Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- import numpy as np
- import matplotlib.pyplot as plt
- import seaborn as sns
- # --- Parameters --------------------------------------------------------------
- N = 2
- lambda_A = 3
- alpha_P, beta_P = 2, 38
- mu_I, sigma_I = 13.8, 0.9
- N_SIM = 100_000
- rng = np.random.default_rng(42)
- # --- Helper to summarise a loss vector ---------------------------------------
- def loss_stats(losses: np.ndarray, label: str) -> dict:
- """Return key metrics for one loss distribution."""
- return {
- "Model": label,
- "Mean [$ M]": losses.mean() / 1e6,
- "Median [$ M]": np.median(losses) / 1e6,
- "VaR95 [$ M]": np.quantile(losses, 0.95) / 1e6,
- "VaR99 [$ M]": np.quantile(losses, 0.99) / 1e6,
- "P(L=0)": (losses == 0).mean(),
- }
- # --- Simulate independent‑product model (S1) ---------------------------------
- losses_s1 = np.empty(N_SIM)
- for s in range(N_SIM):
- A = rng.poisson(lambda_A)
- P = rng.beta(alpha_P, beta_P)
- I = rng.lognormal(mu_I, sigma_I)
- losses_s1[s] = N * A * P * I
- # --- Simulate generative Binomial model (S2) ---------------------------------
- losses_s2 = np.empty(N_SIM)
- for s in range(N_SIM):
- total = 0.0
- for _ in range(N):
- A_i = rng.poisson(lambda_A)
- if A_i == 0:
- continue
- P_i = rng.beta(alpha_P, beta_P)
- successes = rng.binomial(A_i, P_i)
- if successes:
- impacts = rng.lognormal(mu_I, sigma_I, size=successes)
- total += impacts.sum()
- losses_s2[s] = total
- # --- Collect results ---------------------------------------------------------
- results = pd.DataFrame([
- loss_stats(losses_s1, "Independent product"),
- loss_stats(losses_s2, "Generative (Binomial layer)")
- ]).set_index("Model")
- "Comparison of loss models", results.round(3)
- # Convert to millions
- losses_s1_m = losses_s1 / 1e6
- losses_s2_m = losses_s2 / 1e6
- # --- Plot density histograms -------------------------------------------------
- plt.figure(figsize=(9,8))
- bins = np.linspace(0, 8, 160) # up to $8 M covers >99% of mass
- plt.hist(losses_s1_m, bins=bins, density=True, alpha=0.6, label='S1: Independent product')
- plt.hist(losses_s2_m, bins=bins, density=True, alpha=0.8, label='S2: Generative Binomial')
- plt.xlabel('Annual loss (USD millions)', fontsize = 16)
- plt.ylabel('Density',fontsize = 16)
- #plt.gca().set_xscale('log')
- plt.legend(fontsize = 14, labelcolor = 'linecolor', frameon = False)
- #plt.xlim(1, 8)
- #plt.ylim(0, 0.15)
- plt.xticks(fontsize = 14)
- plt.yticks(fontsize = 14)
- plt.axvline(np.percentile(losses_s1_m, 95), ls = '--')
- plt.axvline(np.percentile(losses_s2_m, 95), color = 'orange', ls = '--')
- plt.tight_layout()
Advertisement
Add Comment
Please, Sign In to add comment