Forecasts based on Bayesian shrinkage combination

Authors:
• Mihaela SIMIONESCU, PhD., email: mihaela_mb1@yahoo.com, Afiliation: Institute for Economic Forecasting of the Romanian Academy

Pages:
• 244|261
Keywords: shrinkage parameter, Bayesian forecasts combination, forecasts accuracy, prior
Abstract:
Abstract. The Bayesian shrinkage combination approach was employed to\r\n\r\n improve the inflation rate predictions in Romania. Forecasters’\r\n\r\n anticipations are used as prior information, the forecasts being provided by\r\n\r\n experts in forecasting. For inflation in Romania during 2012-2014 a fixed\r\n\r\n effects model performed better than other types of econometric models\r\n\r\n (dynamic model, log-linear model, VAR model, Bayesian VAR, simultaneous\r\n\r\n equations model). The Bayesian combinations that employed experts’\r\n\r\n forecasts as priors, when the shrinkage parameter tends to infinite,\r\n\r\n improved the accuracy of all forecasts based on individual models,\r\n\r\n outperforming the naïve predictions and null and equal weights combined\r\n\r\n predictions.\r\n

CCRP Conference