Tokens in the digital landscape can vary in their rarity and value. This document outlines the definition of rare tokens based on their price premiums and delves into offline evaluation methods, specifically utilizing the Mean Absolute Percentage Error (MAPE) to gauge their accuracy.
Offline Evaluations of Pricing Model
Offline evaluations act as a powerful lens, revealing the nuances and performance of our pricing model without the complexities of real-time testing. This approach shines a light on the model's pricing accuracy, enabling a more comprehensive understanding of its strengths and areas of potential refinement.
We employ MAPE (Mean Absolute Percentage Error) as our primary evaluation metric. This statistical measure is tailored to discern the accuracy of forecasts, quantifying the average absolute percentage discrepancy between actual sale prices and our model's predictions. Significantly, a lower MAPE value is indicative of superior accuracy, signaling a model's proficiency in closely approximating real-world prices.
The formula for MAPE is given by:
Where:
n represents the number of transactions.
Sale Price is the real sale price of the token
Predicted Price stands for the price projected by our pricing model.
In our evaluation process, we utilize the Predicted Price derived from our models to calculate MAPE. Our objective remains clear: to identify and adopt models that minimize the MAPE, ensuring the highest forecasting accuracy for our token pricing endeavors.