieve roughly $$\begin{pmatrix} 1 & \rho_{Sd} & \rho_{Sf}\ \rho_{Sd} & 1 & \rho_{df}\ \rho_{Sf} & \rho_{df} & 1 \end{pmatrix}.$$9 bp relative precision in the Bermudan case, the reported runtimes are $t$0–$t$1 ms for the 2D non-Markov method and $t$2 ms up to $t$3 s for the 4D Markov method.

Taken together, these papers define two implementation styles within LP-Fx. One uses a fully parameterized local-volatility surface calibrated directly to the market skew and then reused for American and Asian pricing. The other treats long-dated stochastic-rate Bermudans through quantization trees in a three-factor model. A plausible implication is that LP-Fx is not tied to a single numerical engine; grid methods, Monte Carlo, PDE-based schemes, and quantization all appear as legitimate implementations once the FX state dynamics have been fixed.

5. Liquidity, hedging, and dealer-control interpretations

In FX-forward hedge optimization, LP-FX is used in a more operational sense. "Optimal FX Hedge Tenor with Liquidity Risk" studies a fund manager who maintains a $t$4 FX hedge ratio using uncollateralised FX forwards and must decide which tenors to use when staggering the hedge book. The optimization is

$t$5

subject to the full-hedge constraint

$t$6

and the bucketwise liquidity constraint

$t$7

With an Ornstein–Uhlenbeck spot model,

$t$8

the bucket-level CFaR is available in closed form: $t$9 The operational rule is to rank tenors by expected carry net of cost and then fill the CFaR capacity of each maturity bucket sequentially. Short-dated hedges are preferred when short-tenor expected carry net of costs is highest and liquidity headroom is ample; long-dated hedges are preferred when the liquidity budget is tight, spot volatility is high, or expected spot moves worsen near-dated settlement outflows (Zhang et al., 2019).

A more microstructural LP-FX interpretation appears in "FX Market Making with Internal Liquidity" (Barzykin et al., 4 Dec 2025). There the dealer streams external OTC quotes

$Q_t$0

while also facing passive internal liquidity. OTC fills arrive with quote-sensitive intensities

$Q_t$1

and the dealer chooses both quote depths and stopping times at which to consume internal liquidity. The objective combines terminal mark-to-market wealth, terminal inventory penalty, running inventory penalty, and running penalty for leaving internal client liquidity unfilled: $Q_t$2 The resulting HJBQVI yields an inventory-dependent execution threshold: the dealer should not automatically internalize as soon as passive internal liquidity appears. Instead, optimal behavior is to wait in some states and skew OTC prices to attract offsetting flow that moves inventory toward the execution region. The paper reports that the optimal policy dominates a naïve benchmark in simulated PnL across all client types and pricing regimes. From Table 1, Iceberg PnL (K$Q_t$3) is optimal $Q_t$4 versus naïve $Q_t$5; Full Amount PnL (K$) is optimal$Q_t$6 versus naïve $Q_t$7. Table 2 simultaneously shows slower expected first-fill times under the optimal policy, for example Iceberg optimal $Q_t$8, $Q_t$9, $P_d(\cdot,t)$0 sec versus naïve $P_d(\cdot,t)$1, $P_d(\cdot,t)$2, $P_d(\cdot,t)$3 sec.

These works extend LP-Fx from pricing into balance-sheet and execution control. One addresses tenor allocation under bucket-level liquidity risk; the other addresses joint quoting-and-stopping under internalization. In both cases, the common pattern is optimization under explicit state-dependent constraints rather than static FX exposure management.

6. Macro resilience and alternative FX dynamics

At the macro-financial level, "FX Resilience around the World: Fighting Volatile Cross-Border Capital Flows" defines an FX resilience measure as the marginal sensitivity of exchange-rate volatility to capital-flow volatility (Chen et al., 2022). In the paper’s notation,

