关于「期末冲量的检测:周度份额检验与折扣率监控」,《提成阶梯与期末冲量》给出的定量结论是什么?
结论:第六章设计周度份额检验与期末折扣率监控,后者只看 4 个人 1 个季度就能以 0.540 的概率发现加速方案。
数据口径:书中的销售员、门店、费率与折扣均为示意参数;全部计算由书中程序生成;本书是非个人化的经营建议,不构成法律、税务或理财建议,涉及法规的表述以上架前人工复核为准。 第四章用蒙特卡洛模拟比较老板利润,线性方案比台阶方案每季度多 0.2266,相当于 0.089;但在同等薪酬与某一努力弹性下,台阶方案反而小胜 0.017,所以命题只能说「往往」。
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: 0.540
| L: discounted share weeks 11-13 0.092, weeks 1-10 0.092 | ||||
| S: discounted share weeks 11-13 0.204, weeks 1-10 0.106 | ||||
| A: discounted share weeks 11-13 0.319, weeks 1-10 0.107 | ||||
| people | quarters | fires L | fires S | fires A |
| 4 | 1 | 0.075 | 0.215 | 0.540 |
| 8 | 1 | 0.065 | 0.282 | 0.760 |
| 4 | 4 | 0.070 | 0.438 | 0.975 |
| 8 | 4 | 0.055 | 0.662 | 1.000 |
查看完整验算程序
# Chapter 6. p62: The end-of-period discount monitor. Besides the pulled
# deals of the timing game, every ordinary deal is discounted with a
# background probability of 0.10 (illustrative). The monitor compares
# the share of discounted deals in weeks 11 to 13 with weeks 1 to 10,
# one-sided two-proportion z-test at the 5% level. Same simulated
# quarters as the shared core (fixed seeds). We print the average
# discount shares and how often the monitor fires for teams of n people
# watched for k quarters (400 team-samples each).
# 期末折扣率监控:期末三周打折单占比与前十周相比(参数示意)。
import math
import random
# ---- shared core (identical in every DP program of this book) ----
# 一个季度 T 周;每周以概率 p 来一单(标价 1),p 由努力决定,
# 在 [P0, P1] 之间,努力成本为 k*(p-P0)^2/2。来单可当季成交,
# 或压到下季度(每单以 hloss 的概率流失)。最后 pullw 周,
# 每周以 acc 的概率可让下季度的一单打 disc 的折扣提前成交。
# 提前成交的单子下季度不再出现(下季度从负数起计)。
T, P0, P1, GM = 13, 0.6, 0.9, 0.30
BMAX, SMAX = 3, 24
BASE = dict(k=0.2, acc=0.5, disc=0.15, pullw=3, hloss=0.1,
beta=0.9, hmax=4)
LIN = (0.05, 0.05, 10, 0.0) # rate below, rate above, quota, bonus
STEP = (0.03, 0.03, 10, 0.25)
ACCEL = (0.03, 0.08, 10, 0.0)
def inc(plan, k, price):
# commission paid for the k-th deal of the quarter at this price
lo, hi, q, b = plan
if k <= 0:
return 0.0
return (lo if k <= q else hi) * price + (b if k == q else 0.0)
def backward(plan, vst, m):
hmax = m["hmax"]
sn, cn = range(-BMAX, SMAX + 1), range(-BMAX, hmax + 1)
nxt = {(s, c): m["beta"] * vst[c] for s in sn for c in cn}
pol = {}
for t in range(T, 0, -1):
pv, pd = {}, {}
for s in sn:
s1 = min(s + 1, SMAX)
for c in cn:
stay, do = nxt[(s, c)], False
best = stay
if c > -BMAX and t > T - m["pullw"]:
v = inc(plan, s + 1, 1 - m["disc"]) + nxt[(s1, c - 1)]
if v > stay + 1e-12:
best, do = v, True
pv[(s, c)] = (1 - m["acc"]) * stay + m["acc"] * best
pd[(s, c)] = do
cur = {}
for s in sn:
s1 = min(s + 1, SMAX)
for c in cn:
hold = False
if s < 0:
a = pv[(s1, c)]
else:
a = inc(plan, s + 1, 1.0) + pv[(s1, c)]
if c < hmax and pv[(s, c + 1)] > a + 1e-12:
a, hold = pv[(s, c + 1)], True
p = pv[(s, c)]
e = min(max((a - p) / m["k"], 0.0), P1 - P0)
cur[(s, c)] = p + (P0 + e) * (a - p) - m["k"] * e * e / 2
pol[(t, s, c)] = (P0 + e, hold, pd[(s, c)])
nxt = cur
return nxt, pol
def solve(plan, m):
cn = range(-BMAX, m["hmax"] + 1)
vst = {c: 0.0 for c in cn}
for _ in range(400):
w1, pol = backward(plan, vst, m)
new = {}
for c in cn:
if c <= 0:
new[c] = w1[(c, 0)]
continue
terms = []
for n in range(c + 1):
pr = (math.comb(c, n) * (1 - m["hloss"]) ** n
* m["hloss"] ** (c - n))
pay = math.fsum(inc(plan, j, 1.0) for j in range(1, n + 1))
terms.append(pr * (pay + w1[(n, 0)]))
new[c] = math.fsum(terms)
gap = max(abs(new[c] - vst[c]) for c in cn)
vst = new
if gap < 1e-7:
break
return pol
def simulate(plan, pol, m, nq, seed):
