Assessing the multielasticity of risk factors for stress testing of investment products on the Russian stock market

Subject. The article deals with the development of advanced risk assessment method for retail investment products (mutual funds, ETFs, and structured notes) purchased on the Russian stock market. Objectives. The main goal is to create a qualitative stress testing model that takes into account the interdependence of risk factors. Methods. The study employs machine learning modeling, in particular, the vector error correction model (VECM) method enabling to take into account the non-stationarity of time series and cointegration relationships between macroeconomic indicators. Statistical tests for stationarity, multicollinearity, and cointegration were conducted to select the factors. We applied the cross-validation approach and calculation of the forecast error using the MAPE metric. Results. We selected and trained a model based on the vector error correction model method. It helped consider the non-stationarity of time series and interrelationship of risk factors, which made the estimates of other models incorrect. The final mean absolute percentage error was 19.41%, which means that the forecast is highly accurate. The risk factors involved in the modeling were inflation [1] (both food basket and wages), exchange rates, oil and the key rate, given their mutual influence on each other [2]. Conclusions. The offered method will enable a deeper and more comprehensive stress test of retail investment products based on the vector error correction model machine learning method. The integrated approach to assessing risk factors will increase investor protection by better informing about the maximum level of losses under adverse market conditions.