Algorithms for estimating labour productivity prediction based on regression and dynamic models

Gross domestic product is one of the most important indicators of the national accounts system. It characterizes the final result of productive activity of domestic economic units and measures the value of goods and services produced by these units in the country for a certain period of time for final use. The paper considers an algorithm for constructing a linear stationary model in terms of the state space for describing behaviour and predicting labour productivity state depending on the capitallabour ratio and labour costs, instead of describing the object given by the three-factor non-linear Cobb-Douglas model. In addition to a general construction of a dynamic model the algorithm includes the description of the procedures in the form of recursive formulae allowing calculation of variance magnitudes for the dynamic noise, the measuring system noise and the initial state of the investigated object behaviour on the basis of statistical time series data related to labour productivity. An example shows that the proposed model provides more efficient prediction estimates of labour productivity values compared with the prediction estimates calculated using a three-factor non-linear regression model. The numerical calculations of the accuracy of prediction estimates were made using an absolute percentage error and the Teil formula for both the three-factor Cobb-Douglas model and the model in the form of the state space. The calculations showed more accurate prediction estimates and more adequate filtering estimates for the model in terms of the state space.