关于「分段幂律与断点扫描:价格表里的台阶怎样被找出来」,《钻石魔术尺寸与培育钻崩价》给出的定量结论是什么?

结论:第二章说明忽略台阶的平滑拟合会把弹性高估 0.227,盲扫台阶须把门槛提高到 3.47。

数据口径:书中的价格、比例与成本参数,除注明出处者外均为示意;全部计算由书中程序生成;本书不是个人化的法律、理财或经营建议,涉及的标注与广告规则以上架前人工复核的现行法条为准。 第三章用切磨者模型说明台阶怎样在重量分布上留下缺口与堆积,示意模型中 1.00 克拉一档有 0.650 的石头是被推上关口的;并比较左位数注意力与焦点从众两种解释。

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: 0.227

(a) single power law: e = 2.167 (se 0.007), resid sd 0.127
(b) power law + steps: e = 1.914 (se 0.037), resid sd 0.123
step attrueestimatese95% interval
0.500.150+0.1690.023+0.124 to +0.214
0.700.030+0.0420.018+0.007 to +0.077
0.900.040+0.0250.020-0.014 to +0.064
1.000.070+0.0970.019+0.059 to +0.135
1.500.060+0.0610.016+0.029 to +0.092
2.000.080+0.1120.015+0.082 to +0.142
bias of the smooth fit in e: +0.227
查看完整验算程序
# Chapter 2. p21: Piecewise power law: fit steps at known focal weights.
# Synthetic listing (fixed seed): 1200 stones, carat uniform on 0.30 to
# 2.20 and rounded to 0.01; log price = ln 5000 + e * ln w + steps + noise.
# True e = 1.94 (Vaillant-Wolff, 112,080 stones); steps in log points
# at 0.50/0.70/0.90/1.00/1.50/2.00 and noise sd 0.12 are illustrative.
# Two OLS fits: (a) a single power law; (b) power law plus step dummies.
# Shows how a smooth fit misreads the elasticity when steps exist.
import math
import random

rng = random.Random(20261112)
STEPS = ((0.50, 0.15), (0.70, 0.03), (0.90, 0.04),
         (1.00, 0.07), (1.50, 0.06), (2.00, 0.08))
E_TRUE, SD, N = 1.94, 0.12, 1200


def draw():
    w = round(rng.uniform(0.30, 2.20), 2)
    y = math.log(5000.0) + E_TRUE * math.log(w)
    y += math.fsum(d for c, d in STEPS if w >= c - 1e-9)
    return w, y + rng.gauss(0.0, SD)


def ols(rows, ys):
    k = len(rows[0])
    xtx = [[math.fsum(r[i] * r[j] for r in rows) for j in range(k)]
           for i in range(k)]
    xty = [math.fsum(r[i] * y for r, y in zip(rows, ys)) for i in range(k)]
    a = [xtx[i][:] + [1.0 if i == j else 0.0 for j in range(k)]
         for i in range(k)]
    for c in range(k):
        p = max(range(c, k), key=lambda r: abs(a[r][c]))
        a[c], a[p] = a[p], a[c]
        piv = a[c][c]
        a[c] = [v / piv for v in a[c]]
        for r in range(k):
            if r != c:
                f = a[r][c]
                a[r] = [vr - f * vc for vr, vc in zip(a[r], a[c])]
    inv = [row[k:] for row in a]
    b = [math.fsum(inv[i][j] * xty[j] for j in range(k)) for i in range(k)]
    res = [y - math.fsum(bi * xi for bi, xi in zip(b, r))
           for r, y in zip(rows, ys)]
    s2 = math.fsum(e * e for e in res) / (len(ys) - k)
    se = [math.sqrt(s2 * inv[i][i]) for i in range(k)]
    return b, se, math.sqrt(s2)


data = [draw() for _ in range(N)]
ys = [y for _, y in data]
smooth = [[1.0, math.log(w)] for w, _ in data]
b0, se0, s0 = ols(smooth, ys)
print("(a) single power law: e = {:.3f} (se {:.3f}), resid sd {:.3f}"
      .format(b0[1], se0[1], s0))
full = [[1.0, math.log(w)] + [1.0 if w >= c - 1e-9 else 0.0
                              for c, _ in STEPS] for w, _ in data]
b1, se1, s1 = ols(full, ys)
print("(b) power law + steps: e = {:.3f} (se {:.3f}), resid sd {:.3f}"
      .format(b1[1], se1[1], s1))
print("step at   true   estimate   se     95% interval")
for i, (c, d) in enumerate(STEPS):
    est, s = b1[2 + i], se1[2 + i]
    print(" {:.2f}    {:.3f}   {:+.3f}    {:.3f}   {:+.3f} to {:+.3f}"
          .format(c, d, est, s, est - 1.96 * s, est + 1.96 * s))
gap = b0[1] - E_TRUE
print("bias of the smooth fit in e: {:+.3f}".format(gap))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《钻石克拉魔术尺寸价格跳跃、培育钻学习曲线崩价与供给卡特尔信号母典》第 2 章「分段幂律与断点扫描:价格表里的台阶怎样被找出来」(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-209 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026209,
  author    = {Fan, Yafu},
  title     = {Diamond Magic-Size Price Jumps, the Lab-Grown Learning-Curve Slide and Supply-Cartel Signalling},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论