The Application of Neural Network Models to Explain the Relationship Between Stock Value, Returns Value, and Information on Prices

Abstract

This study aims to identify features for stock market value predictions based on time series grouping. This research uses various data sets for different activities. We have the balance sheet data set, which includes a series of quarterly balance sheets. Next, there is the ratio data set. The clustering data set consists of a series of daily prices. There are two data sets for testing activities: pilot forecasts and investments with daily data, projections, and investments. To speed up the development and exploration of different methods, testing begins with a subset of the data, with additional shares of witnesses. This research uses ARIMA and artificial neural networks to predict stock prices. Predicted results provide an investment strategy that compares the results obtained with what happens, resulting in generally positive profits. The capacity of artificial neural networks to manage non-linearities in financial data explains this. Second, many experiments demonstrate that not all data can be utilized using predicting techniques. These results indicate that improved estimations do not always result from more significant data. Forecast accuracy and performance on the test set will likely be negatively impacted by noise and competing signals introduced by data from unrelated stocks. Put another way, by including unrelated data, the information about the action supplied by the group of which it is a member is obscured. As such, the forecasts more closely align with the market's average behavior than with the performance of a specific stock.