关于「夏普比率:门槛回报、估计误差与偏态的陷阱」,《达人矩阵的组合方差》给出的定量结论是什么?

结论:第四章说明夏普比率测不准,顶流在八个季度里的估计平均只有 0.0783。

数据口径:书中借用组合理论的方差、相关、夏普比率与均值—方差前沿,只借其算术,讨论的是营销预算的分配,不是证券投资;本书不构成任何个人化的理财或投资建议,不推荐任何金融产品。 书中的达人、回报、相关、成本与预算均为示意;全部计算由书中程序生成;所有合作都须如实标注广告,不讨论购买流量或虚假评价。

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: 0.0783

head lognormal: skewness 3.556, kurtosis 31.771
V per quarter: portfolio 1.0833, head normal 1.0168, head lognormal 0.6066
T se(port) se(head)95% CI of portfolio Sharpe
40.5204 0.3894 -0.612 to 1.428
80.3680 0.2754 -0.313 to 1.129
120.3005 0.2248 -0.181 to 0.997
200.2327 0.1742 -0.048 to 0.864
400.1646 0.1231 0.086 to 0.731
simulated (10,000 runs): mean and sd of the estimated Sharpe
T8: portfolio 0.4631 sd 0.4451; head 0.0783 sd 0.4544
T40: portfolio 0.4144 sd 0.1685; head 0.1718 sd 0.1411
quarters per arm to detect the gap 0.2249 (5% two-sided, 80% power):
normal formula 326, with the head's skew and kurtosis 263
查看完整验算程序
# Chapter 4. p42: how precisely can a few quarters measure a Sharpe ratio?
# Model: for T independent quarters, the estimated Sharpe ratio SR has
# standard error sqrt(V / T), with V = 1 + SR^2 / 2 for normal returns
# (Lo 2002) and V = 1 + SR^2 / 2 - g3 * SR + (g4 - 3) / 4 * SR^2 when
# returns have skewness g3 and kurtosis g4 (the large-sample extension
# for non-normal returns). The head's lognormal return (sigma 0.7851,
# from p31) has large g3 and g4. A seeded simulation checks both at
# T = 8 and T = 40. Values illustrative (示意).
import math
import random
import statistics

SR_PORT, SR_HEAD = 0.4082, 0.1833
S2 = 0.7851 ** 2
w = math.exp(S2)
g3 = (w + 2.0) * math.sqrt(w - 1.0)
g4 = w ** 4 + 2.0 * w ** 3 + 3.0 * w ** 2 - 3.0
v_port = 1.0 + SR_PORT ** 2 / 2.0
v_head_n = 1.0 + SR_HEAD ** 2 / 2.0
v_head = v_head_n - g3 * SR_HEAD + (g4 - 3.0) / 4.0 * SR_HEAD ** 2
print("head lognormal: skewness %.3f, kurtosis %.3f" % (g3, g4))
print("V per quarter: portfolio %.4f, head normal %.4f, head lognormal %.4f"
      % (v_port, v_head_n, v_head))
print("   T  se(port)  se(head)  95% CI of portfolio Sharpe")
for t in (4, 8, 12, 20, 40):
    sp, sh = math.sqrt(v_port / t), math.sqrt(v_head / t)
    print("%4d %9.4f %9.4f   %.3f to %.3f"
          % (t, sp, sh, SR_PORT - 1.96 * sp, SR_PORT + 1.96 * sp))
rng = random.Random(20261113)
MH, HURDLE = -0.0458, 0.08
print("simulated (10,000 runs): mean and sd of the estimated Sharpe")
for t in (8, 40):
    ep, eh = [], []
    for _ in range(10000):
        xs = [rng.gauss(SR_PORT, 1.0) for _ in range(t)]
        ep.append(statistics.fmean(xs) / statistics.stdev(xs))
        ys = [math.exp(MH + 0.7851 * rng.gauss(0.0, 1.0)) - 1.0 - HURDLE
              for _ in range(t)]
        eh.append(statistics.fmean(ys) / statistics.stdev(ys))
    print("  T=%d: portfolio %.4f sd %.4f; head %.4f sd %.4f"
          % (t, statistics.fmean(ep), statistics.stdev(ep),
             statistics.fmean(eh), statistics.stdev(eh)))
nd = statistics.NormalDist()
z = nd.inv_cdf(0.975) + nd.inv_cdf(0.8)
gap = SR_PORT - SR_HEAD
print("quarters per arm to detect the gap %.4f (5%% two-sided, 80%% power):"
      % gap)
print("  normal formula %d, with the head's skew and kurtosis %d"
      % (math.ceil(z * z * (v_port + v_head_n) / gap ** 2),
         math.ceil(z * z * (v_port + v_head) / gap ** 2)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《达人矩阵头部集中度与中腰部组合套利、尾部方差夏普比率与百名垂直达人投资组合母典》第 4 章「夏普比率:门槛回报、估计误差与偏态的陷阱」(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-276 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026276,
  author    = {Fan, Yafu},
  title     = {Head Concentration versus Mid-Tier Portfolios in Creator Matrices: Tail Variance, Sharpe Ratios and the Allocation of a Budget across a Hundred Niche Creators},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论