import sys, numpy as np, matplotlib.pyplot as plt
sys.path.insert(0,'..')
from engine import ch04
from dataclasses import replaceMFAA Chapter 4 Laboratory
Stochastic DCF Engine (book §4.9)
The book’s first full valuation engine: LDCF operator of Theorem 4.8, premium location, regimes, Bayesian parameter uncertainty. Designed as a hurdle-rate autopsy. Seed 20260400.
1. E1 — Hurdle autopsy
Enter a flat 12% and see what the point estimate was hiding.
p = ch04.DCFParams()
au = ch04.hurdle_autopsy(p)
print(f"flat 12% deterministic value: {au['flat_value']:.4f}")
print(f"full-model mean: {au['full']['mean']:.4f} ± {au['full']['se']:.4f}")
print(f"implied honest flat rate: {au['implied']['implied_flat_rate']:.4f}")
import pandas as pd
pd.DataFrame(au['decomposition'])flat 12% deterministic value: 0.9554
full-model mean: 1.1504 ± 0.0028
implied honest flat rate: 0.0758| stage | mean | se | std | quantiles | |
|---|---|---|---|---|---|
| 0 | deterministic DCF | 1.403467 | 1.570132e-18 | 2.220502e-16 | {'q05': 1.4034674333159036, 'q25': 1.403467433... |
| 1 | + rate risk | 1.247307 | 1.684262e-03 | 2.381906e-01 | {'q05': 0.8632866286499349, 'q25': 1.046989656... |
| 2 | + price-of-risk wedges | 1.154518 | 2.512370e-03 | 3.553028e-01 | {'q05': 0.7591024275595836, 'q25': 0.900515645... |
| 3 | + liquidity friction/factor | 1.150428 | 2.840822e-03 | 4.017529e-01 | {'q05': 0.7194574347974299, 'q25': 0.875777502... |
2. The valuation distribution the hurdle was hiding
from engine.ch04 import _simulate_value
v = _simulate_value(p, 'full')
plt.hist(v, bins=60, color='#1E2761', alpha=.85)
plt.axvline(v.mean(), color='crimson', label=f'mean {v.mean():.3f}')
plt.axvline(au['flat_value'], color='orange', label=f"12% point est {au['flat_value']:.3f}")
plt.legend(); plt.title('Valuation distribution vs the single hurdle number');
3. E2 — Double-count demonstration
The single most opinion-changing exercise in Part II: value illiquidity once as a factor in M, once as a friction in the operator, then force both on.
dc = ch04.double_count_check(p)
print(f"factor-in-M: {dc['factor_in_M']['mean']:.4f}")
print(f"friction-in-operator: {dc['friction_in_operator']['mean']:.4f}")
print(f"both active (bug): {dc['both_active']['mean']:.4f}")
print(f"double-count bias: {dc['double_count_bias_vs_factor']:.4f}")factor-in-M: 1.1504
friction-in-operator: 1.1324
both active (bug): 1.1324
double-count bias: -0.01804. E4 — Learning curve (Bayesian panel, Proposition 4.7)
Feed deal outcomes one at a time; watch the posterior premium spread shrink.
bu = ch04.bayesian_update(0.03, 0.02, 0.05, [0.05,0.01,0.04,0.02,0.06,0.03,0.045])
for n,(mn,sd) in enumerate(zip(bu['posterior_mean'], bu['posterior_sd']),1):
print(f"n={n}: posterior {mn:.4f} ± {sd:.4f}")
print('recursion matches batch:', bu['recursion_matches_batch'])n=1: posterior 0.0328 ± 0.0186
n=2: posterior 0.0300 ± 0.0174
n=3: posterior 0.0311 ± 0.0164
n=4: posterior 0.0300 ± 0.0156
n=5: posterior 0.0327 ± 0.0149
n=6: posterior 0.0324 ± 0.0143
n=7: posterior 0.0334 ± 0.0137
recursion matches batch: True5. Zero-coupon curve (closed form vs Monte Carlo) and validation
mats=[1,2,3,5,7,10]
cf=ch04.zcb_closed_form(p,mats); mc=ch04.zcb_monte_carlo(replace(p,M=40000),mats)
for i,T in enumerate(mats): print(f"T={T:2d} closed {cf[i]:.5f} MC {mc['price'][i]:.5f} ± {mc['se'][i]:.5f}")
v = ch04.validation_checks()
print('\nvalidation ALL:', v['all_pass'])T= 1 closed 0.96947 MC 0.96941 ± 0.00003
T= 2 closed 0.93841 MC 0.93826 ± 0.00007
T= 3 closed 0.90738 MC 0.90714 ± 0.00011
T= 5 closed 0.84674 MC 0.84653 ± 0.00019
T= 7 closed 0.78902 MC 0.78877 ± 0.00024
T=10 closed 0.70890 MC 0.70868 ± 0.00030
validation ALL: True