What does Martingales, Stopping Times and Stop-Loss Bounds conclude, in numbers, about “The Learning Value of a Stop Line: Lines of Belief, the Gap Between Good and Bad, and Escalating Sunk Costs”?

Answer: Chapter 5 shows that a stop line is a line of belief whose value rises with the gap between good and bad, capturing at most 0.646 of the value of foresight.

Scope of the numbers: Its parameters are illustrative; all calculations are generated by its programs. It is not individual financial advice, does not encourage gambling of any kind, and analyses doubling only as a counter-example.

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: 0.646

prior good 0.3, deadline 26; result units per trial
drun allbeststopperfectcaptured
0.02-0.416-0.086-10.312-0.275
0.05-1.040-0.108-10.780-0.139
0.10-2.0800.104-11.5600.067
0.15-3.1200.569-12.3400.243
0.20-4.1601.278-23.1200.410
0.30-6.2403.025-24.6800.646
a negative best means: do not start, or test more cheaply

Key formula

\[\frac{\pi}{1-\pi}\times 1.5^{s}.\]
Show the full program
# Chapter 5. p52: How much a stop line can learn depends on how far
# apart good and bad trials are, relative to weekly noise.
# Model: good p = 0.5 + d, bad p = 0.5 - d, prior good 0.3 (most trials
# fail), deadline 26 weeks, stop at -b with b chosen best in 1..26.
# Not starting earns 0; running everything to the deadline earns
# 26 x 2d x (2 x 0.3 - 1); perfect foresight (run only the good ones)
# earns 0.3 x 26 x 2d. "captured" = best stop line / perfect foresight.
# Exact fractions; d and the prior are illustrative.
from fractions import Fraction

N = 26
PI = Fraction(3, 10)


def run(p, b):
    q = 1 - p
    live = {0: Fraction(1)}
    es = Fraction(0)
    for t in range(1, N + 1):
        new = {}
        for s, pr in live.items():
            new[s + 1] = new.get(s + 1, Fraction(0)) + pr * p
            new[s - 1] = new.get(s - 1, Fraction(0)) + pr * q
        live = {}
        for s in sorted(new):
            if s <= -b or t == N:
                es += s * new[s]
            else:
                live[s] = new[s]
    return es


print("prior good 0.3, deadline %d; result units per trial" % N)
print("%5s %9s %9s %6s %9s %9s" % (
    "d", "run all", "best", "stop", "perfect", "captured"))
for d in (Fraction(1, 50), Fraction(1, 20), Fraction(1, 10),
          Fraction(3, 20), Fraction(1, 5), Fraction(3, 10)):
    allin = N * 2 * d * (2 * PI - 1)
    best_b, best_e = None, None
    for b in range(1, N + 1):
        e = PI * run(Fraction(1, 2) + d, b) + \
            (1 - PI) * run(Fraction(1, 2) - d, b)
        if best_e is None or e > best_e:
            best_b, best_e = b, e
    perfect = PI * N * 2 * d
    print("%5.2f %9.3f %9.3f %6d %9.3f %9.3f" % (
        float(d), float(allin), float(best_e), -best_b, float(perfect),
        float(best_e / perfect)))
print("a negative best means: do not start, or test more cheaply")

Single source and citation

Fan, Y. (2026). Doob's Martingales and the Optional Stopping Theorem: the Doubling Fallacy and Rigorous Stop-Loss Bounds for Business Trials (Version 1.0.0), Chapter 5: The Learning Value of a Stop Line: Lines of Belief, the Gap Between Good and Bad, and Escalating Sunk Costs. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-249 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026249,
  author    = {Fan, Yafu},
  title     = {Doob's Martingales and the Optional Stopping Theorem: the Doubling Fallacy and Rigorous Stop-Loss Bounds for Business Trials},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book