atlaie

Untitled

Jun 9th, 2025
92
0
Never
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
Python 2.66 KB | None | 0 0
  1. import numpy as np
  2. import matplotlib.pyplot as plt
  3. import seaborn as sns
  4.  
  5. # --- Parameters --------------------------------------------------------------
  6. N = 2
  7. lambda_A = 3
  8. alpha_P, beta_P = 2, 38
  9. mu_I, sigma_I = 13.8, 0.9
  10. N_SIM = 100_000
  11. rng = np.random.default_rng(42)
  12.  
  13. # --- Helper to summarise a loss vector ---------------------------------------
  14. def loss_stats(losses: np.ndarray, label: str) -> dict:
  15.     """Return key metrics for one loss distribution."""
  16.     return {
  17.         "Model": label,
  18.         "Mean [$ M]": losses.mean() / 1e6,
  19.         "Median [$ M]": np.median(losses) / 1e6,
  20.         "VaR95 [$ M]": np.quantile(losses, 0.95) / 1e6,
  21.         "VaR99 [$ M]": np.quantile(losses, 0.99) / 1e6,
  22.         "P(L=0)": (losses == 0).mean(),
  23.     }
  24.  
  25. # --- Simulate independent‑product model (S1) ---------------------------------
  26. losses_s1 = np.empty(N_SIM)
  27. for s in range(N_SIM):
  28.     A = rng.poisson(lambda_A)
  29.     P = rng.beta(alpha_P, beta_P)
  30.     I = rng.lognormal(mu_I, sigma_I)
  31.     losses_s1[s] = N * A * P * I
  32.  
  33. # --- Simulate generative Binomial model (S2) ---------------------------------
  34. losses_s2 = np.empty(N_SIM)
  35. for s in range(N_SIM):
  36.     total = 0.0
  37.     for _ in range(N):
  38.         A_i = rng.poisson(lambda_A)
  39.         if A_i == 0:
  40.             continue
  41.         P_i = rng.beta(alpha_P, beta_P)
  42.         successes = rng.binomial(A_i, P_i)
  43.         if successes:
  44.             impacts = rng.lognormal(mu_I, sigma_I, size=successes)
  45.             total += impacts.sum()
  46.     losses_s2[s] = total
  47.  
  48. # --- Collect results ---------------------------------------------------------
  49. results = pd.DataFrame([
  50.     loss_stats(losses_s1, "Independent product"),
  51.     loss_stats(losses_s2, "Generative (Binomial layer)")
  52. ]).set_index("Model")
  53.  
  54. "Comparison of loss models", results.round(3)
  55.  
  56. # Convert to millions
  57. losses_s1_m = losses_s1 / 1e6
  58. losses_s2_m = losses_s2 / 1e6
  59.  
  60. # --- Plot density histograms -------------------------------------------------
  61. plt.figure(figsize=(9,8))
  62. bins = np.linspace(0, 8, 160)  # up to $8 M covers >99% of mass
  63.  
  64. plt.hist(losses_s1_m, bins=bins, density=True, alpha=0.6, label='S1: Independent product')
  65. plt.hist(losses_s2_m, bins=bins, density=True, alpha=0.8, label='S2: Generative Binomial')
  66.  
  67. plt.xlabel('Annual loss (USD millions)', fontsize = 16)
  68. plt.ylabel('Density',fontsize = 16)
  69. #plt.gca().set_xscale('log')
  70. plt.legend(fontsize = 14, labelcolor = 'linecolor', frameon = False)
  71. #plt.xlim(1, 8)
  72. #plt.ylim(0, 0.15)
  73. plt.xticks(fontsize = 14)
  74. plt.yticks(fontsize = 14)
  75. plt.axvline(np.percentile(losses_s1_m, 95), ls = '--')
  76. plt.axvline(np.percentile(losses_s2_m, 95), color = 'orange', ls = '--')
  77. plt.tight_layout()
Advertisement
Add Comment
Please, Sign In to add comment