finquant 0.0.59

Experimental Rust Quant Library
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
//! Fang–Oosterlee COS method for European option pricing from a
//! characteristic function.
//!
//! Reference: Fang & Oosterlee (2008), *A Novel Pricing Method for
//! European Options Based on Fourier-Cosine Series Expansions*,
//! SIAM J. Sci. Comput. 31(2). The discounted price of a European
//! payoff on `y = log FX_T(T)` is approximated by
//!
//! ```text
//!   V(0) ≈ Pd(0,T) · Σ_{k=0}^{N-1} ' Re{ φ(k·π/(b−a)) · e^{−i·k·π·a/(b−a)} } · V_k
//! ```
//!
//! where `Σ'` halves the `k = 0` term, `φ` is the forward-measure ChF
//! evaluated at frequencies `u_k = k·π/(b−a)`, and `V_k` are closed-form
//! cosine-coefficients of the payoff.
//!
//! Truncation range `[a, b]` is chosen wide enough to catch the effective
//! support of `f_y(·)`. Defaults here centre on `log FX_T(0)` with a
//! half-width of `L · √(σ̄·T)` — a conservative Heston-style rule with
//! `L = 10`. Callers can override via [`CosPricer::with_range`].

use crate::models::forex::fx_hhw1_chf::FxHhw1ForwardChf;
use crate::models::forex::fx_hlmm1_chf::FxHlmm1ForwardChf;
use num_complex::Complex64;

/// Interface a characteristic function must expose for COS pricing:
/// frequency evaluation plus enough information to build a sensible
/// default truncation range. Implemented for both FX-HHW1 and FX-HLMM1.
pub trait ForwardChf {
    /// `φ_T(u)` at time 0 under the forward domestic measure.
    fn evaluate(&self, u: Complex64) -> Complex64;
    /// Maturity `T` in year-fractions.
    fn expiry(&self) -> f64;
    /// `log FX_T(0)` — the initial forward's log, used to centre the
    /// COS truncation range.
    fn log_forward(&self) -> f64;
    /// Typical instantaneous variance of `log FX_T(T)`. Used to set the
    /// COS half-width; a factor-10 is applied on top. For Heston-type
    /// models this is just the long-run variance `σ̄`.
    fn typical_variance(&self) -> f64;
}

impl<'a> ForwardChf for FxHhw1ForwardChf<'a> {
    fn evaluate(&self, u: Complex64) -> Complex64 {
        FxHhw1ForwardChf::evaluate(self, u)
    }
    fn expiry(&self) -> f64 {
        self.expiry
    }
    fn log_forward(&self) -> f64 {
        let p = self.params();
        let t = self.expiry;
        (p.fx_0 * (-p.rf_0 * t).exp() / (-p.rd_0 * t).exp()).ln()
    }
    fn typical_variance(&self) -> f64 {
        self.params().heston.theta
    }
}

impl<'a> ForwardChf for FxHlmm1ForwardChf<'a> {
    fn evaluate(&self, u: Complex64) -> Complex64 {
        FxHlmm1ForwardChf::evaluate(self, u)
    }
    fn expiry(&self) -> f64 {
        self.expiry
    }
    fn log_forward(&self) -> f64 {
        // FX-HLMM rates are derived from initial Libor levels inside the
        // ChF; the forward offset is already absorbed there, so evaluating
        // at `u = 0` leaves just `log FX_T(0)` under the exp.
        self.params().fx_0.ln()
    }
    fn typical_variance(&self) -> f64 {
        // FX Heston variance is the dominant contributor; LMM contributions
        // are second-order for typical parameter sets.
        self.params().heston.theta
    }
}

impl<'a> ForwardChf for crate::models::forex::fx_fmm1_chf::FxFmm1ForwardChf<'a> {
    fn evaluate(&self, u: Complex64) -> Complex64 {
        crate::models::forex::fx_fmm1_chf::FxFmm1ForwardChf::evaluate(self, u)
    }
    fn expiry(&self) -> f64 {
        self.expiry
    }
    fn log_forward(&self) -> f64 {
        // Same single-curve convention as FX-HLMM: P_f ≈ P_d under the
        // shared-tenor setup, so `log FX_T(0) = log ξ(0)`.
        self.params().fx_0.ln()
    }
    fn typical_variance(&self) -> f64 {
        self.params().heston.theta
    }
}

