Forecast Linear Augmented Projection (FLAP) is a post-forecast adjustment method that can reduce forecast error variance in multivariate time series. In FLAP, components containing information about shared features are constructed as linear combinations of the original time series. The forecasts of the original time series and the components are then projected such that the linear relationship between the historical data is enforced on the forecasts. While forecast error variance reduction has been theoretically proven regardless of the linear combination, the empirical performance of different component types is less clear and is examined in this paper. Components considered in this paper are estimated by maximising measures of information and/or by minimising the dependency between components. Among other methods, using FLAP with Principal Component Analysis is recommended for its stable performance across settings, while Forecastable Component Analysis offers a strong alternative, as demonstrated by simulations and application to Australian tourism data.