Comparative study of hybrid artificial neural network methods under stationary and nonstationary data in stock market

Abstract

In this study, a new methodology is proposed to automatically determine six parameters of artificial neural network using population-based metaheuristics. We considered following three issues: What is the effect of used metaheuristic on performance? Which parameters are mostly selected? Is there a difference between the forecasting results when using stationary or nonstationary dataset that are selected according to the augmented Dickey-Fuller test statistics? Based upon results of performance measures, proposed method leads to significant opportunities to forecast stock market more effectively. We also expect proposed methodology can provide remarkable advantages for other complex, dynamic, and nonlinear forecasting problems. | INTRODUCTIONStock market is one of the most important investment ways due to the potential of earning higher returns. It is also considered as one of the key indicators of economic growth of countries. Over the years, researchers, practitioners, and investors analyzed the stock market, the stocks, and the relationships. However, stock market has high complexity and nonlinear constraints to grasp. Although anybody can own stocks to perform the aspired profit, many investors find stock market hard to detect which stock to buy, or when to buy and sell stocks. Furthermore, it is so difficult to exactly say when the stock market price will increase, how much it will increase or how long the increase of price will remain (Albeladi & Abdullah, 2018).Due to high volatilities and noises in stock market, unilateral traditional forecasting method and analysis cannot meet investors' need any more. Hence, new methods with recent advances in artificial intelligence have been proposed to extract useful information from stock market data. Depending on the nature of the stock market, a particular method can be used to gain profit. However, there is no single model exist in literature. Each method has its own strengths and limitations in terms of performance and reliability. In this case, artificial intelligence methods can be integrated to combine the strength of methods and overcome a single method's weakness. Especially, artificial neural network (ANN) method can be proposed for stock market forecasting in order to improve investors' decision support quality. In this paper, genetic algorithm (GA), differential evolution (DE), and particle swarm optimization (PSO) are integrated with ANN to clarify the dynamic and nonstationary nature of the stock market. GA, DE, and PSO are population-based metaheuristics. The group of populationbased metaheuristics inherently has some significant advantages. First, they ensure information about the "surface" of an objective function.Second, population-based metaheuristics are less sensitive to "improper" pathways of certain individual solutions. Finally, they improve the probability of attaining the global optimum (Feoktistov, 2006). This study involves the following tasks: (a) determine the relevant technical indicators for the stocks, (b) determine the proper transfe...

A Novel Hybrid Artificial Intelligence Based Methodology for the Inventory Routing Problem

“…In addition, some of the main parameters of the ANN, including the number of neurons in hidden layers and training algorithm, are optimized by the GA. The initial pool of the training algorithm is the same as the Dosdogru [41], except for bayesian regularization. Note that log-sigmoid is utilized as an activation function in the proposed ANN method.…”

Short-term prediction of bank deposit flows: do textual features matter?

“…Mai et al, (2019) document that the back-propagation algorithm upgrades the weights of the model throughout the training process. Based on the mathematics, each value of an input pattern A ∈ R N is linked with weight value W ∈ R N which takes values between 0 and 1 (Dosdogru, 2019). Given that F(x) is the function that computes the output from the neurons, this output could be represented with the following mathematical formula:…”