Ekonomi TEK
Yazarlar: Bige KÜÇÜKEFE
Konular:İktisat
Anahtar Kelimeler:Machine learning,Forecast evaluation,Inflation forecasting,Surveys of expectations,Summary statistics
Özet: This paper aims to produce more accurate short-term inflation forecasts based on surveys of expectations by employing machine-learning algorithms. By treating inflation forecasting as an estimation problem consisting of a label (inflation) and features (summary statistics of surveys of expectations data), we train a suite of machine-learning models, namely, Linear Regression, Bayesian Ridge Regression, Kernel Ridge Regression, Random Forests Regression, and Support Vector Machines, to forecast the consumer-price inflation (CPI) in Turkey. We employ the Time Series Cross Validation Procedure to ensure that the training data exclude forecast horizon data. Our results indicate that these machine-learning algorithms outperform the official forecasts of the Central Bank of Turkey (CBT) and a univariate model.