Forecasting Annual Inflation with Seasonal Monthly Data: Using Levels versus Logs of the Underlying Price Index
Title: Forecasting Annual Inflation with Seasonal Monthly Data: Using Levels versus Logs of the Underlying Price Index
Series/Report no.: EUI MWP; 2009/37
This paper investigates whether using natural logarithms (logs) of price indices for forecasting inflation rates is preferable to employing the original series. Univariate forecasts for annual inflation rates for a number of European countries and the USA based on monthly seasonal consumer price indices are considered. Stochastic seasonality and deterministic seasonality models are used. In many cases the forecasts based on the original variables result in substantially smaller root mean squared errors than models based on logs. In turn, if forecasts based on logs are superior, the gains are typically small. This outcome sheds doubt on the common practice in the academic literature to forecast inflation rates based on differences of logs.
Subject: C22; Autoregressive moving average process; forecast mean squared error; log transformation; seasonally integrated process; seasonal dummy variables
Type of Access: openAccess