spb/wp11_uqo

1% Author: Simon-Pierre Boucher — contact@spboucher.ai2% ============================================================================3% UQO Working Paper No. 114% Half a Million Prices, Twenty Models5%6% Build:  latexmk -pdf main.tex   (or `make` in this directory)7% Figures are read from ../figures/, tables from ../results/tables/.8% ============================================================================9\documentclass[12pt,letterpaper]{article}1011\usepackage[utf8]{inputenc}12\usepackage[T1]{fontenc}13\usepackage[english]{babel}1415\usepackage[letterpaper,margin=1in]{geometry}16\usepackage{setspace}17\onehalfspacing1819\usepackage{mathptmx}20\usepackage{microtype}2122\usepackage{amsmath,amssymb,amsthm}2324\usepackage{booktabs}25\usepackage{threeparttable}26\usepackage{makecell}2728\usepackage{graphicx}29\usepackage{subcaption}30\graphicspath{{../figures/}{./}}3132\usepackage[font=small,labelfont=bf,labelsep=period,justification=justified,singlelinecheck=false]{caption}3334\usepackage[dvipsnames]{xcolor}35\usepackage[colorlinks=true,linkcolor=NavyBlue,citecolor=NavyBlue,urlcolor=NavyBlue,breaklinks=true]{hyperref}3637\usepackage[authoryear,round,semicolon]{natbib}38\bibliographystyle{aer}3940\usepackage{titlesec}41\titleformat{\section}{\large\bfseries}{\thesection.}{0.5em}{}42\titleformat{\subsection}{\normalsize\bfseries}{\thesubsection.}{0.5em}{}43\titleformat{\subsubsection}{\normalsize\itshape}{\thesubsubsection.}{0.5em}{}44\usepackage{fancyhdr}45\pagestyle{fancy}\fancyhf{}46\renewcommand{\headrulewidth}{0pt}47\fancyfoot[C]{\thepage}48\usepackage{enumitem}49\setlist{nosep,leftmargin=*}5051% Float placement: let text and floats share pages52\setcounter{topnumber}{3}\setcounter{bottomnumber}{2}\setcounter{totalnumber}{4}53\renewcommand{\topfraction}{0.9}\renewcommand{\bottomfraction}{0.6}54\renewcommand{\textfraction}{0.08}\renewcommand{\floatpagefraction}{0.5}5556% ============================================================================57% METADATA58% ============================================================================59\newcommand{\WPnumber}{11}60\newcommand{\WPtitle}{Half a Million Prices, Twenty Models}61\newcommand{\WPsubtitle}{A Systematic Assessment of Hedonic Specifications and62  Estimation Methods for the Quebec Housing Market, 2021--2026}63\newcommand{\WPdate}{October 2026}64\newcommand{\WPversion}{1.1}65\newcommand{\WPkeywords}{hedonic price function, functional form, spatial fixed effects, machine learning, automated valuation model, forward-in-time holdout, house price index}66\newcommand{\WPjel}{R31, C21, C52, C53}6768\newcommand{\WPauthor}{Simon-Pierre Boucher}69\newcommand{\WPaffiliation}{D\'epartement des sciences administratives\\70  Universit\'e du Qu\'ebec en Outaouais}71\newcommand{\WPemail}{simon-pierre.boucher@uqo.ca}72\newcommand{\WPaddress}{Gatineau -- Pavillon Alexandre-Tach\'e\\73  283, boulevard Alexandre-Tach\'e\\ Gatineau, Qu\'ebec, Canada J9A 1L8}7475\newcommand{\WPabstract}{%76Fifty years after \mbox{Rosen (1974)}, applied hedonic practice still77rests on specification choices---functional form, temporal controls,78spatial controls, estimation method---that are made by convention and79rarely stress-tested jointly. Using 514{,}212 Quebec residential sales80(2021--2026) whose structural descriptors come from the provincial81assessment roll, we run a controlled horse race of twenty hedonic models82organised along those four axes, scoring every model on identical83holdouts and on the same price-level scoreboard with a single84retransformation rule, with a municipality-clustered bootstrap interval85and a Diebold--Mariano-type test on every comparison. Four lessons86emerge. (1)~Functional form is second-order: from linear to splines the87median absolute error spans 17.6--19.2\%, and the data-driven Box--Cox88exponent gains half a point. (2)~Controls are first-order: dropping89spatial fixed effects costs ten error points; a $\sim$1~km grid used90alone overfits, but nested within municipalities it becomes the most91accurate linear specification---the bias--variance trade-off in92absorbing location is resolved by hierarchy, not coarseness. (3)~Under93random validation the estimation method dominates: gradient boosting94reaches 14.4\%, an 18\% improvement over the best linear model95(95\% interval 1.9--4.7 points) that survives holdouts grouped by parcel96and by building and comes almost entirely from a flexible surface over97space. (4)~Under a forward-in-time holdout (train $<$~2025, test982025--26) every model degrades and the gap shrinks to half a point and99to insignificance. Decomposing the error shows that relative pricing100skill transfers forward intact (re-levelled, boosting leads by 2.9101points) while every frozen model under-prices the typical future sale102by 10--20\%, the ensembles slightly more. Random cross-validation hides103this staleness cost; it does not, as a previous version of this paper104claimed on the basis of an asymmetric retransformation, reverse the105ranking. Constant-quality indices agree across functional forms to106within 2.5 points but a boosted imputation index rises nine points less,107and the boosted model's implicit age and area profiles share the shape108of the quadratic OLS gradients with a milder depreciation path.}109110\hypersetup{pdftitle={Half a Million Prices, Twenty Models}, pdfauthor={}}111112% Self-reference neutralised in the anonymous version (main_anon.tex)113\newcommand{\companionpaper}{a companion paper by the same authors (reference withheld for review)}114115\begin{document}116117\input{sections/titlepage_anon}118119\setcounter{page}{1}120\input{sections/introduction}121\input{sections/literature}122\input{sections/data}123\input{sections/methodology}124\input{sections/results}125\input{sections/extensions}126\input{sections/robustness}127\input{sections/discussion}128\input{sections/conclusion}129130\clearpage131\bibliography{references}132133\end{document}134