ABSTRACT
Autoregressive integrated moving average (ARIMA) models, or ARIMA forecasting, are sometimes referred to as the Box-Jenkins approach after the authors of the seminal and authoritative work on the topic. ARIMA models are a generalization of exponential smoothing models. ARIMA models attempt to exploit patterns and trends in the past values of a data series to make forecasts of future values. This approach will be fruitless if there are no patterns and trends to exploit. Recent developments in trend analysis with economic data suggest that there may be patterns, but these patterns are continually changing in an unpredictable way. This chapter considers these developments and their implications for time-series forecasting. It also considers integration when a time-series is differenced one or more times to make it stationary. The chapter describes the correlogram chart showing the correlations between observations in a time-series any given number of spaces apart.
