关于「实证闭环:月度链上仓位体检、蒸馏小工具与预注册证伪」,《做市损耗与三明治上界》给出的定量结论是什么?
结论:第八章给出月度链上仓位体检、蒸馏小工具「做市损耗与滑点体检器」与预注册证伪方案,主检验需要 56 个头寸月。本书只做防御与风险度量,不写任何抢跑或三明治机器人的搭建与运营。书中以标注框区分已发表的证据、可运行的 Python 计算(20 段程序)、原典与作者的假说。
数据口径:书中参数为示意,全部计算由书中程序生成。 本书是非个人化投资风险分析,不构成投资建议,不推荐任何代币、协议或交易;加密资产高风险,可能全部损失。
对照数据表
下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 56
| Test 1: mean ln(coverage) < 0, sd 0.6, alpha 0.05, power 0.80 | |
| delta n (indep.) design effect (m | 3) n positions x 3 |
| 0.1 | 223 1.6 119 |
| 0.2 | 56 1.6 30 |
| 0.3 | 25 1.6 14 |
| Test 2: sign test on coverage < 1 | |
| p1 0.60: n | 158, reject if at least 90 below 1 |
| p1 0.65: n | 69, reject if at least 42 below 1 |
| p1 0.70: n | 37, reject if at least 24 below 1 |
| Test 3: simulated power of test 1 at delta 0.2 | |
| n | 56, power 0.790 |
查看完整验算程序
# Chapter 8. p82: sample sizes for the pre-registered refutation test.
# Test 1 (core claim): over registered position-months, the mean of
# ln(coverage ratio) is below 0. One-sided test at alpha = 0.05 with
# power 0.80 against a true mean of -delta, monthly sd 0.6 (assumed):
# n = ((z_a + z_b) sd / delta)^2, inflated by the design effect
# 1 + (m - 1) rho for m months per position, rho = 0.3.
# Test 2 (sign test): share of position-months with coverage below 1;
# null 0.5, alternative p1; smallest n with exact binomial size <= 0.05
# and power >= 0.80. Test 3: a fixed-seed simulation checks the power
# of test 1 at the computed n. All inputs illustrative.
import math
import random
import statistics
ZA, ZB = 1.6449, 0.8416
def n_mean(delta, sd):
return math.ceil(((ZA + ZB) * sd / delta) ** 2)
def binom_tail(n, k, p):
terms = [math.comb(n, i) * p ** i * (1 - p) ** (n - i)
for i in range(k, n + 1)]
return math.fsum(terms)
def n_sign(p1):
for n in range(10, 2000):
k = next(k for k in range(n + 1) if binom_tail(n, k, 0.5) <= 0.05)
if binom_tail(n, k, p1) >= 0.80:
return n, k
return None, None
def sim_power(rng, n, delta, sd, reps=4000):
hits = 0
for _ in range(reps):
xs = [rng.gauss(-delta, sd) for _ in range(n)]
t = statistics.fmean(xs) / (statistics.stdev(xs) / math.sqrt(n))
if t < -ZA:
hits += 1
return hits / reps
if __name__ == "__main__":
rng = random.Random(20261120)
print("Test 1: mean ln(coverage) < 0, sd 0.6, alpha 0.05, power 0.80")
print(" delta n (indep.) design effect (m=3) n positions x 3")
for delta in (0.1, 0.2, 0.3):
n = n_mean(delta, 0.6)
deff = 1 + (3 - 1) * 0.3
npos = math.ceil(n * deff / 3)
print(" {:.1f} {:>11,} {:>20.1f} {:>17,}".format(
delta, n, deff, npos))
print("Test 2: sign test on coverage < 1")
for p1 in (0.60, 0.65, 0.70):
n, k = n_sign(p1)
print(" p1 {:.2f}: n = {:>4,}, reject if at least {:>4,} below 1"
.format(p1, n, k))
print("Test 3: simulated power of test 1 at delta 0.2")
n = n_mean(0.2, 0.6)
print(" n = {}, power {:.3f}".format(n, sim_power(rng, n, 0.2, 0.6)))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《集中流动性做市商重平衡损耗(LVR)、链上三明治抢跑边界与去中心化金库风控母典》第 8 章「实证闭环:月度链上仓位体检、蒸馏小工具与预注册证伪」(版本 1.0.0). 5D Library. DOI 待发布
DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-344 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026344,
author = {Fan, Yafu},
title = {Concentrated-Liquidity Loss-Versus-Rebalancing, On-Chain Sandwich Bounds and Decentralised Vault Risk Control},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}