"""
Unit tests for the FX risk engine (multi-currency).

These run fully offline on synthetic correlated return series — no network,
no live rates — so the maths is validated deterministically.

    python3 -m unittest -v          # or:  python3 test_risk_engine.py
"""

import math
import random
import unittest

import risk_engine as risk

CCY = ["EUR", "SEK"]


def make_returns(n=250, sigma_eur=0.004, sigma_sek=0.006, rho=0.8, seed=1):
    """Generate n days of correlated EUR/SEK log returns via Cholesky."""
    rng = random.Random(seed)
    cov = [
        [sigma_eur ** 2,              rho * sigma_eur * sigma_sek],
        [rho * sigma_eur * sigma_sek, sigma_sek ** 2],
    ]
    L = risk.cholesky(cov)
    eur, sek = [], []
    for _ in range(n):
        r = risk._matvec(L, [rng.gauss(0, 1), rng.gauss(0, 1)])
        eur.append(r[0]); sek.append(r[1])
    return {"EUR": eur, "SEK": sek}


class TestLinearAlgebra(unittest.TestCase):
    def test_cholesky_reconstructs_matrix(self):
        M = [[4.0, 2.0], [2.0, 3.0]]
        L = risk.cholesky(M)
        recon = [[sum(L[i][k] * L[j][k] for k in range(2)) for j in range(2)] for i in range(2)]
        for i in range(2):
            for j in range(2):
                self.assertAlmostEqual(recon[i][j], M[i][j], places=10)

    def test_cholesky_lower_triangular(self):
        self.assertEqual(risk.cholesky([[4.0, 2.0], [2.0, 3.0]])[0][1], 0.0)

    def test_covariance_symmetric(self):
        cov = risk.covariance_matrix(make_returns(), CCY)
        self.assertAlmostEqual(cov[0][1], cov[1][0], places=12)

    def test_correlation_diagonal_and_bounds(self):
        corr = risk.correlation_matrix(risk.covariance_matrix(make_returns(), CCY))
        self.assertAlmostEqual(corr[0][0], 1.0, places=9)
        self.assertTrue(-1.0 <= corr[0][1] <= 1.0)

    def test_correlation_recovers_input(self):
        corr = risk.correlation_matrix(risk.covariance_matrix(make_returns(n=2000, rho=0.8), CCY))
        self.assertAlmostEqual(corr[0][1], 0.8, delta=0.05)

    def test_percentile_interpolates(self):
        xs = [0.0, 1.0, 2.0, 3.0, 4.0]
        self.assertAlmostEqual(risk._percentile(xs, 0.5), 2.0)
        self.assertAlmostEqual(risk._percentile(xs, 0.0), 0.0)
        self.assertAlmostEqual(risk._percentile(xs, 1.0), 4.0)

    def test_cholesky_handles_three_assets(self):
        M = [[4, 2, 1], [2, 3, 0.5], [1, 0.5, 2]]
        L = risk.cholesky(M)
        recon = [[sum(L[i][k] * L[j][k] for k in range(3)) for j in range(3)] for i in range(3)]
        for i in range(3):
            for j in range(3):
                self.assertAlmostEqual(recon[i][j], M[i][j], places=9)


class TestVaR(unittest.TestCase):
    def setUp(self):
        self.returns = make_returns(n=2000, rho=0.8)
        self.cov = risk.covariance_matrix(self.returns, CCY)
        self.exposure = [60.0, 40.0]

    def test_var99_exceeds_var95(self):
        p = risk.parametric_var(self.exposure, self.cov, CCY)
        self.assertGreater(p["levels"]["0.99"], p["levels"]["0.95"])

    def test_component_var_sums_to_total(self):
        # Euler/additive property — compare within the 4-dp rounding tolerance.
        p = risk.parametric_var(self.exposure, self.cov, CCY)
        self.assertAlmostEqual(sum(p["components"].values()), p["levels"]["0.95"], delta=1e-3)

    def test_parametric_and_montecarlo_agree(self):
        p = risk.parametric_var(self.exposure, self.cov, CCY)
        mc = risk.monte_carlo_var(self.exposure, self.cov, n_sims=40_000, seed=7)
        rel = abs(p["levels"]["0.95"] - mc["levels"]["0.95"]) / p["levels"]["0.95"]
        self.assertLess(rel, 0.07)

    def test_expected_shortfall_exceeds_var(self):
        mc = risk.monte_carlo_var(self.exposure, self.cov, n_sims=40_000, seed=3)
        self.assertGreaterEqual(mc["es"]["0.95"], mc["levels"]["0.95"])

    def test_horizon_scales_as_sqrt(self):
        p1 = risk.parametric_var(self.exposure, self.cov, CCY, horizon=1)
        p10 = risk.parametric_var(self.exposure, self.cov, CCY, horizon=10)
        self.assertAlmostEqual(p10["levels"]["0.95"] / p1["levels"]["0.95"], math.sqrt(10), delta=0.01)

    def test_historical_var_positive(self):
        self.assertGreater(risk.historical_var(self.exposure, self.returns, CCY)["levels"]["0.95"], 0)

    def test_montecarlo_histogram_shape(self):
        mc = risk.monte_carlo_var(self.exposure, self.cov, n_sims=20_000, bins=31)
        self.assertEqual(len(mc["histogram"]["centers"]), 31)
        self.assertEqual(sum(mc["histogram"]["counts"]), 20_000)


class TestHedge(unittest.TestCase):
    def setUp(self):
        self.cov = risk.covariance_matrix(make_returns(n=1500, rho=0.8), CCY)

    def test_effectiveness_bounded(self):
        h = risk.best_hedge([60.0, 40.0], self.cov, CCY)["best_single"]
        self.assertTrue(0.0 <= h["effectiveness_pct"] <= 100.0)

    def test_residual_below_unhedged(self):
        h = risk.best_hedge([60.0, 40.0], self.cov, CCY)
        self.assertLessEqual(h["best_single"]["residual_sigma"], h["unhedged_sigma"])

    def test_single_currency_book_fully_hedged(self):
        # A pure-EUR book is perfectly hedged by the EUR/USD forward.
        h = risk.best_hedge([100.0, 0.0], self.cov, CCY)["best_single"]
        self.assertEqual(h["instrument"], "EUR")
        self.assertAlmostEqual(h["effectiveness_pct"], 100.0, delta=0.01)


class TestStress(unittest.TestCase):
    def test_negative_shock_gives_loss(self):
        results = risk.stress_test([60.0, 40.0], CCY)
        gfc = next(r for r in results if "GFC" in r["name"])
        self.assertLess(gfc["impact_usd"], 0)

    def test_riskon_gives_gain(self):
        results = risk.stress_test([60.0, 40.0], CCY)
        self.assertGreater(next(r for r in results if "Risk-on" in r["name"])["impact_usd"], 0)

    def test_jpy_safe_haven_in_gfc(self):
        # JPY exposure should *gain* in the GFC scenario (safe-haven rally).
        results = risk.stress_test([100.0], ["JPY"])
        gfc = next(r for r in results if "GFC" in r["name"])
        self.assertGreater(gfc["impact_usd"], 0)


if __name__ == "__main__":
    unittest.main(verbosity=2)