/// Pricer that glues any `ForwardChf` onto the COS quadrature.
pub struct CosPricer<'a, C: ForwardChf> {
    chf: &'a C,
    /// Number of Fourier-cosine terms. More terms → better accuracy.
    pub n_terms: usize,
    /// Truncation range `[a, b]` in log-forward space.
    pub range: (f64, f64),
}

impl<'a, C: ForwardChf> CosPricer<'a, C> {
    /// Build a pricer with default `N = 128` terms and a cumulant-based
    /// truncation range centred on `log FX_T(0)` with half-width
    /// `10 · √(σ̄·T)`.
    pub fn new(chf: &'a C) -> Self {
        let t = chf.expiry();
        let log_forward = chf.log_forward();
        let half_width = 10.0 * (chf.typical_variance() * t).sqrt().max(1.0e-6);
        let a = log_forward - half_width;
        let b = log_forward + half_width;
        Self {
            chf,
            n_terms: 128,
            range: (a, b),
        }
    }

    /// Override number of Fourier-cosine terms.
    pub fn with_n_terms(mut self, n: usize) -> Self {
        assert!(n >= 2, "n_terms must be ≥ 2");
        self.n_terms = n;
        self
    }

    /// Override truncation range explicitly (log-forward units).
    pub fn with_range(mut self, a: f64, b: f64) -> Self {
        assert!(b > a, "upper bound must exceed lower");
        self.range = (a, b);
        self
    }

    /// European call price on `FX_T(T)` struck at `K`. The pricer is
    /// forward-measure internally: the discount factor `Pd(0, T)` is
    /// applied once at the end, so callers in a constant-rates test
    /// pass `exp(−rd · T)`; calibration callers pass whatever market
    /// discount they've already stripped.
    pub fn call(&self, strike: f64, discount: f64) -> f64 {
        self.price_european(strike, discount, true)
    }

    /// European put price on `FX_T(T)` struck at `K`.
    pub fn put(&self, strike: f64, discount: f64) -> f64 {
        self.price_european(strike, discount, false)
    }

    fn price_european(&self, strike: f64, discount: f64, is_call: bool) -> f64 {
        assert!(strike > 0.0);
        let (a, b) = self.range;
        let width = b - a;
        let log_k = strike.ln();

        // Payoff cosine coefficients Vk (closed form).
        // For a call:  Vk = (2/(b−a)) · [χk(log K, b) − K · ψk(log K, b)].
        // For a put:   Vk = (2/(b−a)) · [−χk(a, log K) + K · ψk(a, log K)].
        let (c, d) = if is_call { (log_k, b) } else { (a, log_k) };

        let mut total = 0.0_f64;
        for k in 0..self.n_terms {
            let u = k as f64 * std::f64::consts::PI / width;
            let chi = chi_k(k, a, b, c, d);
            let psi = psi_k(k, a, b, c, d);
            let vk = if is_call {
                (2.0 / width) * (chi - strike * psi)
            } else {
                (2.0 / width) * (-chi + strike * psi)
            };
            let uu = Complex64::new(u, 0.0);
            let phi = self.chf.evaluate(uu);
            let shift = Complex64::new(0.0, -u * a).exp();
            let term_cplx = phi * shift;
            // `Σ'` convention: halve the k=0 term.
            let weight = if k == 0 { 0.5 } else { 1.0 };
            total += weight * term_cplx.re * vk;
        }
        discount * total
    }
}

