Current models for predicting personal consumption expenditures (PCE) employ statistical techniques and rely upon traditional economic features. We compare vector autoregression and random forest regression models using traditional economic features as inputs to predict PCE. Additionally, we develop novel features derived from the earnings call transcripts of publicly traded U.S. companies using natural language processing (NLP) techniques. These new features reduce the mean square error (MSE) of the vector autoregression model by 7% and the random forest model by 23%. We find the random forest models outperformed the vector autoregression models, with a MSE reduction of 68%. We conclude the new features improve PCE predictions.
Kinskey, Ian; Oswald, Glenn; McCann, Charles; Finch, Travis; and Tanaydin, Anthony
"Improvements to Consumption Prediction: Machine Learning Methods and Novel Features,"
SMU Data Science Review: Vol. 1:
4, Article 3.
Available at: https://scholar.smu.edu/datasciencereview/vol1/iss4/3
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License