What does Debiasing Collectible Returns conclude, in numbers, about “Who Sells When: Resale Timing, Exit and Censoring”?
Answer: Chapter 3 separates resale timing, permanent exit and censoring in a twenty-year dynamic market: timing alone biases the index by about 0.0113 a year, permanent exit lifts this to 0.0263 or more, and correction needs a model of who sells.
Scope of the numbers: Its parameters are illustrative; all calculations are generated by its programs; it is not personalised legal or financial advice, and fee schedules and rules are subject to manual review before release. Marked boxes separate published evidence, runnable Python calculations (20 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: 0.0113, 0.0263
| 12 market paths of 20 years, 3000 pieces each | |||||
| A | B | Q | pairs | mean bias | sd |
| 0.00 | 0.00 | 0.00 | 1307 | -0.0002 | 0.0055 |
| 0.30 | 0.03 | 0.00 | 3759 | 0.0113 | 0.0035 |
| 0.60 | 0.03 | 0.00 | 5025 | 0.0087 | 0.0044 |
| 0.00 | 0.00 | 0.30 | 1136 | 0.0123 | 0.0044 |
| 0.30 | 0.03 | 0.30 | 3264 | 0.0263 | 0.0065 |
| 0.30 | 0.03 | 0.60 | 2990 | 0.0331 | 0.0079 |
| 0.60 | 0.03 | 0.60 | 3928 | 0.0334 | 0.0086 |
| bias = BMN annual return minus true market annual return |
Show the full program
# Chapter 3. p31: a 20-year market where owners decide when to resell
# Model (illustrative parameters, fixed seed): 3000 pieces enter at random
# years with a first sale. Each year a piece's log value moves with the
# market (drift 0.05, sd 0.10) plus its own shock (sd 0.20). Each year
# the owner resells with probability P0 + A*max(g,0) + B*max(-g,0), where
# g is the log gain since the last sale; a resale records a pair and the
# clock restarts. The Bailey-Muth-Nourse regression is then run on all
# pairs. A piece may also leave the public market for good (private sale,
# re-attribution, a lesser venue) with yearly probability Q*max(-g,0).
# Every row reuses the same 12 market paths and random streams.
import math
import random
import statistics
T = 20
NPIECES = 3000
MU_M, SD_M, SD_I, SD_N = 0.05, 0.10, 0.20, 0.05
P0 = 0.04
REPS = 12
SETTINGS = [(0.0, 0.0, 0.0), (0.30, 0.03, 0.0), (0.60, 0.03, 0.0),
(0.0, 0.0, 0.3), (0.30, 0.03, 0.3), (0.30, 0.03, 0.6),
(0.60, 0.03, 0.6)]
def solve(a, y):
n = len(y)
m = [row[:] + [y[i]] for i, row in enumerate(a)]
for c in range(n):
p = max(range(c, n), key=lambda r: abs(m[r][c]))
m[c], m[p] = m[p], m[c]
for r in range(n):
if r != c:
f = m[r][c] / m[c][c]
m[r] = [x - f * z for x, z in zip(m[r], m[c])]
return [m[i][n] / m[i][i] for i in range(n)]
def bmn(pairs):
cells = [[[] for _ in range(T)] for _ in range(T)]
rhs = [[] for _ in range(T)]
for s, t, y in pairs:
for i, si in ((s, -1.0), (t, 1.0)):
if i == 0:
continue
rhs[i - 1].append(si * y)
for j, sj in ((s, -1.0), (t, 1.0)):
if j:
cells[i - 1][j - 1].append(si * sj)
return solve([[math.fsum(c) for c in row] for row in cells],
[math.fsum(r) for r in rhs])
def simulate(path, a, b, q, rng):
pairs = []
for _ in range(NPIECES):
t0 = rng.randint(0, T - 2)
own = 0.0
last_t, last_v = t0, path[t0] + rng.gauss(0.0, SD_N)
for t in range(t0 + 1, T + 1):
own = math.fsum([own, rng.gauss(0.0, SD_I)])
true_v = path[t] + own
g = true_v - last_v
p = P0 + a * max(g, 0.0) + b * max(-g, 0.0)
u, x = rng.random(), rng.random()
if x < q * max(-g, 0.0):
break
if u < min(1.0, p):
v = true_v + rng.gauss(0.0, SD_N)
pairs.append((last_t, t, v - last_v))
last_t, last_v = t, v
return pairs
def main():
rng = random.Random(20261112)
out = {k: ([], []) for k in SETTINGS}
for rep in range(REPS):
path = [0.0]
for _ in range(T):
path.append(path[-1] + rng.gauss(MU_M, SD_M))
true_ann = math.expm1(path[T] / T)
for k in SETTINGS:
pairs = simulate(path, k[0], k[1], k[2],
random.Random(20261113 + rep))
est = math.expm1(bmn(pairs)[T - 1] / T)
out[k][0].append(est - true_ann)
out[k][1].append(len(pairs))
print("{} market paths of {} years, {} pieces each".format(
REPS, T, NPIECES))
print("{:>6}{:>6}{:>6}{:>8}{:>11}{:>9}".format(
"A", "B", "Q", "pairs", "mean bias", "sd"))
for k in SETTINGS:
bias, npairs = out[k]
print("{:>6.2f}{:>6.2f}{:>6.2f}{:>8}{:>11.4f}{:>9.4f}".format(
k[0], k[1], k[2],
int(math.fsum(npairs) / REPS), math.fsum(bias) / REPS,
statistics.pstdev(bias)))
print("bias = BMN annual return minus true market annual return")
if __name__ == "__main__":
main()Single source and citation
Fan, Y. (2026). Repeat-Sales Selection Bias in Collectibles, Forgery Priors and the Illiquidity Discount of Alternative Assets (Version 1.0.0), Chapter 3: Who Sells When: Resale Timing, Exit and Censoring. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-208 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026208,
author = {Fan, Yafu},
title = {Repeat-Sales Selection Bias in Collectibles, Forgery Priors and the Illiquidity Discount of Alternative Assets},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · After the Correction: Selection Bias in Repeat-Sales Samples
- Chapter 2 · The Bailey-Muth-Nourse Regression: from Pairs of Sales to a Price Index
- Chapter 4 · Forgery Priors: Likelihood Ratios of Authentication Evidence and the Value of Information
- Chapter 5 · The Round Trip: Break-Even Years and the Chance of Loss on a Single Piece
- Chapter 6 · The Illiquidity Discount: a Perfect-Timing Bound and Sales under a Deadline
- Chapter 7 · Classics East and West: Mi Fu's History of Painting and Vasari's Lives
- Chapter 8 · Closing the Loop: an Allocation Cap, a Due-Diligence Checklist and a Falsification Plan