/// Closed-form χ_k(c, d) (Fang-Oosterlee eq. 22).
fn chi_k(k: usize, a: f64, b: f64, c: f64, d: f64) -> f64 {
    let width = b - a;
    let k_pi_w = k as f64 * std::f64::consts::PI / width;
    let arg_d = k_pi_w * (d - a);
    let arg_c = k_pi_w * (c - a);
    let denom = 1.0 + k_pi_w * k_pi_w;
    let term_cos = arg_d.cos() * d.exp() - arg_c.cos() * c.exp();
    let term_sin = arg_d.sin() * d.exp() - arg_c.sin() * c.exp();
    (term_cos + k_pi_w * term_sin) / denom
}

/// Closed-form ψ_k(c, d) (Fang-Oosterlee eq. 23).
/// `ψ_k = ∫_c^d cos(u(y-a)) dy` with `u = k·π/(b-a)`,
/// giving `(sin(u(d-a)) − sin(u(c-a))) / u` for `k ≥ 1`, and
/// `d − c` for `k = 0`.
fn psi_k(k: usize, a: f64, b: f64, c: f64, d: f64) -> f64 {
    if k == 0 {
        return d - c;
    }
    let width = b - a;
    let k_pi_w = k as f64 * std::f64::consts::PI / width;
    ((k_pi_w * (d - a)).sin() - (k_pi_w * (c - a)).sin()) / k_pi_w
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::math::normal::cdf;
    use crate::models::common::cir::CirProcess;
    use crate::models::forex::fx_hhw::{Correlation4x4, FxHhwParams, FxHhwSimulator};
    use crate::models::forex::fx_hhw1_chf::FxHhw1ForwardChf;
    use crate::models::interestrate::hull_white::HullWhite1F;

    /// Black–Scholes call on a forward: `Pd·[F·Φ(d₁) − K·Φ(d₂)]` with
    /// `d₁,₂ = (ln(F/K) ± σ²T/2)/(σ√T)`. Inline helper for the BS-limit
    /// unit test below — we don't ship a separate BS module.
    fn bs_call_forward(forward: f64, strike: f64, sigma: f64, t: f64, discount: f64) -> f64 {
        let sqrt_t = t.sqrt();
        let d1 = ((forward / strike).ln() + 0.5 * sigma * sigma * t) / (sigma * sqrt_t);
        let d2 = d1 - sigma * sqrt_t;
        discount * (forward * cdf(d1) - strike * cdf(d2))
    }

    fn paper_params() -> FxHhwParams {
        FxHhwParams {
            fx_0: 1.35,
            heston: CirProcess {
                kappa: 0.5,
                theta: 0.1,
                gamma: 0.3,
                sigma_0: 0.1,
            },
            domestic: HullWhite1F {
                mean_reversion: 0.01,
                sigma: 0.007,
            },
            foreign: HullWhite1F {
                mean_reversion: 0.05,
                sigma: 0.012,
            },
            rd_0: 0.02,
            rf_0: 0.05,
            theta_d: 0.02,
            theta_f: 0.05,
            correlations: Correlation4x4 {
                rho_xi_sigma: -0.40,
                rho_xi_d: -0.15,
                rho_xi_f: -0.15,
                rho_sigma_d: 0.30,
                rho_sigma_f: 0.30,
                rho_d_f: 0.25,
            },
        }
    }

    /// Pure Black-Scholes limit: `γ = 0`, `η_d = η_f = 0`, no correlations.
    /// The COS call price must match the BS call to `1e-5` or better.
    #[test]
    fn cos_matches_black_scholes_in_gbm_limit() {
        let p = FxHhwParams {
            fx_0: 1.35,
            heston: CirProcess {
                kappa: 1.0,
                theta: 0.04,
                gamma: 0.0,
                sigma_0: 0.04,
            },
            domestic: HullWhite1F {
                mean_reversion: 0.01,
                sigma: 0.0,
            },
            foreign: HullWhite1F {
                mean_reversion: 0.05,
                sigma: 0.0,
            },
            rd_0: 0.02,
            rf_0: 0.05,
            theta_d: 0.02,
            theta_f: 0.05,
            correlations: Correlation4x4 {
                rho_xi_sigma: 0.0,
                rho_xi_d: 0.0,
                rho_xi_f: 0.0,
                rho_sigma_d: 0.0,
                rho_sigma_f: 0.0,
                rho_d_f: 0.0,
            },
        };
        let t = 1.0_f64;
        let chf = FxHhw1ForwardChf::with_simpson_steps(&p, t, 256);
        let pricer = CosPricer::new(&chf).with_n_terms(256);

