关于「按节庆对齐的同比:事件时间映射、斋月切分与日期误差」,《马来西亚移动节庆与促销排期》给出的定量结论是什么?

结论:第六章给出按节庆对齐的月度同比,全年平均绝对误差由 0.054 降到 0.017。

数据口径:书中销售序列与提前天数为示意;全部计算由书中程序生成。 第二章量化同比失真:在一家示意门店中,开斋节从四月移入三月,使三月同日同比为加 0.289、四月为减 0.218,而真实增长只有百分之四;用真实节日日期,十二对比较的平均绝对误差为 0.180。

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: 0.017, 0.054

true growth +0.040; monthsame-date YoYaligned YoY
Jan+0.157+0.044
Feb-0.089+0.055
Mar+0.164+0.040
Apr-0.096+0.071
May+0.055+0.048
Jun+0.049+0.056
Jul+0.050+0.047
Aug+0.066+0.066
Sep+0.057+0.075
Oct+0.086+0.070
Nov+0.064+0.057
Dec+0.039+0.050
mean abs error: same-date 0.054, aligned 0.017
worst month:same-date 0.136, aligned 0.035
查看完整验算程序
# Chapter 6. p61: a festival-aligned calendar map for monthly YoY.
# Model (illustrative): a shop serving all three communities. Daily
# sales = 1000 x (1 + G) per year x weekday factor x (1 + sum of
# weight x (lift - 1)) x lognormal noise (sigma 0.08, fixed seed),
# with the lift profile of p21 around each festival and weights
# CNY 0.5, Aidilfitri 0.6, Deepavali 0.3. Dates: the official 2024
# and 2025 dates. Aligned map: a day within -28..+6 of a festival is
# paired with the same event day last year; any other day is paired
# with the day 364 days earlier (same weekday).
import math
import random
from datetime import date, timedelta

G, SIG, LO, HI = 0.04, 0.08, -28, 6
WEEK = [1.0, 0.95, 0.95, 1.0, 1.1, 1.25, 1.2]
FEST = {"CNY": (0.5, {2024: date(2024, 2, 10), 2025: date(2025, 1, 29)}),
        "RAYA": (0.6, {2024: date(2024, 4, 10), 2025: date(2025, 3, 31)}),
        "DEEP": (0.3, {2024: date(2024, 10, 31),
                       2025: date(2025, 10, 20)})}
MON = "Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec".split()


def lift(k):
    if -21 <= k <= -1:
        return 1.0 + 1.2 * (22 + k) / 21
    if 0 <= k <= 2:
        return 0.35
    if 3 <= k <= 6:
        return 0.8
    return 1.0


def sales(rng):
    s, d = {}, date(2023, 12, 1)
    while d <= date(2025, 12, 31):
        x = [1.0] + [w * (lift((d - t.get(d.year, t[2024])).days) - 1)
                     for w, t in FEST.values()]
        v = 1000.0 * (1 + G) ** (d.year - 2024) * WEEK[d.weekday()]
        s[d] = v * math.fsum(x) * math.exp(rng.gauss(0.0, SIG))
        d += timedelta(days=1)
    return s


def partner(d, fests):
    best = None
    for name, (w, t) in fests.items():
        k = (d - t[d.year]).days
        if LO <= k <= HI and (best is None or abs(k) < abs(best[1])):
            best = (name, k)
    if best is None:
        return d - timedelta(days=364)
    return fests[best[0]][1][d.year - 1] + timedelta(days=best[1])


if __name__ == "__main__":
    s = sales(random.Random(20261113))
    print("true growth {:+.3f}; month  same-date YoY  aligned YoY".format(G))
    en, ea = [], []
    for m in range(1, 13):
        days = [d for d in s if d.year == 2025 and d.month == m]
        cur = math.fsum(s[d] for d in days)
        old = math.fsum(s[d] for d in s if d.year == 2024 and d.month == m)
        ali = math.fsum(s[partner(d, FEST)] for d in days)
        en.append(abs(cur / old - 1 - G))
        ea.append(abs(cur / ali - 1 - G))
        print("                  {}   {:+.3f}        {:+.3f}".format(
            MON[m - 1], cur / old - 1, cur / ali - 1))
    print("mean abs error: same-date {:.3f}, aligned {:.3f}".format(
        math.fsum(en) / 12, math.fsum(ea) / 12))
    print("worst month:    same-date {:.3f}, aligned {:.3f}".format(
        max(en), max(ea)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《马来西亚多元节庆移动日历叠加、开斋节农历新年屠妖节浮动日期与促销排期对齐母典》第 6 章「按节庆对齐的同比:事件时间映射、斋月切分与日期误差」(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-313 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026313,
  author    = {Fan, Yafu},
  title     = {Moving Festival Calendars in Multicultural Malaysia: Overlaps of Hari Raya, Chinese New Year and Deepavali and the Alignment of Promotional Schedules},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论