?
Algorithms for Labour Income Share Forecasting: Detecting of Intersectoral Nonlinearity
We examine different algorithms to forecast labour share for 18 KLEMS-classified economic sectors, 12 European countries. The choice is driven by data availability. For each sector 11 specifications of time component in CES production function with factor-augmenting technical change are tested. This includes comparing models with linear, nonlinear time and the same with structural breaks. Then, three degrees of models ‘power’ are proposed to characterize whether a model is consistent and valid for prediction. Here, residuals stationarity and autocorrelation as well as regressors and structural breaks statistical significance are investigated. To sum up main results, models with structural break in nonlinear time component show better predictive power according to the derived criteria. Next, overall labour share decline cannot be stated as only 7 sectors out of 18 have decreasing trend in more than one third of cases (countries). Additionally, each country sectors are grouped by LS forecast average value into four interval categories.