        // Domestic discount for a constant rd: Pd(0,T) = exp(-rd·T).
        let discount = (-p.rd_0 * t).exp();
        // Forward: F = S₀·exp((rd-rf)·T); BS takes F, strike, discount, σ, T.
        let fwd = p.fx_0 * ((p.rd_0 - p.rf_0) * t).exp();
        let sigma = p.heston.sigma_0.sqrt();

        for strike in [1.20_f64, 1.35, 1.50] {
            let cos = pricer.call(strike, discount);
            let bs = bs_call_forward(fwd, strike, sigma, t, discount);
            let err = (cos - bs).abs();
            assert!(
                err < 1.0e-4,
                "K={}: COS {} vs BS {}, err {}",
                strike,
                cos,
                bs,
                err
            );
        }
    }

    /// Put-call parity in the forward-measure framework:
    /// `C − P = Pd(0,T) · (F − K)` where `F = FX_T(0)`.
    #[test]
    fn put_call_parity_holds() {
        let p = paper_params();
        let t = 1.0_f64;
        let chf = FxHhw1ForwardChf::new(&p, t);
        let pricer = CosPricer::new(&chf);
        let discount = (-p.rd_0 * t).exp();
        let fwd = p.fx_0 * ((p.rd_0 - p.rf_0) * t).exp();
        for strike in [1.20_f64, 1.35, 1.45] {
            let c = pricer.call(strike, discount);
            let q = pricer.put(strike, discount);
            let parity = discount * (fwd - strike);
            let diff = (c - q - parity).abs();
            assert!(
                diff < 1.0e-4,
                "K={}: C-P={}, parity={}, diff {}",
                strike,
                c - q,
                parity,
                diff
            );
        }
    }

    /// Convergence: doubling `N` should leave the price essentially
    /// unchanged at reasonable parameter sets.
    #[test]
    fn cos_converges_with_n_terms() {
        let p = paper_params();
        let t = 1.0_f64;
        let chf = FxHhw1ForwardChf::new(&p, t);
        let discount = (-p.rd_0 * t).exp();
        let p128 = CosPricer::new(&chf).with_n_terms(128).call(1.35, discount);
        let p512 = CosPricer::new(&chf).with_n_terms(512).call(1.35, discount);
        assert!((p128 - p512).abs() < 1.0e-6, "{} vs {}", p128, p512);
    }

    /// Monte Carlo agreement in the deterministic-rate limit. Pricer and
    /// simulator agree on an ATM European call to within MC noise.
    #[test]
    fn cos_matches_monte_carlo_in_deterministic_rate_limit() {
        let mut p = paper_params();
        p.domestic.sigma = 0.0;
        p.foreign.sigma = 0.0;
        p.correlations.rho_xi_d = 0.0;
        p.correlations.rho_xi_f = 0.0;
        p.correlations.rho_sigma_d = 0.0;
        p.correlations.rho_sigma_f = 0.0;
        p.correlations.rho_d_f = 0.0;
        let t = 0.5_f64;
        let chf = FxHhw1ForwardChf::new(&p, t);
        let pricer = CosPricer::new(&chf).with_n_terms(256);
        let discount = (-p.rd_0 * t).exp();
        let strike = 1.35;

        let cos_call = pricer.call(strike, discount);

