Testing for Non-linearity in the Balancing Item of Balance of Payments Accounts: The Case of 20 Industrial Countries

The non-linearity of financial and economic time series is becoming a fundamental issue both at the theoretical and empirical level. This applies equally to balance of payments statistics such as balancing item (net errors and omissions), which is a residual variable needed to ensure that all debit and credit entries in the balance of payments statement sum to zero. This univariate variable is only one of a number of accounting balances that can be established in economic statistics that are used in assessing the accuracy of the balance of payments statistics. Except for Tang et al (forthcoming), the existing works which have studied the balancing item either disregard the presence of non-linearity or assume at the outset that the non-linearity takes a particular form. An AR(p) model is employed in the present paper to remove any linear structure from the balancing item series. Applying a battery of non-linearity tests, the assumption of linearity is tested for 20 industrial countries' balancing item. In general, the statistical tests reveal the presence of non-linear dependencies in the balancing item series for 16 out of the 20 industrial countries. An implication is that the non-linear dynamics of these balancing item series should be incorporated in the modelling and forecasting exercises for a majority of the countries.