Reparameterization Of Autoregressive Distributed Lag To Vector Error Correction Model To Study Youth Unemployment In Kenya

Abstract

The research provides statistical basis for assessing and prioritizing investment policies, initiatives and projects to maximise youth employment by scrutinizing in uence of macroeconomic variables. The macroeconomic variables considered are gross domestic product (GDP), external debt (ED), foreign domestic investment (FDI), private investment(PI), youth unemployment(YUN), literacy rate (LR), and youth population (POP). The research approach taken uses predictive analytics such as impulse response functions and variance decomposition from vector error corrections model (VECM) and cointegration regression in autoregressive distributed lag (ARDL) to identify key determinants of youth unemployment to prioritize investment. This research analyzes reparameterization of ARDL to VECM through cointegration of time series. First, the time series data undergo logarithm transformation to reduce outlier e ects and have elasticity interpreted in terms of percentage. The study scrutinizes the e ects of macroeconomic shocks on youth unemployment in Kenya. For this purpose, the Augmented Dickey-Fuller test is conducted to assess stationarity of the variables used. Then Johansen Cointegration test is carried out to establish the rank at which the series are cointegrated. The unit root test has been performed on YUN, GDP, ED, FDI, PI, and LR, and POP to assess stationarity. The cointegrated dynamic ARDL model is estimated using ordinary least squares (OLS) and e ects of variables and their lags interpreted. The results reveal that Gross Domestic Product (GDP) and its second lag have negative e ect on youth unemployment, that is, one unit increase in (GDP) and GDP lag 2 reduce youth unemployment by 0.207922% and 0.2052705% respectively. Also, one unit of External Debt (ED) and ED lag 2 reduce youth unemployment by 0.07303% and 0.009116% respectively. Furthermore, unit increase in one year lag of youth literacy rate reduces youth unemployment by 0.0892691%. Lastly, lag one and three of population reduce youth unemployment by 0.2590455% and 4.3093119% respectively. The Johansen Cointegration Analysis has revealed three long run relationships which can be interpreted as a GDP e ect; External Debt e ect and Foreign Direct Investment e ect relations. A structural VECM has been described through restrictions taken from the Cointegration Analysis. Based on the results of the Impulse-Response Function and variance decomposition analyses of the Structural VECM, it is concluded that GDP, literacy level, population, and FDI shocks have signi cant iii e ects on Kenyan youth unemployment in the long run. On the superiority of the two models, whereas ARDL captures the in uence of past shocks through coe cients of lags, VECM predicts the e ects of current shocks and resulting movement of variables more than 10 unit steps ahead. Also, Granger causality present in ARDL does not exist in reparameterized VECM. The F-test and t-test reveal that the two models are signi cant at 95% con dence level. However CUSUM test shows that the estimated ARDL is more stable.