        // MC: E_Q[exp(-rd·T) · max(ξ(T) - K, 0)]. Under constant rates
        // this is the forward-measure expectation times Pd(0,T).
        let mut sim = FxHhwSimulator::new(p, 2020).unwrap();
        let n_paths = 50_000_usize;
        let terminals = sim.simulate(t, 200, n_paths);
        let mc_call = discount
            * terminals
                .iter()
                .map(|s| (s.fx - strike).max(0.0))
                .sum::<f64>()
            / n_paths as f64;
        let se = discount
            * (terminals
                .iter()
                .map(|s| ((s.fx - strike).max(0.0) - mc_call / discount).powi(2))
                .sum::<f64>()
                / (n_paths as f64).powi(2))
            .sqrt();
        let err = (cos_call - mc_call).abs();
        assert!(
            err < 4.0 * se + 5.0e-4,
            "COS {} vs MC {} — err {} > 4·SE+5e-4 = {}",
            cos_call,
            mc_call,
            err,
            4.0 * se + 5.0e-4
        );
    }

    /// COS prices from the FX-HLMM1 ChF. In the pure-Heston limit
    /// (LMM vol → 0), HLMM and HHW should give identical prices.
    #[test]
    fn cos_works_with_fx_hlmm1_chf_in_pure_heston_limit() {
        use crate::models::forex::fx_hlmm::{
            DdSvLmm, FxHlmmCorrelations, FxHlmmParams, LiborTenor,
        };
        use crate::models::forex::fx_hlmm1_chf::FxHlmm1ForwardChf;

        let p_hhw = FxHhwParams {
            fx_0: 1.35,
            heston: CirProcess {
                kappa: 0.5,
                theta: 0.04,
                gamma: 0.3,
                sigma_0: 0.04,
            },
            domestic: HullWhite1F {
                mean_reversion: 0.01,
                sigma: 0.0,
            },
            foreign: HullWhite1F {
                mean_reversion: 0.05,
                sigma: 0.0,
            },
            rd_0: 0.02,
            rf_0: 0.02,
            theta_d: 0.02,
            theta_f: 0.02,
            correlations: Correlation4x4 {
                rho_xi_sigma: -0.3,
                rho_xi_d: 0.0,
                rho_xi_f: 0.0,
                rho_sigma_d: 0.0,
                rho_sigma_f: 0.0,
                rho_d_f: 0.0,
            },
        };

        // Construct a matching FX-HLMM with LMM vols set to ~zero so both
        // ChFs describe the same process (pure Heston with flat rates).
        let tenor = LiborTenor::new(vec![0.0, 0.5, 1.0], vec![0.02, 0.02]);
        let lmm = DdSvLmm {
            sigmas: vec![1e-8; 2],
            betas: vec![0.5; 2],
            lambda: 1.0,
            eta: 1e-8,
            v_0: 1.0,
            libor_corr: vec![vec![1.0, 0.0], vec![0.0, 1.0]],
        };
        let p_hlmm = FxHlmmParams {
            fx_0: 1.35,
            heston: p_hhw.heston,
            tenor,
            domestic: lmm.clone(),
            foreign: lmm,
            correlations: FxHlmmCorrelations {
                rho_xi_sigma: -0.3,
                rho_xi_d: vec![0.0; 2],
                rho_xi_f: vec![0.0; 2],
                libor_cross_corr: vec![vec![0.0; 2]; 2],
            },
        };

        let t = 1.0_f64;
        let chf_hhw = FxHhw1ForwardChf::new(&p_hhw, t);
        let chf_hlmm = FxHlmm1ForwardChf::new(&p_hlmm, t);
        let pricer_hhw = CosPricer::new(&chf_hhw).with_n_terms(256);
        let pricer_hlmm = CosPricer::new(&chf_hlmm).with_n_terms(256);
        let discount = (-p_hhw.rd_0 * t).exp();

        for strike in [1.20_f64, 1.35, 1.50] {
            let hhw = pricer_hhw.call(strike, discount);
            let hlmm = pricer_hlmm.call(strike, discount);
            // Both models should agree on the pure-Heston limit to a few
            // basis points — there's a slight forward-offset difference
            // (HLMM derives rates from the Libor grid) so allow ~5e-3.
            assert!(
                (hhw - hlmm).abs() < 5.0e-3,
                "K={}: HHW {} vs HLMM {}, diff {}",
                strike,
                hhw,
                hlmm,
                hhw - hlmm
            );
        }
    }
}