LDA vs PCA in analyzing companies with OFC activity

Alexandra Georgiana SIMA
, Gheorghe HURDUZEU
, Stefan Alexandru IONESCU

Keywords

LDA, PCA, OFC, classification, financial indicators

Abstract

Continuing previous research and aiming to build a system of indicators and risk assessment procedures for companies operating in CFOs, we have extracted and processed financial data for the years 2017-2019 for approximately 8300 companies in over 40 jurisdictions. The data has been processed so that values can be compared, regardless of jurisdiction, currency, or accounting standard. Given the fact that we follow companies from several industries, for a period of more than a year, we chose to use financial data that can be found for as many companies as possible, just to be able to obtain a sufficiently large sample. We performed the PCA analysis using the covariance matrix. The first two principal components preserve about 76% of the variance of the initial causal space, and if we add the third, we have over 84% of the initial information. Next, we employed LDA techniques in order to determine which of the selected indicators provides a better classification of the companies selected in the analysis. Last but not least, we were able to make predictions about the membership of new companies, based on the built model.