# one record per quarter: owner profit, pay, deals by week, etc.
rng = random.Random(20261112 + seed)
c, out = 0, []
for _ in range(nq):
prof, pay, rev = [], [], []
week, pulled = [0] * (T + 1), [0] * (T + 1)
holds = lost = 0
s = c
if c > 0:
s = 0
for _ in range(c):
if rng.random() < 1 - m["hloss"]:
s += 1
else:
lost += 1
for j in range(1, s + 1):
x = inc(plan, j, 1.0)
pay.append(x)
prof.append(GM - x)
rev.append(1.0)
week[1] += s
cn = 0
for t in range(1, T + 1):
p, hold, _ = pol[(t, s, cn)]
if rng.random() < p:
if s < 0:
s += 1
elif hold:
cn += 1
holds += 1
else:
x = inc(plan, s + 1, 1.0)
pay.append(x)
prof.append(GM - x)
rev.append(1.0)
s = min(s + 1, SMAX)
week[t] += 1
if rng.random() < m["acc"] and pol[(t, s, cn)][2]:
x = inc(plan, s + 1, 1 - m["disc"])
pay.append(x)
prof.append(GM - m["disc"] - x)
rev.append(1 - m["disc"])
s = min(s + 1, SMAX)
cn -= 1
week[t] += 1
pulled[t] += 1
c = cn
out.append(dict(profit=math.fsum(prof), pay=math.fsum(pay),
rev=math.fsum(rev), week=week, pulled=pulled,
holds=holds, lost=lost))
return out
def mean(xs):
return math.fsum(xs) / len(xs)
# ---- end of shared core ----
BG = 0.10
def window(recs, n, k, rng):
late = [0, 0]
early = [0, 0]
for _ in range(n):
i = rng.randrange(len(recs) - k)
for r in recs[i:i + k]:
for t in range(1, T + 1):
reg = r["week"][t] - r["pulled"][t]
bg = 0
for _ in range(reg):
bg += rng.random() < BG
box = late if t > T - 3 else early
box[0] += r["week"][t]
box[1] += bg + r["pulled"][t]
return late, early
def z_one_sided(late, early):
p1, p2 = late[1] / late[0], early[1] / early[0]
pp = (late[1] + early[1]) / (late[0] + early[0])
se = math.sqrt(pp * (1 - pp) * (1 / late[0] + 1 / early[0]))
return (p1 - p2) / se if se > 0 else 0.0
if __name__ == "__main__":
m = dict(BASE)
recs = {}
for name, plan in (("L", LIN), ("S", STEP), ("A", ACCEL)):
recs[name] = simulate(plan, solve(plan, m), m, 24000, 8)
rng = random.Random(20261112)
for name in "LSA":
late, early = window(recs[name], 400, 1, rng)
print("{}: discounted share weeks 11-13 {:.3f}, weeks 1-10 {:.3f}"
.format(name, late[1] / late[0], early[1] / early[0]))
print("people quarters fires L fires S fires A")
for n, k in ((4, 1), (8, 1), (4, 4), (8, 4)):
row = []
for name in "LSA":
hits = 0
for _ in range(400):
hits += z_one_sided(*window(recs[name], n, k, rng)) > 1.645
row.append(hits / 400)
print(" {:2d} {} {:.3f} {:.3f} {:.3f}"
.format(n, k, *row))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《销售提成阶梯非线性、期末冲量操纵与业绩配额博弈扭曲母典》第 6 章「期末冲量的检测:周度份额检验与折扣率监控」(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-213 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026213,
author = {Fan, Yafu},
title = {Nonlinear Commission Ladders, End-of-Period Timing Manipulation and Quota-Induced Game Distortion},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}