Cross–Temporal Coherent Forecasts for Gross Domestic Product

Timely and accurate forecasts aligning different views of economic agents are of utmost importance in macroeconomic forecasting to facilitate effective policy decisions. Thus, this study investigates the ability of a reconciliation approach to align different viewpoints regarding forecasts and thereby increasing the forecast performance specifically related to GDP forecasting. The proposed methodology is based on forecasting hierarchical time series which is a collection of time series that follow an inherent aggregation structure. The aggregation constraints can be cross-sectional or temporal dimension. Thus, this method attempts to reconcile forecasts so that they follow aggregation constraints in both dimensions. This property is referred to as cross-temporal coherency. As the initial step forecasts are obtained for each of the series in the cross-temporal hierarchy. These are referred to as base forecasts and are often incoherent. These forecasts are then revised so that they become crosstemporally coherent. This is referred to as cross-temporal forecast reconciliation. Empirical applications based on disaggregated economic activities of the production approach for Sri Lankan GDP reveal that this approach brings improvement in forecast accuracy by blending different viewpoints in a data driven way. These cross-temporal coherent forecasts align decisions within an organisation transparently towards one number by aligning short term forecasts with long term forecasts and aligning views at different levels within the GDP hierarchy. As the proposed method is independent of forecasting models different short term forecasting models and long term forecasting models can be used to reflect different viewpoints.