What does Sizing Systems and Return Minimisation conclude, in numbers, about “Vanity-Size Drift: Tracking a Moving Brand Offset with a Kalman Filter”?
Answer: Chapter 5 tracks vanity-size drift with a Kalman filter and shows that under drift a 12-season average, with error 0.6015, does worse than a 4-season average at 0.4640.
Scope of the numbers: Its brands, shops, prices and body dimensions are illustrative; all calculations are generated by its programs; the book is not individual business advice, deals only with the measurement of garment sizes, and makes no health judgement. Marked boxes separate published evidence, runnable Python calculations (21 programs), classical sources and the author's hypotheses.
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.4640, 0.6015, 12
drift sd gain K RMSE kalman last only mean of 4 mean of 12 0.0 0.000 0.0000 0.7390 0.3695 0.2133 0.3 0.332 0.4257 0.7390 0.4640 0.6015 0.6 0.547 0.5464 0.7390 0.6720 1.1448 1.0 0.718 0.6263 0.7390 1.0058 1.8866 (drift 0: the Kalman error tends to 0 as seasons accumulate) steady trend d = 0.25 cm per season: drift sd 0.3: gain 0.332, lag bias 0.5034 cm drift sd 0.6: gain 0.547, lag bias 0.2073 cm mean of 12 seasons lags by 1.3750 cm
Show the full program
# Chapter 5. p51: Tracking a drifting brand offset. Local-level model:
# u_t = u_{t-1} + w_t, w_t ~ N(0, q) (drift per season)
# y_t = u_t + v_t, v_t ~ N(0, r) (r = 1.28^2 / 3: three garments
# measured each season, as p41)
# Steady-state Kalman: P = (q + sqrt(q^2 + 4 q r)) / 2 (prior variance),
# gain K = P / (P + r), error variance P r / (P + r). Compared with the
# last season alone (r) and with plain means over the last W seasons,
# whose error variance is r / W + q / W^2 * sum_{i<W} (W - i)^2.
# A steady vanity trend of d cm per season leaves an exponentially
# weighted estimate behind by d (1 - K) / K. Illustrative values.
import math
R = 1.28 ** 2 / 3
def kalman(q, r=R):
if q == 0:
return 0.0, 0.0
p = (q + math.sqrt(q * q + 4 * q * r)) / 2
return p / (p + r), p * r / (p + r)
def window(q, w, r=R):
return r / w + q / w ** 2 * math.fsum((w - i) ** 2 for i in range(1, w))
print("drift sd gain K RMSE kalman last only mean of 4 mean of 12")
for sd in (0.0, 0.3, 0.6, 1.0):
q = sd * sd
k, v = kalman(q)
print("%.1f %.3f %.4f %.4f %.4f %.4f"
% (sd, k, math.sqrt(v), math.sqrt(R), math.sqrt(window(q, 4)),
math.sqrt(window(q, 12))))
print("(drift 0: the Kalman error tends to 0 as seasons accumulate)")
d = 0.25
print("steady trend d = %.2f cm per season:" % d)
for sd in (0.3, 0.6):
k, _ = kalman(sd * sd)
print(" drift sd %.1f: gain %.3f, lag bias %.4f cm" % (sd, k,
d * (1 - k) / k))
print(" mean of 12 seasons lags by %.4f cm" % (d * 5.5))Single source and citation
Fan, Y. (2026). Anthropometric Principal-Component Sizing, Vanity-Size Drift and Cross-Border Return-Rate Minimisation (Version 1.0.0), Chapter 5: Vanity-Size Drift: Tracking a Moving Brand Offset with a Kalman Filter. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-215 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026215,
author = {Fan, Yafu},
title = {Anthropometric Principal-Component Sizing, Vanity-Size Drift and Cross-Border Return-Rate Minimisation},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · A Label Is Not a Measurement: Brand Dispersion of the Same "M" as a Random Effect
- Chapter 2 · A Finer Size Chart or a Brand Residual: the Error Budget of Five Sizing Policies
- Chapter 3 · A Two-Dimensional Principal-Component Grid of Height and Girth: What a One-Axis Label Throws Away
- Chapter 4 · Shrinkage Estimates of Brand Offsets: How Far to Trust Two Measured Garments
- Chapter 6 · The Expected Cost of Returns: Sizing Up, Cross-Border Freight and Ordering Several Sizes
- Chapter 7 · Sources East and West: the Palace Tailors of the Zhou li and Quetelet's Average Man
- Chapter 8 · Closing the Loop: a Buyer's Direct-Message Size Recommender, a Return-Reason Code Table and a Pre-Registered Test