关于「跃升:累积和、指数加权平均与贝叶斯在线变点」,《收单风控阶跃与准备金生存》给出的定量结论是什么?
结论:第三章用累积和、指数加权平均与贝叶斯在线变点检测交易额跃升,一年一次误报时,单位标准差的抬升平均 8.5 天被看见。
数据口径:书中参数为示意,全部计算由书中程序生成。 本书是非个人化的经营与风险分析,不构成法律、融资或投资建议,不向任何人承诺收入。
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: 8.5
| upper CUSUM, K = 0.5, SIGMA = 0.25 on log takings | |||||
| target ARL0 | h | ARL1 at delta = 0.5 | 1.0 | 2.0 | 3.0 |
| 100 | 2.849 | 16.1 | 6.1 | 2.6 | 1.7 |
| 365 | 4.083 | 27.5 | 8.5 | 3.4 | 2.2 |
| 730 | 4.761 | 35.1 | 9.9 | 3.8 | 2.5 |
| takings factor: 0.5 -> 1.133, 1.0 -> 1.284, 2.0 -> 1.649, 3.0 -> 2.117 | |||||
| false alarms a year at ARL0 365: about 1.00 |
核心公式
\[S_t = \max(0,\; S_{t-1} + z_t - K),\]
查看完整验算程序
# Chapter 3. p31: Page's CUSUM on daily log takings, threshold by ARL.
# x_t = log(daily takings); in control x_t ~ N(mu, SIGMA), SIGMA = 0.25
# (illustrative). z_t = (x_t - mu) / SIGMA. Upper CUSUM:
# S_t = max(0, S_{t-1} + z_t - K), alarm when S_t > h, K = 0.5.
# Average run lengths (ARL) by the Brook-Evans Markov chain with M
# states, solved by Gaussian elimination. ARL0 = mean days between
# false alarms with no change; ARL1 = mean days to alarm after a step
# of delta standard deviations (delta = 1 is takings x exp(0.25)).
import math
SIGMA, K, M = 0.25, 0.5, 90
def phi(x):
return 0.5 * (1.0 + math.erf(x / math.sqrt(2.0)))
def solve(a, b):
n = len(b)
a = [row[:] + [b[i]] for i, row in enumerate(a)]
for c in range(n):
piv = max(range(c, n), key=lambda r: abs(a[r][c]))
a[c], a[piv] = a[piv], a[c]
for r in range(c + 1, n):
f = a[r][c] / a[c][c]
if f:
for j in range(c, n + 1):
a[r][j] -= f * a[c][j]
x = [0.0] * n
for r in range(n - 1, -1, -1):
tail = math.fsum([a[r][j] * x[j] for j in range(r + 1, n)])
x[r] = (a[r][n] - tail) / a[r][r]
return x
def cusum_arl(h, delta):
w = 2.0 * h / (2 * M - 1)
mat = []
for i in range(M):
ci = i * w
row = []
for j in range(M):
if j == 0:
pij = phi(0.5 * w - ci + K - delta)
else:
pij = (phi((j + 0.5) * w - ci + K - delta)
- phi((j - 0.5) * w - ci + K - delta))
row.append((1.0 if i == j else 0.0) - pij)
mat.append(row)
return solve(mat, [1.0] * M)[0]
def h_for(arl0):
lo, hi = 1.0, 8.0
for _ in range(40):
mid = 0.5 * (lo + hi)
if cusum_arl(mid, 0.0) < arl0:
lo = mid
else:
hi = mid
return 0.5 * (lo + hi)
def main():
print("upper CUSUM, K = 0.5, SIGMA = 0.25 on log takings")
print("target ARL0 h ARL1 at delta = 0.5 1.0 2.0 3.0")
for arl0 in (100, 365, 730):
h = h_for(arl0)
a1 = [cusum_arl(h, d) for d in (0.5, 1.0, 2.0, 3.0)]
print("{:>10} {:.3f} {:>7.1f} {:>5.1f} {:>5.1f}"
" {:>5.1f}".format(arl0, h, *a1))
print("takings factor: " + ", ".join(
"{} -> {:.3f}".format(d, math.exp(d * SIGMA))
for d in (0.5, 1.0, 2.0, 3.0)))
h = h_for(365)
print("false alarms a year at ARL0 365: about {:.2f}".format(
365.0 / cusum_arl(h, 0.0)))
if __name__ == "__main__":
main()唯一数据源与引用
范雅福(Fan, Y.). (2026). 《数字商品收单风控阶跃模型、多网关热备切换与防冻卡滚动准备金母典》第 3 章「跃升:累积和、指数加权平均与贝叶斯在线变点」(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-397 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026397,
author = {Fan, Yafu},
title = {Payment-Acquirer Risk Step-Functions, Multi-Gateway Hot Failover and Rolling-Reserve Survival for Digital Fleets},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}