Executive Summary

Disclaimer: This article has been written purely for educational and informational purposes as an explanation of how Shareland’s price feeds operate. This article is not intended to be investment advice of any kind.

The Shareland price oracle is the authority on the real-world price of real estate for each neighborhood on the Shareland exchange. This article provides the technical details of the Shareland neighborhood-level oracle price calculation. Using real estate transaction data, we identify the most common property type in each neighborhood. We then calculate the average price per square foot of all similar recently completed transactions. This approach provides an estimate of the price per square foot of a typical home in each neighborhood, which becomes the oracle price for that neighborhood. The oracle estimate is updated daily as real estate transactions close.

The method we use and discuss below offers a compromise between accuracy and ease of replicability. Our primary intention is to use basic, publicly available information and an easy to understand and replicate estimation method. Our secondary goal is to generate an oracle price per square foot estimate for a typical home in a consistent way through time. Our approach does not maximize pricing accuracy for all homes. In other words, the oracle methodology is designed to capture market evolution of a typical home over time. It is not an automated valuation system.

Source Data

We obtain real estate transaction data that contains all transactions as reported by the county recorder offices. As such, we rely on transactions that have closed escrow, and land title has been transferred to the buyer. We do not include transactions that are under contract, as not all transactions in this category close.

Table 1 lists the variables we use in our model. Our driving objective in selecting the variables is to ensure that all information is commonly available and accurately computed and recorded. In other words, we emphasize ease of use and replication.

Table 1: Variables used for home price estimation

‍

We recognize that many other physical and neighborhood characteristics influence the price of a home. Factors such as school quality and walkability are critical determinants of home prices. However, since our neighborhoods are small, these geographic factors affect the price of all properties within a neighborhood in a similar way. Therefore, there is no need to use these variables to select the sub-set of observations that are most alike the typical house.

This does not mean neighborhood characteristics are ignored, however. On the contrary, changes in neighborhood characteristics over time affect transaction prices, which, in turn, determine the oracle price.

We further use neighborhood boundary data which allows us to identify the transactions that fall within each neighborhood for which Shareland token is available.

If a particular neighborhood does not have enough transactions to estimate our model, we broaden the area to include neighboring communities.

Data Filters

Since our goal is to capture the evolution of the price of a typical home over time, rather than estimate the value of a wide range of homes, we employ the following filters to the data:

To find our typical home, we take all the homes in a specific neighborhood and apply the following filters:

• Filter out observations whose house size and lot size are below 10th percentile or above 90th percentile of the sample.
• Filter out observations whose number of bedrooms and number of bathrooms are below the 1st percentile or above the 80th percentile.

We then compute the median characteristics of the remaining homes in each neighborhood. A property with the median characteristics defines the typical home for each neighborhood.

To estimate the mean price per square foot, we take all completed sales from the last 90 days and select all observations whose size, lot size, number of bedrooms, and number of bathrooms are within 75% of the same characteristics for the typical home.

Oracle Price Estimation

The oracle price for a neighborhood is the weighted average price per square foot for the transactions over the past 90 days that satisfy the above filters. We employ the Epanechnikov Kernel weighting function, which puts more weight on recent sales and less weight on sales that completed further in the past.

If a neighborhood does not have a minimum of 50 observations that satisfy the above filters, then we include the nearest observations to the center of the neighborhood needed to reach the desired 50 observations. This includes all observations within the neighborhood and augments the sample with other nearby observations if needed.

The above model selection is driven by our primary objective to capture the evolution of the price per square foot for a typical home through time and ensure easy replication. The above approach is straightforward. However, it is not suitable for obtaining estimates for all homes within a neighborhood or for automated valuation systems.

Note that once we define the physical characteristics and location of a typical home for our purposes, we do not change those characteristics even if the composition of sales or homes changes over time. This ensures that changes in the oracle price reflect changes in the market prices in a neighborhood, rather than changes in the characteristics of the typical home.

We use the above method to provide daily updates to the oracle price for each neighborhood.

Model Validation

To validate our model, we manually consider real estate transactions that are similar to the typical home for each neighborhood. For instance, the typical home in Beverly Hills, as defined above, has the following characteristics:

Table 2: Variables used for home price estimation

‍

The estimated price per square foot for this typical home as of March 10, 2023 is $1168.

Table 3 provides a sample list of real estate transactions that occurred prior to that date and meet the filters described above. The average price per square foot of the 8 transactions reported in Table 3 is $1,225. As stated above, the weighted average of the full sample is $1168, which constitutes the oracle price per square foot on March 10, 2023.

Table 3: Sample of Beverly Hills Real Estate Transactions prior to March 10, 2023

‍

A user could draw on their experience or obtain past transactions to perform their own average price per square foot calculation. This would likely generate a numerically different average price per square foot estimate from what our method reports. Such a difference, which sometimes can be substantial, arises from the filters we employ, the potentially incomplete set of transactions a user obtains, or the time weighting we use. The guiding goal of our method is to capture the time variation of the price per square foot for a typical home, as defined by its size, number of rooms, and other physical characteristics. Our filters eliminate homes with extremely high or low values or with atypical characteristics. We note again that our method is not meant to estimate the price per square foot for all homes and is not suitable for automated valuation systems.

Conclusion

This article describes the methodology we use to derive the oracle price for each neighborhood. The data selection and modeling choices we've made are primarily driven by our objective to capture the time evolution of typical home prices, rather than estimate the prices of a wide range of homes. We were also careful to ensure that our data selection and model is easy to understand and replicate.