The robustness and accuracy of Box-Jenkins ARIMA in modeling and forecasting household debt in South Africa

Abstract

Abstract: This paper adopted the Box-Jenkins methodology to estimate a univariate time series model. Quarterly data collected from the South African Reserve Bank covering the period 1994 to 2014 was used. The initial plot of the series revealed that household debt is explained by an irregular and non-seasonal component. Owing to the non stationarity of the series, first differencing was applied to induce stationarity. The ACFs and PACFs identified six models. Of the six identified models,ð´ð‘…ð¼ð‘€ð´ 3, 1, 0 was selected according to the standard error estimates and the information criteria. The proposed model passed all the diagnostic tests and was further used for producing ten period forecasts of household debt. The forecasted household debt rates obtained were above 75% and within confidence bounds of 95%. Insample and out-of-sampling forecasts moved together confirming the reliability of the model in forecasting household debt and vigour in predictive ability. The proposed model exhibited the best performance in terms of Max APE and Max AE and ascertained the robustness and accuracy of the BoxJenkins ARIMA in forecasting. Both a trend of the data captured and non-seasonal peaks were predicted by the model. These forecasts were proven to be realistic and a true reflection of economic reality in the country. The paper recommended a non-seasonalð´ð‘…ð¼ð‘€ð´ 3, 1, 0 be used by researchers, policy makers and decision makers of different countries to make forecasts of household debt. The South African authorities were also encouraged to use this model to produce further forecasts of the series when making long term planning.