Specification of regression analysis of the impact of the information environment on the company's financial indicators

The subject of the research is the development and experimental validation of a comprehensive regression specification designed for the quantitative assessment of the elasticity of market stock values to thematic information flows. The object of the research includes daily time series of thematic intensities, extracted by the Latent Dirichlet Allocation algorithm from the industry news corpus, and the stock exchange differential "closing-opening" of the shares of PJSC "GMK Norilsk Nickel". The author thoroughly examines aspects such as Corr–γ–split normalization, which eliminates the bimodality of distributions, the orthogonalization of "scale-asymmetry," which reduces multicollinearity, Partial Least Squares projection for aggregating features, and regularized Ridge regression for robust forecasting. Special attention is given to how the combination of these stages forms a statistically sound and interpretable bridge between textual signals and financial metrics, ensuring the practical applicability of the model to the dynamics of high-frequency informational disturbances. The methodological foundation consists of Corr–γ–split normalization, "Sum/Diff" orthogonalization, Partial Least Squares projection, and Ridge regression with cross-validation, combined in a full factorial experiment of forty-five alternative specifications. The main conclusions of the conducted research are the confirmation that only a comprehensive integration of Corr–γ–split normalization, "Sum/Diff" orthogonalization, PLS projection, and Ridge regression forms a statistically robust and practically applicable model of the influence of news background on market price. The novelty of the work lies in the introduction of a metrically justified threshold T*, which eliminates the inherent bimodality of LDA intensity distributions, as well as in the development of interpretable decompositions of flows into size and asymmetry, which enhances the explanatory power of elasticity coefficients. The empirical testing on data from PJSC "GMK Norilsk Nickel" showed a reduction in RMSE by 13%, an increase in CV-R² to 0.78, and an improvement in the aggregate quality score by 0.32 compared to the baseline model. The obtained results prove that the proposed specification is scalable to various corporate or industry information flows and can serve as a reliable tool for monitoring and forecasting market indicators in the context of high-frequency informational disturbances.

Information environment, Regression analysis, Elasticity, Financial performance, Bayesian priors, Multi-criteria calibration, Heteroskedasticity, Time lags, PLS projection, Ridge regression