$P_d(\cdot,t)$4

and, for country rankings,

$P_d(\cdot,t)$5

A more negative value indicates a smaller effect of $P_d(\cdot,t)$6 and a stronger FX resilience. The significant moderating fundamentals included in the composite factor are trade openness, FX reserves, total foreign investment, short-term interest rate, fiscal surplus, and financial development. The paper’s quarterly regressions show that capital-flow volatility increases FX volatility, regardless of whether capital controls have been put in place, while stronger macroeconomic fundamentals attenuate that pass-through. Among the reported interaction coefficients, financial development $P_d(\cdot,t)$7, fiscal surplus $P_d(\cdot,t)$8, and short-term interest rate $P_d(\cdot,t)$9 are especially important. Country rankings place Hong Kong $\widetilde C(K,t)=\frac{C(K,t)}{P_d(0,t)} =\mathbb{E}^{Q_t}\bigl[(S(t)-K)^+\bigr],$0, Singapore $\widetilde C(K,t)=\frac{C(K,t)}{P_d(0,t)} =\mathbb{E}^{Q_t}\bigl[(S(t)-K)^+\bigr],$1, and Switzerland $\widetilde C(K,t)=\frac{C(K,t)}{P_d(0,t)} =\mathbb{E}^{Q_t}\bigl[(S(t)-K)^+\bigr],$2 among the most resilient examples, and Egypt $\widetilde C(K,t)=\frac{C(K,t)}{P_d(0,t)} =\mathbb{E}^{Q_t}\bigl[(S(t)-K)^+\bigr],$3, India $\widetilde C(K,t)=\frac{C(K,t)}{P_d(0,t)} =\mathbb{E}^{Q_t}\bigl[(S(t)-K)^+\bigr],$4, and Brazil $\widetilde C(K,t)=\frac{C(K,t)}{P_d(0,t)} =\mathbb{E}^{Q_t}\bigl[(S(t)-K)^+\bigr],$5 among the least resilient.

A different broadening of LP-Fx appears in "Alternative models for FX: pricing double barrier options in regime-switching Lévy models with memory" (Boyarchenko et al., 2024). This work departs from diffusion-centric modeling and introduces regime-switching Lévy models with finite memory after truncation. The regime history is

$\widetilde C(K,t)=\frac{C(K,t)}{P_d(0,t)} =\mathbb{E}^{Q_t}\bigl[(S(t)-K)^+\bigr],$6

with

$\widetilde C(K,t)=\frac{C(K,t)}{P_d(0,t)} =\mathbb{E}^{Q_t}\bigl[(S(t)-K)^+\bigr],$7

The state variable $\widetilde C(K,t)=\frac{C(K,t)}{P_d(0,t)} =\mathbb{E}^{Q_t}\bigl[(S(t)-K)^+\bigr],$8 is monitored against lower and upper barriers $\widetilde C(K,t)=\frac{C(K,t)}{P_d(0,t)} =\mathbb{E}^{Q_t}\bigl[(S(t)-K)^+\bigr],$9 and $dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$00, and the Laplace-domain valuation system is

$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$01

with $dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$02 outside the barriers. The computational contribution is a normalization using

$dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$03

so that only $dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$04 distinct operator blocks $dS(t) = \bigl(r_d(t)-r_f(t)\bigr)S(t)\,dt + \sigma(t,S(t))S(t)\,dW^{DRN}_S(t),$05 remain on the left-hand side, even though the history space may be much larger. The paper therefore claims the same number of the main time-consuming blocks as in the Markovian regime-switching case. It also states that the memory-dependent structure is flexible and suitable for application of the machine-learning tools.

These two extensions show how far the LP-Fx label can stretch. In one direction it becomes a macro diagnostic for the transmission of capital-flow volatility into exchange-rate volatility. In another it denotes exotic-option pricing under non-Markovian regime-switching Lévy dynamics. The commonality is not a single algorithmic template but a persistent focus on state dependence, regime dependence, and the inadequacy of one-factor or purely static representations of FX behavior.

LP-Fx, taken across this literature, is therefore best understood as a family resemblance rather than a single model class. Its central core is the three-factor local-volatility theory for long-dated FX derivatives, especially the stochastic-rate analogue of Dupire, the associated calibration problem, and the hybrid local/stochastic-volatility extension. Around that core sit numerical engines for American, Asian, Bermudan, and barrier exotics; optimization problems for hedging ladders and internal dealer liquidity; and macro measures of exchange-rate shock absorption. This suggests that the unifying idea of LP-Fx is not a fixed acronymic expansion but the use of structured quantitative models to encode how FX dynamics interact with rates, liquidity, and state-dependent constraints across pricing and risk management.