Metaheuristic enabled intelligent model for stock market prediction via integrating volatility spillover: India and its Asian and European counterparts

Abstract

Recently, the price of a stock market changes often owing to a variety of factors. As a result, making an accurate stock price prediction is a difficult process. Hence, this research work proposes a novel intellectual stock market prediction model that incorporates the volatility spillover over Indian and its Asian countries. This intellectual model mainly involves two phases like data library construction and stock market prediction. For stock market prediction, a Neural Network (NN) model is employed and this model intake the data of calculated indicators in the data library and makes the prediction of Indian market. To attain more precise prediction, the NN weight is optimally chosen via novel hybrid algorithm namely Fly Updated Whale Optimization Algorithm (FU-WOA) that is the hybridization of WOA and Firefly Algorithm (FF). At last, the suggested model performance is exploited by comparing other conventional models in the view of various metrics. Especially, the computational cost of the proposed hybrid FU-WOA–NN model is 38.12%, 15.96%, 15.52%, 41.22%, 16.07% and 16.33% better than existing LM-NN, FF-NN, GWO-NN, WOA-NN, PCA as well as ARIMA methods respectively.

Introduction

The international incorporation of financial markets [1], [2] has encouraged quite a lot of fresh empirical learning for mechanism investigation, by which the transmission of stock market movements are wandered around the world [3], [4], [5]. Financial markets are utilized all over the world for many purposes, such as product recommendation [6] and so on. Because of the increasing interconnection of major financial markets around the world [7], [8], stock market information transmission between major Asian markets has emerged as the most interesting topic. The previous researches have concerned mainly on the first-moment spillover of stock prices, i.e. between the main stock exchanges termed as a return. However, the stock market co-movement exploration is associated learning of information spillover [9], [10], [11] regarding the returns and also the return’s volatility. The understanding of the ‘spillover’ component in the market is also an imperative task to be performed by the investors at the time of portfolio allocation and risk management. Spillovers are imitated as transform in domestic volatilities or stock market returns, because of content’s information transmission from one market to another by diverse transmission channels.

Over a period of time, numerous researchers have tried to capture co-movement and return volatility spillovers [12], [13], [14] among the markets to account diversification benefits. Many researchers have made to incarcerate the explained effects between developed and emerging markets over a crisis period. Adding further to this literature, the current learning tries to incarcerate the dynamic co-movement and return volatility spillovers among the most challenging and opportunities instilled emerging equity markets ‘BRIC’1, which comprised of Brazil, Russia, India and China by employing Asymmetric BEKK [15] and DCC Asymmetric GARCH (1, 1) models [16], [17] and by taking VAR [18] as the ‘mean’ equation model in a multivariate framework, i.e. under four-dimensional framework. To capture the time-varying correlation dynamics during the global financial crisis (2007–2009) period, a heat map and Markov regime switching model (two regimes with a switch at the ‘mean’ level only) are employed, making the study first of its kind in all the perspectives.

In previous years, deep learning, machine learning & optimization models are used in the field of stock market prediction. For instant, Stoean et al. [19] designed a LSTM and CNN models with the HC heuristic strategy for prediction of the close price. Sedighi et al. [20] use ANFIS, a SVM, and ABC algorithm to estimate market values for the 50 largest US companies. In the Turkish stock market, Göçken et al. [21] evaluated the efficiency of technical indicators such as a simple moving average close price and a close momentum price. In this study, the hybrid ANN is coordinated with HSAas well as GA to find the relation between the stock exchange & technical indicators. CRO algorithm was used by Nayak and Misra [22] to find optimal weights and biases of MLP with a single hidden layer which was applied to the BSE, NASDAQ, DJIA, FTSE, and TAIEX indices for close price prediction.

Based on the above information, the optimization methods are better and it have shown to be more effective than traditional methods in handling stock price forecasting difficulties. The reported results, however, can still be improved to develop more reliable answers to forecasting difficulties. Local search challenges are the most common problems that existing search methods encounter, and there is a lack of variation in the answers. Furthermore, as stated in [23], the primary flaw in this problem is that utilizing only one forecasting model is ineffective due to the problem’s complexity [24]. The nonlinear models for price forecasting have been proved in the majority of recent studies, but only a handful have incorporated advanced machine learning and optimization techniques. This assertion is backed up by [25], which uses a new sophisticated optimization technique as a prediction system based on unstructured data to produce a more realistic and consistent forecast. All of the published research in the literature agreed that forecasting stock prices is a difficult topic that requires a practical solution [26].

The main contribution of this study is given below:

• A novel stock market prediction phase dependent on NN is proposed to present an optimal prediction scheme with enhanced accuracy. The NN model uses data from calculated indicators in the data library to predict the Indian market.

• Proposes a hybrid optimization, namely Fly Updated Whale Optimization Algorithm (FU-WOA) for fine-tuning the weights of NN.

• FU-WOA is introduced to overcome the drawbacks of traditional methods like premature convergence of WOA as well as slow convergence speed of FF.

Organization of the paper: Section 2 discusses the literature survey on stock market prediction. The proposed stock market prediction model influencing volatility spillover is expressed in Section 3. The results as well as their comments are evaluated in Section 4. The paper comes to a close with Section 5.