Decentralmate is a price estimator for single parcels that are currently on sale in Decentraland. The streamlit web application is here.
Many of these listed parcels are asking for exorbitant prices, but what price would they realistically sell at? To answer this question, I built a price estimator model using a random forest trained on features such as the x,y coordinates of the parcel, its proximity to roads, plazas and certain districts. In order to compare the relative value of two parcels in a given time period, I normalized the parcel sale price by the 5-day rolling average sale price of all parcels. The normalized price distribution shows the higher priced parcels near the center:
The resulting model also captures the characteristic that parcel values are higher near the center and near roads and plazas:
Using the data from the past year 2021 as the testing set and all prior data as the training set, the model yields a testing R^2 value of 0.65 with MAPE ~13%.
run generate_parcel_csv.py
This will download the data as csv files including road parcels, genesis plazas, Decentraland University, Decentraland Convention Center, District X, Gambling District, get transaction information on all previously sold parcels, and also download MANA-USD exchange rate from yahoo finance.
run generate_model.py
This will generate normalized parcel price predictions and save it to parcel_model.csv, which is a map of all predicted normalized prices of all parcels. Note this module needs parcel_model.csv and roads.csv to run.
run streamlit run app.py