关于「退款:争议发生之前的最优拦截阈值」,《收单风控阶跃与准备金生存》给出的定量结论是什么?

结论:第四章求出争议前退款的阈值,示意参数下为 0.443,并比较严格、全退与定向三种政策。

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。

English version

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。 与结论吻合的打印值: 0.443

A. refund-now threshold q* (all disputes lost)
price STEP0 STEP=20 STEP=60
100.367 0.220 0.122
300.620 0.443 0.282
1000.842 0.721 0.561
3000.941 0.885 0.792
B. price 30, STEP 20: q* by dispute win rate W
W0.0 q* = 0.443
W0.2 q* = 0.517
W0.4 q* = 0.620
W0.6 q* = 0.775
C. expected cost of one complaint, price 30, STEP 20
q refund now wait
0.131.00 7.00
0.331.00 21.00
0.531.00 35.00
0.731.00 49.00

核心公式

\[q > q^* = \frac{P + 1}{P + 15 + 5 + \text{STEP}}.\]
查看完整验算程序
# Chapter 4. p41: when does a refund before a dispute cost less?
# A buyer complains about a digital order of price PRICE. If we refund
# now we give back PRICE and lose the processing fee FEE that is not
# returned. If we do not, the buyer opens a dispute with probability q;
# a lost dispute costs PRICE + CB_FEE (dispute fee) + OPS (evidence
# work) + STEP (the marginal expected cost of one more dispute on the
# ratio band, from chapter 1). We assume, against ourselves, that every
# dispute is lost; the win rate W enters in part B.
# Refund now is cheaper when q > q* = (PRICE + FEE) /
#   (PRICE + CB_FEE + OPS + STEP) for W = 0, and in general when
#   q (1 - W) (PRICE + CB_FEE + OPS + STEP) + q W (CB_FEE + OPS) >
#   PRICE + FEE.
# Units are illustrative and carry no currency.

FEE, CB_FEE, OPS = 1.0, 15.0, 5.0


def q_star(price, step, win=0.0):
    lose = price + CB_FEE + OPS + step
    keep = CB_FEE + OPS
    per_q = (1.0 - win) * lose + win * keep
    return min(1.0, (price + FEE) / per_q)


def main():
    print("A. refund-now threshold q* (all disputes lost)")
    print("  price   STEP=0   STEP=20   STEP=60")
    for price in (10.0, 30.0, 100.0, 300.0):
        row = [q_star(price, s) for s in (0.0, 20.0, 60.0)]
        print("  {:>5.0f}   {:.3f}    {:.3f}     {:.3f}".format(price, *row))
    print("B. price 30, STEP 20: q* by dispute win rate W")
    for win in (0.0, 0.2, 0.4, 0.6):
        print("  W = {:.1f}   q* = {:.3f}".format(win, q_star(30.0, 20.0,
                                                            win)))
    print("C. expected cost of one complaint, price 30, STEP 20")
    print("     q    refund now   wait")
    for q in (0.1, 0.3, 0.5, 0.7):
        wait = q * (30.0 + CB_FEE + OPS + 20.0)
        print("  {:.1f}    {:>7.2f}    {:>7.2f}".format(q, 30.0 + FEE, wait))


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 4 章「退款:争议发生之前的最优拦截阈值」(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}
}

本书其他章节结论