Modelling and forecasting of Tunisian current account: Aggregate versus disaggregate approach

While there is considerable literature attempting to model current account, there are not many studies to forecast current account balance. This study gives a comprehensive way to model and predict current account deficit (CAD) by evaluating the forecasting performance of direct and indirect approach. At the disaggregated level, I use two variants to model current account components; in the first alternative I apply different ARIMA models with exogenous variables (ARIMA-X) to account for the pattern of the data and exogenous factors. In the second alternative, I integrate the cointegration relationship between exports and imports with ARIMA-X models. With respect to the direct approach, I use error correction model to allow for dynamics in current account. The data used spans from January 2000 to December 2014 and comes from the Central Bank of Tunisia, the Tunisian National Institute of Statistics, and the OECD database. I find that for one-step ahead forecast, both ARIMA-X and reduced form model produce accurate forecast. However, with respect to dynamic forecasts, direct method is more accurate when compared to ARIMA-X. When cointegrating relationship between exports and imports is combined with ARIMA-X models, the indirect approach outperforms the direct approach. I also show that, as volatility of underlying components increase disaggregate approach using time series models become less reliable. In addition, I found that current account is mainly affected by domestic GDP, trade openness, fiscal deficit, exchange rate, credit to the private sector and partner GDP. Estimation of ECM indicates that persistent effect is high and can take more than three quarters to die out. In addition I assess the performance of direct and indirect approach over time using naïve approach as benchmark. It appears that the MSE of naïve approach lies between direct and indirect approach in average up to horizon 12, but then worsen.