Predicting High-Growth Firms in Kazakhstan with Machine Learning Methods

Abstract

In this paper, we study the effectiveness of popular machine learning methods for predicting high-growth firms in Kazakhstan and analyze this question with a set of 2012-2018 panel datasets. Moreover, we study the most important variables for the prediction of high-growth firms out of 50 variables included in the analysis. We develop a predictive design, where the past values are used to train classifiers that are applied in predicting future outcomes. Hereto, a test sample was used to evaluate the predictive performance of the classifiers. The results indicate that the best performing classifier increases the area under the curve equal to 0.8746. In terms of the variable importance, the firm’s past growth in size, past growth in employment, past growth in revenue, and second derivative of the growth of financial variables contributed the most to predicting high-growth firms.