ChrionML® Research Archive · October 20264 completed city studies · 2 matchups
Independent quantitative research

Research archive.

Four completed city-level machine-learning studies, organized into two comparable matchups. Each publication separates Zillow’s underlying ZHVI history from ChrionML®’s dataset construction, quality assurance, modeling, evaluation and interpretation.

4 cities published320 monthly observations analyzed176 rolling backtests64-city research program

Completed ML projects

Every city is evaluated independently before results are compared at the matchup level.

Project 01 · Boston 02114

High-value condo market with measured near-term growth

Eighty monthly condo/co-op observations evaluated across five candidate forecasting approaches.

$779KForecast
+0.06%Momentum
DriftWinner
View Boston evidence →
Project 02 · Providence 02903

Lower price level with stronger projected momentum

The same chronological protocol selects Drift and measures substantially stronger growth since 2020.

$438KForecast
+0.44%Momentum
DriftWinner
View Providence evidence →
Project 03 · Cambridge 02139

Highest forecast value with a local Drift advantage

Drift improves upon the last-value baseline under expanding-window evaluation.

$889KForecast
+0.19%Momentum
DriftWinner
View Cambridge evidence →
Project 04 · Ithaca 14850

Strongest Week 2 trajectory with a baseline win

The simplest candidate—Last Value—outperforms more complex alternatives in rolling backtests.

$290KForecast
+29.5%Since 2020
Last ValueWinner
View Ithaca evidence →

Published matchups

Research narratives designed for both technical and nontechnical readers.

Week 1 · Final

Boston vs Providence

Boston leads on forecast value. Providence leads on projected momentum and observed appreciation since January 2020.

Read the research publication →
Week 2 · Final

Cambridge vs Ithaca

Cambridge leads on value and near-term momentum. Ithaca leads on cumulative growth—and demonstrates why local model selection matters.

Read the research publication →
Research standard

Comparable evidence, without pretending every market behaves the same.

The protocol stays consistent while model selection remains local. That balance makes the archive useful as it scales from four cities toward sixty-four.

80Months per city
44Backtests per city
5Candidate models
1Shared protocol