Template-type: ReDIF-Article 1.0
Author-Name: Zhanna Shuvalova
Author-Email: shuvalovazhd@cbr.ru
Author-Workplace-Name: Bank of Russia
Title: Forecasting Fixed Capital Investment with Patent Activity Indicators
Abstract: The paper examines the potential of patent activity indicators for forecasting fixed capital investment. We rely on comparative analysis of accuracy of classical econometric models (Random Walk, ARIMA, ARIMAX) and machine learning techniques (Ridge Regression, Random Forest, as well as various types of boosting such as AdaBoost, Gradient Boosting, XGBoost, LightGBM, CatBoost), with and without patent variables, to conclude that the inclusion of these variables in models improves the accuracy of short-term investment forecasts. CatBoost, which incorporates patent activity variables, demonstrates the lowest forecast error compared to the other models. Our findings also confirm that patent activity variables have a significant and positive impact on fixed capital investment.
Classification-JEL: C53, E22, O31, O32, O34
Keywords: investments, patent activity, machine learning, CatBoost, HLN test, Kalman filter, forecasting
Journal: Russian Journal of Money and Finance
Pages: 37-66
Volume: 85
Issue: 2
Year: 2026
Month: June
DOI:
File-URL: https://rjmf.econs.online/upload/documents/RJMF-85-2-Forecasting-Fixed-Capital-Investment.pdf
Handle: RePEc:bkr:journl:v:85:y:2026:i:2:p:37-66