An increasing hybrid morphological-linear perceptron with pseudo-gradient-based learning and phase adjustment for financial time series prediction

The recent work in the author's research group on using independent component analysis (ICA) for the analysis and prediction of financial time series is reviewed. ICA belongs to the group of linear transform methods, with the goal to make a transform from the observed signals into a signal space in which the signals are statistically independent. Sometimes independence can be attained, especially in blind source separation in which the observed signals are assumed to be linear mixtures of independent source components. Then the goal of ICA is to invert the unknown mixing operation. Even when independence is not possible, as is often the case in financial time series, the ICA transformation produces useful component signals whose dependence is reduced, and that are nonGaussian with a density allowing sparse coding. The ICA transformation is also related to the temporal structure of the found signals as measured by Kolmogorov complexity or its approximations. The signals are structured and hence may be easier to interpret and predict. After discussing the ICA criterion within the context of linear signal transforms, the FastICA algorithm is reviewed. It is a computationally efficient method for finding a subset or all of the component signals. Then, two financial applications are covered: decomposition of parallel financial time series of weekly sales into basic factors, and prediction of mixture time series by linear combinations of predictions on the ICA components.

Financial time series are chaotic that, in turn, leads their predictability to be complex and challenging. This paper presents a novel financial time series prediction hybrid that involves Chaos Theory, Convolutional neural network (CNN), and Polynomial Regression (PR). The financial time series is first checked in this hybrid for the presence of chaos. The chaos in the series of times is later modeled using Chaos Theory. The modeled time series is input to CNN to obtain initial predictions. The error series obtained from CNN predictions is fit by PR to get error predictions. The error predictions and initial predictions from CNN are added to obtain the final predictions of the hybrid model. The effectiveness of the proposed hybrid (Chaos+CNN+PR) is tested by using three types of Foreign exchange rates of financial time series (INR/USD, JPY/USD, SGD/USD), commodity prices (Gold, Crude Oil, Soya beans), and stock market indices (S&P 500, Nifty 50, Shanghai Composite). The proposed hybrid is superior to Auto-regressive integrated moving averages (ARIMA), Prophet, Classification and Regression Tree (CART), Random Forest (RF), CNN, Chaos+CART, Chaos+RF and Chaos+CNN in terms of MSE, MAPE, Dstat, and Theil’s U.

GMRVV m–SVR model for financial time series forecasting

Financial time series are nonlinear, volatile and chaotic. Inspired by quantum computing, this paper proposed a new model, called primary ensemble empirical mode decomposition combined with quantum neural network (PEEMD-QNN) in predicting the stock index. PEEMD-QNN takes the advantages of the PEEMD which retains the main component of modal component and QNN. To demonstrate that our PEEMD-QNN model is robust, we used the new model to predict six major stock index time series in China at a specific time. Detailed experiments are implemented for both of the proposed prediction models, in which empirical mode decomposition combined with QNN (EMD-QNN), QNN and BP neural network are compared. The results demonstrate that the proposed PEEMD-QNN model has higher accuracy than BP neural network, QNN model and EMD-QNN model in stock market prediction.

Prediction of financial time series using artificial neural networks has been the subject of many publications, even if the predictability of financial series remains a subject of scientific debate in the financial literature. Facing this difficulty, analysts often consider a large number of exogenous indica- tors, which makes the fitting of neural networks extremely difficult. In this paper, we analyze how to aggregate a large number of indicators in a smaller number using -possibly nonlinear- projection methods. Nonlinear projection methods are shown to be equivalent to the linear Principal Component Analysis when the prediction tool used on the new variables is linear. Furthermore, the computation of the nonlinear projection gives an objective way to evaluate the number of resulting indicators needed for the prediction. Finally, the advantages of nonlinear projection could be further exploited by using a subsequent nonlinear prediction model. The methodology developed in the paper is validated on data from the BEL20 market index, using systematic cross-validation results. Classification Codes: G00, G14.

Financial time series prediction using artificial neural network based on Levenberg-Marquardt algorithm

A dynamic neural network (DNN) and a new computationally efficient functional link artificial neural network (CEFLANN) combination optimised with differential evolution (DE) is presented in this paper to predict financial time series like stock price indices and stock return volatilities of two important Indian stock markets, namely the Reliance Industries Limited (RIL), and NIFTY from one day ahead to one month in advance. The DNN comprises a set of 1st order IIR filters for processing the past inputs and their functional expansions and its weights are adjusted using a sliding mode strategy known for its fast convergence and robustness with respect to chaotic variations in the inputs. Extensive computer simulations are carried out to predict simultaneously the stock market indices and return volatilities and it is observed that the simple IIR–based DNN–FLANN model hybridised with DE produces better forecasting accuracies in comparison to the more complicated neural architectures.

