Abstract
Monetary aggregates are at present thought of, at best, as a variable that is redundant with respect to interest rates and, at worst, a measure unrelated to aggregate economic activity. We examine recent monetary policy in Switzerland and investigate the performance of a Divisia monetary aggregate in a price level forecasting experiment. Specifically, we investigate whether Swiss Divisia aggregates could accurately forecast the Swiss price level out-of-sample using a Multi-recurrent Neural Network model (MRN).
Results indicate that models based on interest rates alone reflect only past movements in the price level. In contrast, models that include the aggregate quantity of money yield results that are forward looking. This implies that policy decisions made on the basis of interest rates alone could miss structural shifts in the underlying data generating process of prices. That an MRN including money and calibrated to data ending in 2006 is forward-looking is supported by accurate forecasts of the price level through the financial crisis.
Results indicate that models based on interest rates alone reflect only past movements in the price level. In contrast, models that include the aggregate quantity of money yield results that are forward looking. This implies that policy decisions made on the basis of interest rates alone could miss structural shifts in the underlying data generating process of prices. That an MRN including money and calibrated to data ending in 2006 is forward-looking is supported by accurate forecasts of the price level through the financial crisis.
| Original language | English |
|---|---|
| Publication status | Unpublished - 2017 |