Year of Graduation
Benchmark Models Development for Forecasting Short Macroeconomic Time Series
The identification of an appropriate time-series predictive model and an accurate estimation of its parameters can become an extraordinarily difficult issue when working with a small number of observations. Modern information criteria often produce ambiguous selection of model specification for short datasets, which are regularly used in applied macroeconomic research. The proposed work focuses on the consideration of basic predictive models (so-called benchmarks) for forecasting short macroeconomic time series. These models have a small number of configurable parameters, that may give an advantage in computational time and a quality of the forecasts over more complex methods. The current study proposes an improvement of the standard theta method, that has become a very popular predictive model for its performance in the M3-Competition (2000). To demonstrate the superior quality of the optimized theta method, a numerical experiment using more than 5,000 short macroeconomic time series has been carried out with respect to other forecasting models.