How do we calculate all the numbers you see in the Valuate module
The Valuate module uses searches in our Legentic open-source document repository based on the make, model, variant and years defined by the user. The outcome of the searches are a lot of numbers and graphs which are displayed. Within the page we try to outline the calculations behind the models for reference.
Results and Estimated price: The Market Value is obtained by a linear regression (i.e. fitting a line) and extrapolating its value to the current date. To gauge the confidence in our estimate, we construct a prediction band in the following way: First, we de-trend the prices — i.e., take the differences between the price and the regression line at the time when the ad was published — and compute the standard deviation of those de-trended prices. One way to think about this is that we look into how close to our model the actual prices were in the past.
Tracking Price Movements (1): The prediction band of the linear regression model is visualised by shifting the regression line up and down by the amount of the computed uncertainty. Dots in the graph are the actual advertisements found based on the search query.
Tracking Price Movements (2): We also visualize the uncertainty by plotting a kernel density estimation (KDE) of the prices adjusted for the time trend. In other words, we correct each historical price by the trend. For example, if a price in the ad was 100 EUR below the model at the time of its publication, we add a bump to our KDE plot at the price that is the current predicted price minus 100 EUR.