LAND Valuation Algorithm
The MGH DAO has been striving since its inception to create tools that will help its community to navigate through the open metaverse in a seamless way. One of the main goals of the DAO is to establish a decentralized state by developing multiple projects and experiences within the most relevant virtual worlds. One of the biggest hurdles MGH faced when acquiring the LANDs needed to fullfill its vision was finding fairly valued parcels. The AI and algorithmic trading background of our collaborator, ITRM, allowed the DAO to develop a valuation model that prices LAND parcels in a highly accurate way.
Firstly, all the historic LAND sales from multiple metaverses are retrieved from Opensea's and Etherscan's API. The data is used to calculate the daily median of historic sales on any given day. This statistical measure is used as a benchmark for the model's accuracy.
After having gathered all the data, the prices are fed to a machine learning algorithm that is then trained using all the historic price movements for the asset in question. The training process helps predicting the daily median of LAND sales on t+1. The daily median is used as the reference for the valuation of all individual parcels.
The virtual real estate market has some similarities to the real estate market. One of them is that prices vary depending on the area where a parcel is located. For this reason, we implemented a neighboring weighting algorithm which significantly increased the accuracy of the model. The algorithm analyzes the historic prices for the neighbors around the LAND being evaluated and gives them a weight depending on how many times the neighboring LANDs have been sold in the past. The more a LAND has been sold, the higher its weight and the more influence it has on the LAND being evaluated.
The weighting process is optimized by training neural networks with different combinations of neighbors. The AI was trained to test combinations between 1 and 34. After multiple training rounds, it was identified that the optimal combination of neighbors that minimized the median average percentage error (MAPE) was 11.
The model is currently set up to evaluate LANDs in The Sandbox, Decentraland and Axie Infinity and we are constantly looking to expand to other metaverses.
For more information please refer to the model's KPIs
MGH Valuation Algorithm KPIs.pdf
3MB
PDF