关于「准备金:滚动扣留、结算冻结与现金不断血的最低缓冲」,《收单风控阶跃与准备金生存》给出的定量结论是什么?
结论:第六章算出一成、九十天的滚动准备金只需 2,900 的缓冲,而冻结三十天需要 93,000,缓冲留足 50,000 时仍有 0.0175 的年份会用光。
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: 2900, 93000
| scenario | buffer | worst day | held peak |
|---|---|---|---|
| no measure | 0 | - | 0 |
| reserve 10% x 90d from day 1 | 2,900 | 32 | 28,500 |
| reserve 10% x 180d from day 1 | 2,900 | 32 | 42,000 |
| reserve 25% x 180d from day 1 | 13,500 | 90 | 105,000 |
| reserve 10% x 90d from day 45 | 0 | - | 25,500 |
| reserve 25% x 180d from day 45 | 4,600 | 90 | 97,500 |
| freeze 14d from day 45 | 43,400 | 58 | 0 |
| freeze 30d from day 45 | 93,000 | 74 | 0 |
| freeze 30d, spend cut day 47 | 59,400 | 74 | 0 |
核心公式
\[Z = \left(\frac{3\,B\,\text{SIG}^2}{4\,I}\right)^{1/3},\]
查看完整验算程序
# Chapter 6. p61: the cash a shop must hold before a surge, by scenario.
# Daily takings: 1,000 on days 1-30, 4,000 on days 31-90 (a campaign),
# 1,500 on days 91-365. Settlement two days after the sale. Costs paid
# the same day: variable 0.65 of takings during the campaign (paid
# advertising, delivery) and 0.35 otherwise, plus a fixed 500 a day.
# Scenarios (the acquirer's measures are imposed, not chosen):
# rolling reserve r of each settlement, released T days later,
# either from day 1 or imposed on day 45; a payout freeze of 14 or 30
# days from day 45 (held payouts paid at the end of the freeze),
# with the campaign spend kept, or cut to the base share on day 47.
# Buffer = the largest fall of cash below its level on the eve of the
# measure (day 0 when there is no measure): the cash that must already
# sit aside, whatever profit came before, so that no bill goes unpaid.
# All amounts are illustrative units without currency.
import math
DAYS, LAG, FIXED = 365, 2, 500.0
def takings(t):
if t <= 30:
return 1000.0
if t <= 90:
return 4000.0
return 1500.0
def var_share(t, cut_day=None):
if 31 <= t <= 90 and (cut_day is None or t < cut_day):
return 0.65
return 0.35
def path(r=0.0, hold=0, r_from=1, frz_from=None, frz_len=0, cut=None):
cash, low, low_day = 0.0, 0.0, 0
event = frz_from or (r_from if r > 0 else 1)
eve = 0.0
release = {}
frozen = 0.0
held_peak, held = 0.0, 0.0
for t in range(1, DAYS + 1):
flows = [-FIXED, -var_share(t, cut) * takings(t)]
gross = takings(t - LAG) # days before 1 sold 1,000 a day
keep = r * gross if t >= r_from else 0.0
if keep:
release[t + hold] = release.get(t + hold, 0.0) + keep
held = held + keep
pay = gross - keep
if frz_from and frz_from <= t < frz_from + frz_len:
frozen = frozen + pay
else:
flows.append(pay)
if frz_from and t == frz_from + frz_len:
flows.append(frozen)
frozen = 0.0
if t in release:
flows.append(release[t])
held = held - release[t]
held_peak = max(held_peak, held)
if t == event:
eve = cash
cash = cash + math.fsum(flows)
if t >= event and cash - eve < low:
low, low_day = cash - eve, t
return 0.0 - low, low_day, held_peak
def main():
rows = [("no measure", {}),
("reserve 10% x 90d from day 1", dict(r=0.10, hold=90)),
("reserve 10% x 180d from day 1", dict(r=0.10, hold=180)),
("reserve 25% x 180d from day 1", dict(r=0.25, hold=180)),
("reserve 10% x 90d from day 45",
dict(r=0.10, hold=90, r_from=45)),
("reserve 25% x 180d from day 45",
dict(r=0.25, hold=180, r_from=45)),
("freeze 14d from day 45", dict(frz_from=45, frz_len=14)),
("freeze 30d from day 45", dict(frz_from=45, frz_len=30)),
("freeze 30d, spend cut day 47",
dict(frz_from=45, frz_len=30, cut=47))]
print("scenario buffer worst day held peak")
for name, kw in rows:
need, day, peak = path(**kw)
print("{:<32} {:>8,.0f} {:>5} {:>9,.0f}".format(
name, need, day if day else "-", peak))
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 6 章「准备金:滚动扣留、结算冻结与现金不断血的最低缓冲」(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}
}