Free «The Role of Accounting Indicators Based on Artificial Neural Networks to Predict Stock Prices» Essay

1. Introduction

The use of artificial neutral networks has contributed significantly towards prediction of stock market prices, because they have shown the capability to predict stock prices, future stocks and the trends in stocks performance (Trippi & Turban, 1996). Many accounting indicators implement artificial neural networks (ANN) method to predict stocks prices. This has resulted into the need to understand the manner in which these accounting indicators contribute to accomplishment of this function while ensuring accuracy of predicting stocks prices. Other accounting indicators that have been used to understand stock prices based on ANN include earnings per share return ration on assets, current ratios and net cash flow per share (Cao, Leggio & Schniederjans). This paper presents the role of various accounting indicators and their relationship with ANN towards prediction of stocks prices. Also, this study provides the first attempt towards application of ANN during the process of predicting Saudi Arabia stocks prices and comparing its accuracy with other methods that have been reportedly used to predict stocks prices in literature. Furthermore, it investigates the impacts of moving averages as a method that can be used to investigate the future value of stocks prices in the Saudi Stock Exchange.

2. Purpose of the Study

Current study is focused on finding the roles of accounting indicators in the Saudi Stock Exchange based on artificial neural networks. It investigates how various accounting indicators implement a multilayer artificial neural network design for prediction of technical indicators and variables aimed at finding effectiveness of ANN on prediction of Tawadul Stock prices and its advantage over autoregressive models of predicting stocks prices.

3. Discussion

3.1. Artificial Neural Networks

Artificial neural network is a strategy that is analogous to the neural network of human body where neuron enables conveyance of information in a particular direction while units in these directions receive the information. In ANN, the units are trained to perform specific tasks by adjusting these weights (Huang, Chen & Wu, 2004). It also involves continuous adjustments of weights through comparison of the network with the target until a match between the network and the target is achieved while the error function existing between the network and the target is minimized. It involves the use of a number of pairs of inputs and outputs during network training. In order to minimize errors, several approaches are used, for example, a standard optimization method like Network-type approach.

Training of network is accomplished by use of two forms: incremental training and batch training. During incremental training, pairs are fed individually which is followed by comparing the output with the expectation of each input and weights are adjusted using training algorithm such as backpropagating algorithm (Jasic & Wood, 2004). This process is followed by feeding the next pair to the network after adjusting the previous pair and repeating the process. During batch-training, pairs of input and out-put are put into the network at the first stage then the weights are adjusted. When ANN is trained well, it enables exploitation of the basic nonlinear relationships which contribute to accuracy of security prices.

In the context of financial management, ANN is used to forecast the prices of stocks since it contributes to lower out-of-sample short-term predictions for each return in a day.

3.2. Various Accounting Indicators that Apply ANN

An example of accounting indicator that applies the use of artificial neural networks is the time series forecasting that focuses on analysis of past data and comes up with a future forecast of data values for estimation of stock prices. It involves modeling of a non-linear function as a result of recurrence relations from past values (Cheng & Titterington, 1994). This relation is then used to predict future values of stocks. The main types of time series forecasting techniques are univariate and multivariate techniques where univariate technique is composed of one variable in the recurrent equation. The values used in equations are those corresponding to past values of the moving averages. The multivariate model contains more than one equation in its model.

Neural networks can work in parallel with these variables and contribute to handling of large combinations of data within a short duration of time. The network enables finding the patterns and irregularities in addition to identification of detecting where there are irregularities or nonlinearity of data (Zhang, 1998). These attributes of ANN are important in enabling the process of designing a system in stock market. Application of neural networks allows prediction of stock market prices through learning of non-linear mappings between variables used in equation and the resulting values.

The reasons for studying stock prices in Saudi Arabia include the fact that it is a young market and has a number of asymmetric information statuses, which result into lack of implementation of market hypothesis to its affection (Tkacz, 2001). In addition, stock prices are subject to change as from annual reports. There are also few institute investors in Saudi Arabia that makes Saudi Arabian Stock a subject to changes. Consequently, tracking these changes ensures the need to track the future values of stocks by use of accounting principles which apply ANN as a method of prediction.