Abstract: Quantitative evaluation of financial variables plays a foundational role in financial price modeling, economic prediction, risk evaluation, portfolio management, etc. However, the problem suffers from high dimensionality. Thus, financial variables should be selected in a way to reduce the dimensionality of the financial model and make the model more efficient. In addition, it is quite common for financial datasets to contain missing data due to a variety of limitations. Consequently, in practical situations, it is difficult to choose the best subset of financial variables due to the existence of missing values. The two problems are interrelated. Therefore, the central idea in this research is to develop and examine new techniques for financial variable selection based on estimating the missing values, while accounting for all the longitudinal and latitudinal information.
This research proposes a novel methodology to minimize the problem associated with missing data and the best subset of financial variables that could be used for effective analysis. There are two major steps; the first step concentrates on estimating missing data using Bayesian updating and Kriging algorithms. The second step is to find the best subset of financial variables. In this step a novel feature subset selection is proposed (LmRMR) which ranks the financial variables and the best subset of variables is chosen by employing statistical techniques through SSTC measurement. Some tests have been done to demonstrate the applicability and effectiveness of the ideas presented in this research. In particular, the potential application of the proposed methods in stock market trading model and stock price forecasting are studied. The experimental studies are conducted on Dow Jones Industrial Average financial variables.