Forecasting, Cointegration and Causality Analysis of Unemployment Using Time Series Models

Abstract

One of the major issue for policy makers is handling with continues increase in the level of unemployment in Pakistan. Thus forecasting unemployment rate is imperative to policy makers. This study aims to explore the best forecasting model among ARIMA, ARFIMA and exponential smoothing for forecasting unemployment. Secondly this study analyzed unemployment using time series techniques, measured long & short run relationship with population growth, labor force participation rate and crop production, and also investigated the causality between unemployment and other variables. Time series data ranging from 1965 to 2014 is collected from Pakistan Economic Survey for analysis. This study evaluate the forecasting performance of three models by using the forecast accuracy criterion such mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE) and Theil’s U statistics. Double Exponential Smoothing model is chosen as a best forecasted model for unemployment rate on the basis of forecast accuracy criterion. Augmented Dickey Fuller (ADF) and Phillips-Perron (PP) test is used for checking stationarity in the variables. At level the variables were non stationary and become stationary at first difference. The results of Johnson cointegration and Vector Error Correction model (VECM) indicated that there exists long & short run cointegration relationship between unemployment rate and other variables. Granger Causality test shows bi-directional causality running from crop production toward population growth.