关于「固定种子的订单模拟:四种付款政策在同一批买家身上的账」,《货到付款押金与预付筛选》给出的定量结论是什么?

结论:第六章以固定种子模拟四种政策,押金规则加申诉复核比只做货到付款多赚 3535.44。

数据口径:本书把一单的账拆开:示意参数下,签收一单的毛利为 4.5342,拒收一单要付出双程运费、货损与资金占用共 4.3522,盈亏平衡拒收率为 0.4202;在买家自愿付款的保守假设下,追讨要打平需要联系率与付款率之积达到 0.2068,现实中很难做到。 东南亚拒收率没有找到经同行评议的估计,书中参数全部为示意,全部计算由书中程序生成。

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。 程序只用 Python 标准库与结论吻合的打印值: 3535.44

buyers 4000 (committed 2995, casual 1005)
pol orders fails rate profit total per order flag H flag L
P031843 3949 0.1240 118763.04 3.7296 0.0000 0.0000
P131843 3569 0.1121 120472.25 3.7833 0.0000 0.0000
P231277 3204 0.1024 122358.96 3.9121 0.0097 0.4876
P331285 3211 0.1026 122298.48 3.9092 0.0060 0.4826
analytic P0 profit per order 3.7134
P1 minus P01709.20 u
P2 minus P03595.92 u
P3 minus P03535.44 u
查看完整验算程序
# Chapter 6. p61: a fixed-seed simulation of four payment policies.
# 4000 buyers, a quarter casual; each places 2 to 14 orders with a
# basket drawn from {10, 15, 20, 30, 45} u. Failures per COD attempt:
# buyer refusal 0.025 (committed) or 0.325 (casual), buyer absent
# 0.015, carrier or address error 0.01; a fifth of the last two are
# wrongly recorded as refusals. Trust cost phi is uniform on
# [0, 0.16]; surplus per order uniform on [0, 6] u. Policies:
#   P0 COD only;  P1 COD plus a prepay discount of 0.0278 of the basket
#   (p41); P2 = P1 plus the deposit rule of p32 from the buyer's own
#   record (flagged buyers pay a deposit, deterred as in p51, or leave
#   when the expected deposit cost exceeds their surplus); P3 = P2
#   with appeal and human review that strikes wrongly recorded
#   refusals from the record. Costs follow p13. The same random draws
#   serve every policy. Illustrative parameters throughout.
import math
import random

SO = SR = 1.8
RATE, DR, DF, FEE, A, E = 0.12, 21, 10, 0.03, 0.8, 0.015
QH, PHI, RHO, H, THETA = 0.05, 0.08, 0.6, 1.8, 6.0
QD = (0.025, 0.325)
QA, QC, MIS = 0.015, 0.01, 0.2
DISC = 0.0278


def costs(p):
    c = 0.65 * p
    m = p - c - SO - FEE * p - p * RATE * DF / 365
    k = math.fsum([SO, SR, 0.05 * c, (c + SO) * RATE * DR / 365])
    return c, m, k


def dstar(p, ql):
    _, _, k = costs(p)
    d0 = (ql * k + A) / (1 - ql) - (QH * k + A) / (1 - QH)
    return (1 - ql) * d0 / (ql + PHI - (1 - ql) * QH / (1 - QH))


def deposit(p, n, k):
    qhat = (k + 1) / (n + 10)
    if n < 5 or qhat <= 0.15:
        return 0.0
    return math.floor(min(dstar(p, qhat), costs(p)[2]) * 2) / 2


def draw(seed):
    rng = random.Random(20261113)
    rng.seed(seed)
    buyers = []
    for _ in range(4000):
        t = 1 if rng.random() < 0.25 else 0
        phi, surplus = rng.uniform(0, 0.16), rng.uniform(0, 6)
        orders = [(rng.choice((10, 15, 20, 30, 45)), rng.random(),
                   rng.random(), rng.random())
                  for _ in range(rng.randint(2, 14))]
        buyers.append((t, phi, surplus, orders))
    return buyers


def run(buyers, pol):
    prof, n_ord, n_fail, flagged = [], 0, 0, [0, 0]
    for t, phi, sur, orders in buyers:
        rec_n, rec_k, was = 0, 0, False
        for p, u1, u2, u3 in orders:
            c, m, k = costs(p)
            d = DISC * p if pol >= 1 else 0.0
            qtot = QD[t] + QA + QC
            if pol >= 1 and d >= phi * (p - d) + RHO * qtot * H:
                n_ord = n_ord + 1
                if u1 < RHO * QD[t]:
                    prof.append(-math.fsum([SO, 0.05 * c, A, E * (p - d),
                                            (c + SO) * RATE * DR / 365]))
                else:
                    prof.append(p - d - c - SO - E * (p - d) - A)
                continue
            dep = deposit(p, rec_n, rec_k) if pol >= 2 else 0.0
            if dep > 0:
                was = True
                if (qtot + phi) * dep > sur:
                    break
            qd = QD[t] * math.exp(-dep / THETA)
            n_ord, rec_n = n_ord + 1, rec_n + 1
            if u1 < qd:
                n_fail, rec_k = n_fail + 1, rec_k + 1
                prof.append(dep - k - A)
            elif u1 < qd + QA + QC:
                n_fail = n_fail + 1
                if u2 < MIS and pol != 3:
                    rec_k = rec_k + 1
                prof.append(-k - A)
            else:
                prof.append(m - A)
        if was:
            flagged[t] = flagged[t] + 1
    return math.fsum(prof), n_ord, n_fail, flagged


if __name__ == "__main__":
    buyers = draw(20261113)
    nt = [0, 0]
    for b in buyers:
        nt[b[0]] = nt[b[0]] + 1
    print("buyers {} (committed {}, casual {})".format(
        len(buyers), nt[0], nt[1]))
    print("pol orders  fails  rate    profit total per order flag H flag L")
    for pol in range(4):
        tot, no, nf, fl = run(buyers, pol)
        print("P{} {:6d} {:6d} {:6.4f} {:13.2f} {:9.4f} {:6.4f} {:6.4f}"
              .format(pol, no, nf, nf / no, tot, tot / no, fl[0] / nt[0],
                      fl[1] / nt[1]))
    ana = []
    for p in (10, 15, 20, 30, 45):
        _, m, k = costs(p)
        for t, w in ((0, 0.75), (1, 0.25)):
            q = QD[t] + QA + QC
            ana.append(0.2 * w * ((1 - q) * m - q * k - A))
    print("analytic P0 profit per order {:.4f}".format(math.fsum(ana)))
    base = run(buyers, 0)[0]
    for pol in (1, 2, 3):
        print("P{} minus P0: {:9.2f} u".format(pol, run(buyers, pol)[0]
                                                - base))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《货到付款拒收博弈与东南亚电商信任赤字、押金阈值与预付折扣激励设计母典》第 6 章「固定种子的订单模拟:四种付款政策在同一批买家身上的账」(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-314 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026314,
  author    = {Fan, Yafu},
  title     = {Cash-on-Delivery Refusal Games and the Trust Deficit in Southeast Asian E-Commerce: Deposit Thresholds and Prepayment-Discount Incentive Design},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论