Always being aware of future economic and investment is important and significant for investors, companies, politicians and public people, in order to come to know the future economic perspective and accordingly how particular policies perform. There are different models in the field of prediction of financial time series that we can perform investment with lower risk by using them. Whatever prediction quantity is more accurate, the investment performs with lower risk. The proposed method of this paper is MA-SVR, which is a combination of memetic algorithm and support vector regression. We use memetic computation to estimate support vector regression method parameters and perform prediction with optimisation of SVR method parameters by using memetic algorithm. We evaluate the proposed MA-SVR method by use of prediction error with RMSE criterion, and compare it with other effective algorithms and we show that the proposed method causes improvement in the result from the view point of closeness of predicted quantities to reality.

This paper presents the prediction of financial time series using an adaptive neural network, which is called the self-organised multilayer perceptrons inspired by the immune algorithm. The simulation results were compared with the multilayer perceptrons and the functional link neural networks. The prediction capability of the various neural networks was tested on ten different data sets; the US/UK exchange rates, the JP/US exchange rate, the US/EU exchange rates, NASDAQO time series, NASDAQC time series, DJIAO time series, DJIAC time series, DJUAO time series, DJUAC time series and the oil price. We carried out two sets of experiments. In the first set, the input data are pre-processed between 0 and 1 and passed to the neural networks as non-stationary data. In the second set of experiments, the non-stationary input signals are transformed into stationary signals and passed to the neural networks. The predictions demonstrated that all neural networks generate profit using stationary data and fail to generate any profit when using non-stationary data. Furthermore, the experiment results for the stationary signals showed that the (SOMLP) outperforms the MLP neural network based on the profit. However, it produced a lower profit when predicting stock price in comparison with the Functional Link neural networks.

BP neural network is promising methods for the prediction of financial time series because it use a risk function consisting of the empirical error and a regularized term which is derived from the structural risk minimization principle. This study applies BP neural network to predicting the stock price index. In addition, this study examines the feasibility of applying BP neural network in financial forecasting. The experimental results show that BP neural network provides a promising alternative to stock market prediction.

The financial time series is inherently nonlinear and hence cannot be efficiently predicted by using linear statistical methods such as regression. Hence, intelligent predictor has been developed and reported which is suitable for nonlinear time series. But such predictors require that the past financial data are available at the location of the predictor which is not the case in many real-life situations. Hence, when the financial data are available at different places and a single intelligent predictor needs to be developed, the task becomes challenging. In the current work, this problem has been addressed and solved using a low-complexity artificial neural network and employing incremental and diffusion learning strategies. In the current study, distributed prediction of three different types of time series such as exchange rates, stock indices and net asset values has been carried using incremental and diffusion-based learning strategies. The results of different days ahead prediction of two proposed low computational complexity-based functional link artificial neural network are compared with those obtained by conventional intelligent method. The results of simulation-based experiments reveal similar or improved prediction performance of the proposed distributed predictors compared to conventional one. In addition, saving in band width, memory and power are achieved in this method.

It is a critical challenge in financial time series analysis to reduce noise and forecast future stock prices. In this paper, we propose Opemod, an optimal performing selection model to predict and adaptively select prediction modes based on performance. Opemod is designed with three parts: two-ends extension mode decomposition (TEEMD) algorithm, attention based encoder and decoder (AED) model, and optimal performing selection (OPS) algorithm. Firstly, we propose TEEMD algorithm to restrain the end effect of sequence decomposition by differently extending and truncating both two ends of the sequence, and the original financial time series are decomposed into intrinsic mode functions (IMFs) more accuracy by TEEMD. Secondly, we design a novel encoding and decoding model based on both LSTM and multi-head attention mechanism (AED) to capture both long-term and short-term dependence information. Thus the trends of IMFs can be predicted separately. Finally, after the processes of decomposition and prediction, OPS is used to select modes with the best performance. Extensive experiments have been carried out on CSI 300 index and Dow Jones index (DJI) datasets, and the results show that Opemod can get better investment return than other state-of-the-art methods.KeywordsFinancial time seriesForecastingEmpirical mode decompositionDeep learning

