TURNING POINTS IN ECONOMIC TIME SERIES, LOSS STRUCTURES AND BAYESIAN FORECASTING

Methods for forecasting turning points and future values of economic time series are developed which take account of a forecaster's loss structure. For example, it is found that the decision to forecast a downturn in an economic series is very sensitive to the form of the forecaster's loss structure as well as to the predictive probability of a downturn. Using an autoregressive-leading indicator model and data on real output growth rates for eighteen countries, turning point forecasts were made for each year, 1974-84. Overall, 66% of the 68 downturn and no-downturn forecasts were correct and 75% of the 82 upturn and no-upturn forecasts were correct