关于「预算约束下的剪枝:背包、贪心比率与十年政策回测」,《域名池续费剪枝》给出的定量结论是什么?

结论:第四章把剪枝写成背包,并做十年回测:剪枝加等待的平均现值为 11,121,全部续费加净现值法则只有 1,344,但等待也加大了离散。

数据口径:示意参数下,触发倍数为 2.893,续费 15 时下边界为 599、上边界为 16,849;42 枚示意名字里有 23 枚应当弃续,期权部分占最优价值的 0.861,合理续费只需全额 1,340 中的一小部分。 核对原文之后,本书做了蓝图修正:2012 年的精确匹配域名更新针对的是低质量网站,名字不带排名加成只是稳健性假定;林登塔尔的重复交易指数只反映成交过的名字;不依赖排名的访问在示意参数下约占 0.309,是可靠的地板而不是流量的大头。

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: 11121, 1344

A. 42 names, 2,000 paths, ten years, PV at r = 0.08
policymean PVs.e.5th pctfees PVbuilds
prune+wait11,121185-371,6750.89
keep+wait4,114188-7,3279,4940.91
keep+NPV1,34450-1,4469,1503.65
prune+NPV8,510465,9761,3543.39
B. prune+wait minus keep+NPV: 9,777 per pool
查看完整验算程序
# Chapter 4. p42: ten years of four renewal policies on the same paths.
# Model: the 42 names of p13; V follows geometric Brownian motion with
# drift r - y (y = 0.06) and volatility 0.35, one step a year, 2,000
# seeded scenarios, the same draws for every policy. Fees are paid at
# the start of each held year, builds pay V - K, all discounted at r.
# After ten years a still-held name is marked at its model value W(V).
# Policies: prune + wait (drop at L, build at H), keep all + wait,
# keep all + build at V >= K (NPV rule), prune + NPV rule.
import math
import random


def roots(r, d, s):
    a = 0.5 - (r - d) / (s * s)
    b = math.sqrt(a * a + 2 * r / (s * s))
    return a + b, a - b


def solve(c, k, r=0.08, d=0.06, s=0.35):
    b1, b2 = roots(r, d, s)
    x = -b2 * (c / r) / (b1 - b2)
    y = b1 * (c / r) / (b1 - b2)

    def rhs(z):
        return (b1 - 1) * x * z ** b1 + (b2 - 1) * y * z ** b2 + c / r

    lo, hi = 1.0, 64.0
    for _ in range(200):
        mid = (lo + hi) / 2
        lo, hi = (mid, hi) if rhs(mid) < k else (lo, mid)
    z = (lo + hi) / 2
    low = (b1 * x * z ** b1 + b2 * y * z ** b2) / z
    return low, z * low, x, y, b1, b2, c / r


def w_of(v, k, sol):
    low, high, x, y, b1, b2, pv = sol
    if v <= low:
        return 0.0
    if v >= high:
        return v - k
    return x * (v / low) ** b1 + y * (v / low) ** b2 - pv


CATS = [("geo-svc", 10, 1500), ("geo", 8, 900), ("brand", 12, 250),
        ("generic", 12, 500)]
FEES = (12, 15, 35, 60)


def make_pool():
    rng = random.Random(20261120)
    pool = []
    for cat, n, vmed in CATS:
        for _ in range(n):
            v = vmed * math.exp(rng.gauss(0, 1.1))
            rng.gauss(0, 0.6)
            c = FEES[rng.randrange(4)]
            k = rng.choice((4000, 6000, 8000))
            pool.append((v, c, k))
    return pool


def run(paths=2000, years=10, r=0.08, y=0.06, s=0.35):
    pool = make_pool()
    sols = [solve(c, k) for _, c, k in pool]
    rng = random.Random(20261121)
    pols = ("prune+wait", "keep+wait", "keep+NPV", "prune+NPV")
    res = {p: ([], [], []) for p in pols}
    drift = r - y - s * s / 2
    for _ in range(paths):
        zs = [[rng.gauss(0, 1) for _ in range(years)] for _ in pool]
        for p in pols:
            pv, fees, builds = [], [], 0
            for (v0, c, k), sol, z in zip(pool, sols, zs):
                v, low, high = v0, sol[0], sol[1]
                trig = high if p.endswith("wait") else k
                for t in range(years + 1):
                    disc = math.exp(-r * t)
                    if t == years:
                        pv.append(disc * w_of(v, k, sol))
                        break
                    if v >= trig:
                        pv.append(disc * (v - k))
                        builds += 1
                        break
                    if p.startswith("prune") and v <= low:
                        break
                    fees.append(disc * c)
                    pv.append(-disc * c)
                    v *= math.exp(drift + s * z[t])
            res[p][0].append(math.fsum(pv))
            res[p][1].append(math.fsum(fees))
            res[p][2].append(builds)
    return pols, res


if __name__ == "__main__":
    pols, res = run()
    print("A. 42 names, 2,000 paths, ten years, PV at r = 0.08")
    print("   policy        mean PV   s.e.   5th pct   fees PV  builds")
    for p in pols:
        pv, fees, b = res[p]
        n = len(pv)
        m = math.fsum(pv) / n
        sd = math.sqrt(math.fsum((x - m) ** 2 for x in pv) / (n - 1))
        p5 = sorted(pv)[n // 20]
        print("   {:<11} {:>8,.0f}  {:>5.0f}  {:>8,.0f}  {:>7,.0f}  {:.2f}"
              .format(p, m, sd / math.sqrt(n), p5,
                      math.fsum(fees) / n, math.fsum(b) / n))
    base = math.fsum(res["keep+NPV"][0]) / len(res["keep+NPV"][0])
    best = math.fsum(res["prune+wait"][0]) / len(res["prune+wait"][0])
    print("B. prune+wait minus keep+NPV: {:,.0f} per pool".format(
        best - base))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《精确地域域名实物期权估值、续费剪枝最优停时与自然搜索垄断护城河母典》第 4 章「预算约束下的剪枝:背包、贪心比率与十年政策回测」(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-341 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026341,
  author    = {Fan, Yafu},
  title     = {Real-Options Valuation of Geo-Exact Domain Portfolios, Optimal Renewal Pruning and the Moats of Organic Search},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论