Forecasting Residential Property Appreciation in NYC Using Machine Learning

This project addresses the challenge of predicting next-year residential property appreciation in New York City using census tract-year observations. Housing appreciation is shaped by local property-market conditions, neighborhood socioeconomic and housing structure, and broader macro-financial conditions. Forecasting short-term price changes, therefore, requires a framework that connects neighborhood-level indicators with wider market conditions.

The project combines building and neighborhood characteristics with macro-financial variables to predict the following year’s percentage change in residential sale price per square foot. The goal is to evaluate whether these variables contain meaningful predictive signals for next-year residential appreciation in New York City.

The project uses census tract-year as the main analytical unit because it allows housing market outcomes to be connected with neighborhood-level indicators. For each census tract and year, residential sales are aggregated into a median sale price per square foot. The target variable is then defined as the following year’s percentage change in tract-level median sale price per square foot.

The feature set combines three types of information. First, ACS-derived variables describe neighborhood socioeconomic data. Second, transaction-derived variables show property attributes. Third, macroeconomic variables capture broader market conditions that affect housing appreciation across New York City.

Using these data, the model can estimate how neighborhood conditions and overall market forces jointly relate to next-year residential appreciation. The purpose is not only to fit a prediction model, but also to test whether the selected variables contain enough signal to explain short-term changes in residential sale price per square foot.