This paper continues research within the framework of the scientific direction in econophysics at the Department of Information Systems and Mathematical Methods in Economics of the faculty of Economics of PSU. Modeling and prediction of financial time series is quite a perspective area of research, because it allows participants of financial processes to reduce risks and make effective decisions. For example, we could research financial processes with the help of fractal analysis. In the article there is studied and worked out in detail one of the methods of fractal analysis of financial time series – the box-counting method for assessment of the fractal dimension. This method is often used in studies conducted by domestic authors, but the authors do not delve into the characteristics and problems of using the box-counting method for analysis of time series, that means that the answers to the interested questions have not yet been given. The main problem is that, as a rule, the analyzed object in the tasks of applying the box-counting method to time series is a computer image of the plot of series. In the article there is proposed the procedure of adaptation of the box-counting method for assessment of the fractal dimension of time series, the procedure does not require the formation of a computer image of the plot. In the article there is considered following difficulties developed from this adaptation: 1) high sensitivity of the resulting estimation of the dimension to the input parameters of the method (the ratio of the sides of the covered by cells plane with the plot; the used range of lengths of the side of the cell; the number of partitions of the plane into cells); 2) the non-obviousness of choosing the optimal values ​​of these parameters. In the article there are analyzed approaches to the selection of these parameters that were proposed by other authors, and there are determined the most suitable approaches for the adapted box-counting method. Also there are developed unique methods for determining the ratio of the sides of the plane with the plot. In the paper there is written the computer program that implements the developed method, and this program is tested on the generated data. The study obtained the following results. The fact of sensitivity of the adapted box-counting method to input parameters is confirmed, that indicates the high importance of the correct choice of these parameters. According to the study, there is found out inability of the proposed methods of automatic determination the ratio of the sides of the plane in relation to artificial time series. There are obtained the most precise (in a statistical sense) estimates of fractal dimension, those found by means of the adapted box-counting method, with the fixed ratio of the sides 1:1. According to comparing the adapted box-counting method and R/S analysis, there are obtained the most precise estimates by the second method (R/S analysis). Finally in the paper there are formulated the possible directions for further research: 1) comparison of the accuracy of various methods for assessment of the fractal dimension on series of different lengths; 2) comparison of the methods of fractal analysis and p-adic analysis for modeling and prediction of financial time series; 3) determination of the conditions of applicability of various methods; 4) approbation of the developed methods for determining of the ratio of the sides of the plane with the plot on real economic data.

Discontinuities in financial time-series data caused by market holidays can significantly affect the accuracy of predictive models. This study explores interpolation strategies to address holiday-induced gaps in daily stock prices of eight major NASDAQ technology companies (2019-2024), using data from Yahoo Finance. Three approaches-linear interpolation, nearest interpolation, and no interpolation-were compared using machine learning models to forecast next-day closing prices. Input features included 7-day lagged price data and technical indicators such as SMA, EMA, and Bollinger Bands. Model performance was evaluated using MAE and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{2}$</tex>. Results show that linear interpolation consistently yielded the lowest mean MAE, reducing error by 23.9% compared to no interpolation, and improved mean <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex> by 1.15%, demonstrating higher explanatory power. Linear interpolation outperformed both alternative methods for every company examined, highlighting its substantial benefit for forecasting accuracy and model robustness in financial time-series prediction.

1: Architectures.- Scale Independence in the Visual System.- Dynamic Neuronal Information Processing of Vowel Sounds in Auditory Cortex.- Convolutional Spiking Neural Network for Robust Object Detection with Population Code using Structured Pulse Packets.- Networks Constructed of Neuroid Elements Capable of Temporal Summation of Signals.- Predictive Synchrony Organized by Spike-Based Hebbian Learning with Time-Representing Synfire Activities.- Improving Chow-Liu Tree Performance by Mining Association Rules.- A Reconstructed Missing Data-Finite Impulse Response Selective Ensemble (RMD-FSE) Network.- Higher Order Multidirectional Associative Memory with Decreasing Energy Function.- Fast Indexing of Codebook Vectors Using Dynamic Binary Search Trees with Fat Decision Hyperplanes.- 2: Learning Algorithms.- On Some External Characteristics of Brain-like Learning and Some Logical Flaws of Connectionism.- Superlinear Learning Algorithm Design.- Extension of Binary Neural Networks for Multi-class Output and Finite Automata.- A Memory-Based Reinforcement Learning Algorithm to Prevent Unlearning in Neural Networks.- Structural Optimization of Neural Networks by Genetic Algorithm with Degeneration (GAd).- Adaptive Training for Combining Classifier Ensembles.- Combination Strategies for Finding Optimal Neural Network Architecture and Weights.- 3: Applications.- Biologically Inspired Recognition System for Car Detection from Real-Time Video Streams.- Financial Time Series Prediction Using Non-Fixed and Asymmetrical Margin Setting with Momentum in Support Vector Regression.- A Method for Applying Neural Networks to Control of Nonlinear Systesm.- Robot Manipulator Control via Recurrent Neural Networks.- Gesture Recognition Based on SOM Using Multiple Sensors.- Enhanced Phrase-Based Document Clustering Using Self-Organizing Map (SOM) Architectures.- Discovering Gene Regulatory Networks from Gene Expression Data with the Use of Evolving Connectionist Systems.- Experimental Analysis of Knowledge Based Multiagent Credit Assignment.- Implementation of Visual Tracking System Using Artificial Retina Chip and Shape Memory Alloy Actuator.