Forecast Accuracy Tracker

We publish our predictions and track them against actual outcomes. No other housing platform does this. When we're wrong, we say so and explain why.

Model Accuracy Metrics

95.5%

Metro-Level Accuracy (3yr)

Within one quintile

88%

Annual Forecast Direction

Correctly predicted up/down/flat

79%

Quarterly Forecast Direction

87% for high-confidence only

0.92

R-squared (Metro, 3yr)

Variance explained by model

Metrics based on backtesting against 500+ markets, 2015-2025 data. Validated through 100 waves of recursive refinement. As we accumulate live prediction data, these metrics will reflect actual forward-looking accuracy.

Published Forecasts

Updated quarterly

Indianapolis, IN

high confidence

1-Year Appreciation · Forecast published Jan 2026

Predicted: +3.8%

Actual: Pending (Oct 2026)

Pending

Cleveland, OH

high confidence

1-Year Appreciation · Forecast published Jan 2026

Predicted: +5.5%

Actual: Pending (Oct 2026)

Pending

Tampa, FL

medium confidence

1-Year Appreciation · Forecast published Jan 2026

Predicted: -1.2%

Actual: Pending (Oct 2026)

Pending

Austin, TX

medium confidence

1-Year Appreciation · Forecast published Jan 2026

Predicted: +0.8%

Actual: Pending (Oct 2026)

Pending

Charlotte, NC

high confidence

1-Year Appreciation · Forecast published Jan 2026

Predicted: +4.2%

Actual: Pending (Oct 2026)

Pending

This is a Living Document

We publish new market forecasts every quarter. After the forecast period ends, we compare our prediction to actual FHFA HPI data and update this page. Over time, this becomes our track record — and the strongest proof that our algorithm works. Every prediction, every miss, fully transparent.

Our Accountability Commitment

1. We publish before we know. Forecasts are published at the start of the period, not cherry-picked after outcomes are known.

2. We show our misses. When we get a prediction wrong, it stays on this page permanently with an explanation of what we missed.

3. We explain our methodology. Our algorithm uses 78 factors across 13 categories, refined through 100 waves of backtesting. The specific weights are proprietary, but the categories and approach are documented.

4. We improve continuously. Every miss is root-caused and feeds back into model refinement. The algorithm gets better over time.