Forecast Accuracy

Every forecast is logged with a timestamp and a predicted price for a specific vehicle segment. Once 4 weeks have elapsed, we look up the actual market median price (from live dealer inventory) that was recorded after the forecast date.

Average Error % measures how far off the predicted price was from the actual median: |predicted − actual| / actual × 100. Lower is better. Used car markets are inherently noisy, so errors under 5% are considered strong performance.

Trend-Classification Accuracy measures whether the directional call matched actual price movement: a Rising classification is correct if the price actually rose; a Softening classification is correct if it fell; a Stable classification is counted as correct when prices stayed roughly flat.

We do not cherry-pick results. Every forecast in our database is included once it has a corresponding actual price to compare against.