Best value ?
β
Fair price ?
β
Model fit (RΒ²) ?
β
Typical error (RMSE) ?
β
Price model
View
Items to buy
Price model terms
Listings against the model
What this assumes
- The listings are a fair sample of the market. The model only knows the prices you give it, so a few odd sales or a thin sample move the answer.
- Each feature has a steady effect. Linear adds a fixed amount per unit, Log-linear a fixed percentage, Quadratic lets the effect bend, and Ridge pulls each effect slightly toward zero.
- Older prices are restated in today's money at one flat inflation rate, from the listing's data year to the current year. A blank data year counts as the current year.
- A year becomes an age, measured at the listing's data year for the market data and at the current year for the items to buy.
- Yes/No features count as 1 or 0, and every feature is standardised before fitting, so a large unit such as kilometres cannot swamp a small one. This changes how the terms are scaled, not what Linear predicts.
- The range is one typical error either side, a guide to how far real prices scatter, not a confidence interval. Features outside the range of the listings are extrapolated and flagged.
- Items to buy are never added to the model. Their asking prices are judged against it, nothing more.