Forecasting the Average Fixed Interest Rate of 30-year Mortgages: A Machine Learning Approach

Abstract

This paper analyzes the ability of 4 different multi and univariate forecasting models to predict directional change in the average fixed interest rate of 30-year mortgages on a monthly basis. Data was used from January 1978 till February 2022. Special attention is paid to the ability of these models to correctly predict when the interest rates increase instead of decrease. The 4 models used were ARMA, VAR, ARIMAX, and SVR. The SVR model was the worst at predicting directional change. The VAR model was the most accurate as it was able to correctly predict when interest rates would rise 83.89% of the time over the training data.