Week 2 · Final research publication
Cambridge leads near term. Ithaca leads since 2020.
The same 80-month protocol produces a split result: Cambridge carries the higher forecast value and modest projected momentum, while Ithaca has delivered stronger cumulative appreciation.
Cambridge 02139Ithaca 14850N = 80 per city44 backtests per city
Olu (Tim) Daramola
ChrionML® AI Labs · 64-City Intelligence Program
Published: 2 October 2026Research article: Week 2Housing type: Condo/co-op
Abstract
We analyzed 80 monthly condo/co-op index observations for Cambridge 02139 and Ithaca 14850 under an identical forecasting protocol. Across 44 expanding-window backtests, Drift minimized error in Cambridge while the Last Value baseline performed best in Ithaca. Cambridge retained the higher forecast value and modest projected monthly momentum; Ithaca exhibited stronger cumulative appreciation since January 2020. The results demonstrate that a shared evaluation standard need not produce a universal model winner. Published 2 October 2026.
Research goal
To determine whether the Week 1 evaluation protocol remains useful across two structurally different markets and whether model selection changes when local out-of-sample evidence favors simplicity.
Methods summary
Five interpretable candidates—Last Value, 12-month Seasonal Naive, Drift, 36-month Linear Trend and Holt Damped Trend—were evaluated chronologically. The lowest-MAE candidate was selected independently for each city, with no look-ahead information available during historical forecasts.
Primary finding
The same protocol can—and should—select different models.
Drift performs best in Cambridge, while the Last Value baseline wins in Ithaca. Complexity is not rewarded unless it improves out-of-sample accuracy.
$889KCambridge forecast
$290KIthaca forecast
+17.8%Cambridge since 2020
+29.5%Ithaca since 2020
What the evidence shows
Four findings measured under one repeatable evaluation protocol.
Value
Cambridge’s forecast is roughly three times Ithaca’s
The September 2026 estimates are $888,570 and $289,770, respectively—a 3.07× market-level ratio.
Momentum
Cambridge holds the projected monthly edge
The selected models imply +0.19% for Cambridge and a flat estimate for Ithaca over the one-month horizon.
Trajectory
Ithaca leads cumulative appreciation
Ithaca rose approximately 29.5% since January 2020 versus 17.8% for Cambridge, despite its lower current price level.
Model selection
Local evidence defeats one-size-fits-all modeling
Drift reduces Cambridge’s baseline MAE by 6.8%. In Ithaca, Last Value outperforms Drift and the other candidates.
Data availability and provenance
Zillow publishes the underlying Home Value Index histories. ChrionML® located the condo/co-op view for each ZIP code, compiled the monthly values, checked the observation sequence, and applied an identical forecasting and evaluation protocol to both cities.
Limitations
The study uses a ZIP-level index and a one-month forecast horizon. It does not capture property-specific variation or directly model macroeconomic and policy drivers. A winning historical model can still produce future error.
Interpretation boundary. The forecast comparison describes modeled condo/co-op index levels and short-horizon movement. It is not a property appraisal or a universal declaration of which city is “better.”
References
- Zillow. Home Value Index for Cambridge, Massachusetts 02139, condo/co-op series.
- Zillow. Home Value Index for Ithaca, New York 14850, condo/co-op series.
- ChrionML® AI Labs. Research Methodology, version 1.0.
