Multivariate imputation of xbrl data for financial statement analysis

XBRL company filingsprovide immediate availability and easy accessibility, for both researchers and investors, for financial statement analysis. The objective of this study is to examine whether large scale XBRL data can be used to predict the direction of movement of earnings. The study analyzes companies' XBRL filings of quarterly data using a two-step Logit regression model. The model is then used to arrive at the probability of the directional movement of earnings between current quarter and subsequent quarter. The results classified the companies as ones that would realize an increase, or a decrease, in earnings. Although the final model indicated an ability to predict subsequent earnings changes on average about 67% of the time, (similar to those of previous studies based on COMPUSTAT), it based the models on about 23% of the entire sample examined, and could classify less than 10% of the entire sample. A Multivariate Imputation by Chained Equations (MICE) was implemented to fill in the missing data. This increased the number of useable observations by about 11%, and increased the number of observations in the final models by 150%. The models utilized 56% of the original companies (more than double) and classified 27% of the original companies (about triple), and still increased the accuracy of prediction to 68%. These results suggest that XBRL data with imputation can be used as a financial statement analysis tool.

Amos Baranes and Rimona Palas
Journal Name: 
Int J Inf Res Rev
Volume No: 
Issue No: 
Paper Number: 
Select Subjects: 
Download PDF: