Multivariate Time Series Analysis of Ethiopian International Air Travel Demand

No Thumbnail Available



Journal Title

Journal ISSN

Volume Title


Addis Abeba university


Ethiopian Airline is playing a leading role in transforming Addis Ababa and Ethiopia into a world class aviation hub of the African continent not only for trade and business but also for tourism. For the continual of its success in a better way, forecasting air travel demand plays a crucial role for an overall effective planning. An air travel (passengers‘) demand forecast is a scientifically formed opinion about future air traffic. This demand is most preferably measured by Load Factor which relates the proportion of seats purchased to flight distance covered by a given period. This study utilized a monthly data from January 2009 up to December 2013 to construct multivariate time series model, vector autoregression (VAR) model, and investigation is made on the reaction of a study target variable Load Factor (LF) to Passenger Revenue (PR), Block Hours (BH), and Distance Flown (DF) at international level. First and foremost all series are seasonally adjusted after they were known to be seasonal through standard tests built in X-12 ARIMA program in Eviews 7 Statistical Software. Post-seasonal adjustment tests also assured that all series are non-seasonal. Stationarity is checked before and after differencing using visual inspection, unit root tests, and variance comparison. Each series are found to be integrated of order one (I (1)). The three information criteria AIC, SC and HQ recommended one lag length. Johansen cointegration test indicated only one long-term equilibrium relationship occurred between the variables. This immediately implied the legitimacy of vector error correction (VEC) model of order one to be fitted than a pure VAR (1) model for the time series data. Exogeneity test also indicated that only LF and PR are endogenous variables. But to determine the short-run bonds between series, impulse response functions and variance decompositions are employed. Granger causality test is also conducted to identify the total possible causal effects among the variables. As one footstep before out-of-sample forecasting, the VECM (1) model has been checked for its accuracy with the aid of RMSE, MAE, MAPE and Theil-U statistics. The summary result of VECM (1) shown that Load Factor (as a measure of travel demand) is Granger caused by all variables in the short-run except Passenger Revenue and significantly explained by all variables in the long-run. At last, forecasting is made for Ethiopian International air travel demand (Load Factor). KEYWORDS: Load Factor, Vector Autoregression (VAR), Cointegration, Vector Error Correction model (VECM), and forecasting



Load Factor, Vector Auto regression (VAR), Cointegration, Vector Error Correction Model (VECM), and Forecasting