What does Tail Protection under Kelly Bounds conclude, in numbers, about “Closing the Loop: The Annual Tail-Protection Budget Sheet and a Pre-Registered Test”?

Answer: Chapter 8 gives an annual tail-protection budget sheet, the Tail-Protection Premium Calculator and a pre-registered test plan: backtesting the tail model would need a single record of 1,171 years, so it can be tested only on pooled samples. Marked boxes separate published evidence, runnable Python calculations (20 programs), classical sources and the author's hypotheses.

Scope of the numbers: Marked boxes separate published evidence, runnable Python calculations (20 programs), classical sources and the author's hypotheses.

中文版

Data table

The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 1171

(1) one buyer, years needed to see the drag
  K 0.90: mean -0.294, sd 1.803, years 295
  K 0.80: mean -0.401, sd 2.313, years 262
  K 0.70: mean -0.485, sd 3.014, years 304
(2) P(M < 0.80): no cover 0.0402, cover 0.0059; per arm 300
(3) lognormal VaR 0.975 loss 0.2076; true exceedance rate 0.0376
    years needed: 1,171 (reject if exceedances >= 39)
Show the full program
# Chapter 8. p82: sample sizes for the pre-registered tests.
# (1) Premium drag seen by one buyer: years n so that a 5 % two-sided
#     test of mean(payout / premium - 1) = 0 has power 0.80, normal
#     approximation, moments exact under Merton (P).
# (2) Cover versus no cover across operators: one-year P(M < 0.80) for
#     L = 1.0 without cover and with h = 0.5 at K = 0.90; operator-years
#     per arm for a two-proportion test (5 %, power 0.80).
# (3) Tail-model backtest: a lognormal 97.5 % VaR applied to Merton
#     losses is exceeded more often than 0.025; years of records for an
#     exact one-sided binomial test (5 %) to reach power 0.80.
# Illustrative parameters; quotes are the Q premiums of p11.
import math

MU, SIG = 0.05, 0.18
LAM, JM, JS = 0.10, -0.35, 0.15
A, L = 0.12, 1.0
ZA, ZB = 1.959964, 0.841621


def Phi(x):
    return 0.5 * math.erfc(-x / math.sqrt(2.0))


KAP = math.exp(JM + 0.5 * JS * JS) - 1.0
BASE = MU - 0.5 * SIG * SIG - LAM * KAP
CS = [(math.exp(-LAM) * LAM ** n / math.factorial(n), BASE + n * JM,
       math.sqrt(SIG * SIG + n * JS * JS)) for n in range(13)]


def moments(k):
    # E[(K - S)+] 与 E[((K - S)+)^2] 的闭式
    m1, m2 = [], []
    for w, m, s in CS:
        d = (math.log(k) - m) / s
        e1 = math.exp(m + 0.5 * s * s) * Phi(d - s)
        e2 = math.exp(2 * m + 2 * s * s) * Phi(d - 2 * s)
        m1.append(w * (k * Phi(d) - e1))
        m2.append(w * (k * k * Phi(d) - 2 * k * e1 + e2))
    return math.fsum(m1), math.fsum(m2)


def cdf(x):
    return math.fsum(w * Phi((math.log(x) - m) / s) for w, m, s in CS)


print("(1) one buyer, years needed to see the drag")
for k, pi in ((0.90, 0.04084), (0.80, 0.02032), (0.70, 0.00910)):
    e1, e2 = moments(k)
    mean = e1 / pi - 1.0
    sd = math.sqrt(e2 / pi ** 2 - (e1 / pi) ** 2)
    n = ((ZA + ZB) * sd / mean) ** 2
    print("  K {:.2f}: mean {:+.3f}, sd {:.3f}, years {:.0f}".format(
        k, mean, sd, math.ceil(n)))
h, k, pi = 0.5, 0.90, 0.04084
cut0 = 1.0 - (1.0 + A - 0.80) / L              # 无保护时 M < 0.80 的 S
# 有保护时,S < K 段 M = 1 + A + L(S - 1) + hL(K - S - pi)
cut1 = (0.80 - 1.0 - A + L - h * L * (k - pi)) / (L * (1.0 - h))
p1, p2 = cdf(cut0), cdf(cut1)
pb = 0.5 * (p1 + p2)
nn = (ZA * math.sqrt(2 * pb * (1 - pb))
      + ZB * math.sqrt(p1 * (1 - p1) + p2 * (1 - p2))) ** 2 / (p1 - p2) ** 2
print("(2) P(M < 0.80): no cover {:.4f}, cover {:.4f}; per arm {:.0f}"
      .format(p1, p2, math.ceil(nn)))
mean = math.fsum(w * m for w, m, s in CS)
var = math.fsum(w * (s * s + m * m) for w, m, s in CS) - mean * mean
q_ln = math.exp(mean - 1.959964 * math.sqrt(var))
true_rate = cdf(q_ln)
print("(3) lognormal VaR 0.975 loss {:.4f}; true exceedance rate {:.4f}"
      .format(1.0 - (1.0 + A + L * (q_ln - 1.0)), true_rate))


def pmf(n, p):
    lp, lq = math.log(p), math.log(1.0 - p)
    c = math.lgamma(n + 1)
    return [math.exp(c - math.lgamma(x + 1) - math.lgamma(n - x + 1)
                     + x * lp + (n - x) * lq) for x in range(n + 1)]


for n in range(20, 2001):
    null, alt = pmf(n, 0.025), pmf(n, true_rate)
    c = n + 1
    while c > 0 and math.fsum(null[c - 1:]) <= 0.05:
        c -= 1
    if c <= n and math.fsum(alt[c:]) >= 0.80:
        print("    years needed: {:,} (reject if exceedances >= {})".format(
            n, c))
        break

Single source and citation

Fan, Y. (2026). Tail-Risk Option Moats and Quantile Hedging: Jump-Diffusion Models and Fixed-Premium Protection of Core Cash Flows under Kelly Bounds (Version 1.0.0), Chapter 8: Closing the Loop: The Annual Tail-Protection Budget Sheet and a Pre-Registered Test. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-290 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026290,
  author    = {Fan, Yafu},
  title     = {Tail-Risk Option Moats and Quantile Hedging: Jump-Diffusion Models and Fixed-Premium Protection of Core Cash Flows under Kelly Bounds},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book