Network Model For Time Series Research Articles

Hybrid wavelet-neural network models for time series

The use of wavelet analysis contributes to better modeling for financial time series in the sense of both frequency and time. In this study, S&P500 and NASDAQ data are separated into several components utilizing multiresolution analysis (MRA). Subsequently, using an appropriate neural network structure, each component is modeled. In addition, wavelets are used as an activation function in long short-term memory (LSTM) networks to form a hybrid model. The hybrid model is merged with MRA as a proposed method in this paper. Four distinct strategies are employed: LSTM, LSTM+MRA, hybrid LSTM-Wavenet, and hybrid LSTM-Wavenet+MRA. Results show that the use of MRA and wavelets as an activation function together reduces the error the most.

CLPVG: Circular limited penetrable visibility graph as a new network model for time series.

A visibility graph transforms time series into graphs, facilitating signal processing by advanced graph data mining algorithms. In this paper, based on the classic limited penetrable visibility graph method, we propose a novel mapping method named circular limited penetrable visibility graph, which replaces the linear visibility line in limited penetrable visibility graph with nonlinear visibility arc for pursuing more flexible and reasonable mapping of time series. Tests on degree distribution and some common network features of the generated graphs from typical time series demonstrate that our circular limited penetrable visibility graph can effectively capture the important features of time series and show higher robust classification performance than the traditional limited penetrable visibility graph in the presence of noise. The experiments on real-world time-series datasets of radio and electroencephalogram signals also suggest that the structural features provided by a circular limited penetrable visibility graph, rather than a limited penetrable visibility graph, are more useful for time-series classification, leading to higher accuracy. This classification performance can be further enhanced through structural feature expansion by adopting subgraph networks. All of these results demonstrate the effectiveness of our circular limited penetrable visibility graph model.

Analysis of oil price fluctuation under the influence of crude oil stocks and US dollar index — Based on time series network model

This paper uses the high-frequency oil price after ensemble empirical mode decomposition, combines crude oil stocks and the US dollar index to construct co-movement threshold networks according to different periods divided by multiple structural breakpoint method. The results show that in the period of sharp fluctuations in oil prices, the inverse relationship between the US dollar index and high-frequency oil prices is more obvious, and the co-movement network of the three is closer to the “small world network”. After oil prices returned to stable fluctuations, coupled with the weakening role of the United States in leading the global economy, the reverse relationship between the US dollar index and high-frequency oil prices has gradually weakened, and the reverse relationship between stocks and high-frequency oil prices has become increasingly prominent. The research not only reveals the main co-movement characteristics of the three, but also provides a new idea for future research on oil price fluctuations and its influencing factors.

Investigation of stock price network based on time series analysis and complex network

Complex network is now widely used in a series of disciplines such as biology, physics, mathematics, sociology and so on. In this paper, we construct the stock price trend network based on the knowledge of complex network, and then propose a method based on information entropy to divide the stock network into some communities, that is, a gathering study of stock price trend. We construct time series networks for each stock in Chinese A-share market based on time series network model, and then use these networks to divide the stock market into communities. We find that the average trend of stocks in the same community is the same as the trend of market value weighting, but the average trend of stocks in different communities is quite different and the sequence correlation is low. This conclusion shows that stocks in the same community share the same price trend, while the stock trend in different communities varies. This paper is a successful application of complex network and information entropy in stock trend analysis, which mainly includes two contributions. First, the success of the visibility graph algorithm provides a new perspective for enriching stock price trend modeling. Second, our conclusion proves that the clustering based on information entropy theory is effective, which provides a new method for further research on stock price trend, portfolio construction and stock return prediction.

A novel error-output recurrent neural network model for time series forecasting

It is a well-known fact that improving forecasting accuracy is an important yet often challenging issue. Extensive research has been conducted using neural networks (NNs) to improve their forecasting accuracy. In general, the inputs to NNs are the auto-regressive (i.e. lagged variables) of one or more time series. In addition, either network outputs or network errors have been used as extra inputs to NNs. In this paper, however, we propose a novel recurrent neural network forecasting model which is called the ridge polynomial neural network with error-output feedbacks (RPNN-EOF). RPNN-EOF has two main types of inputs: auto-regressive and moving-average inputs. The former is represented by the lagged variables of a time series, while the latter is represented by feeding back network error to the input layer. In addition, network output is fed back to the input layer. The proposed recurrent model has the ability to produce more accurate forecasts due to the advantages of learning temporal dependence and the direct modelling of the moving-average component. A comparative analysis of RPNN-EOF with five neural network models was completed using ten time series. Simulation results have shown that RPNN-EOF is the most accurate model among all the compared models with the time series used. This shows that employing auto-regressive and moving-average inputs together helps to produce more accurate forecasts.

Exploring the fluctuant transmission characteristics of Air Quality Index based on time series network model

Abstract Although many works have focused on the issues related to Air Quality Index (AQI) from the perspective of statistics or economics, the dynamic and temporal characteristics of AQI time series remain unclear. The main reason is that AQI time series has both non-linearity and complexity, which cannot be characterized by conventional ways. The inherent transmission laws of AQI volatility are of great importance for environment monitoring and management system. Thus, this paper has endeavored to address the issue from the interdisciplinary perspective of complex network theory. First, we collect the annual and the total AQI data of three cities (Shanghai, Wuhan and Guangzhou) in China from January 1, 2015 to December 31, 2017. Then the coarse graining method is employed in combination with the theory of phase space reconstruction to convert AQI time series into complex network. The coarse graining method is used to transform yearly and total AQI time series into the corresponding symbol sequences. In accordance with the theory of phase space reconstruction, we adopt C C method and false nearest neighbor algorithm to estimate the optimal time delay and embedding dimension of AQI time series, respectively. The optimal time delay is integrated with the embedding dimension to map symbol sequences into AQI complex network. Based on complex network theory, we investigate the transmission of AQI fluctuation by the temporal characteristics and dynamics of yearly AQI networks (YAN) and total AQI networks (TAN), which include node strength, betweenness centrality and community structure. The empirical results show: (1) the AQI networks of various periods display small-world effect; (2) network evolution tends to be more complex and the modality in YAN has less alternative transformation paths than that in TAN; (3) the node strength and betweenness centrality in measuring the importance of nodes achieve a relative consistency; (4) the three cities have the same dominant fluctuation modalities both in YAN and in TAN; (5) the uneven distribution of transmission ability between clusters implies a preferential transmission whether in YAN or TAN. These findings not only help us to interpret the frequent volatility of AQI and control some extreme short-term AQI fluctuation, but also contribute to the reliability assessment of AQI and the prediction of AQI fluctuation. In summary, this paper not only reveals the temporal dynamic properties of AQI time series, but also provides some policy recommendations for environmental monitoring and management.

A new linear & nonlinear artificial neural network model for time series forecasting

Artificial neural network approach is a well-known method that is a useful tool for time series forecasting. Since real life time series can generally contain both linear and nonlinear components, hybrid approaches which can model both these two components have also been proposed in the literature. The hybrid approaches suggested in the literature generally have two phases. In the first phase, linear component of time series is modeled with a linear model. Then, nonlinear component is modeled by utilizing a nonlinear model in the second phase. In two-phase methods, it is assumed that time series has only a linear structure in the first phase. Also, it is assumed that time series has only a nonlinear structure in the second phase. Therefore, this causes model specification error. In order to overcome this problem, a novel neural network model, which consists of both linear and nonlinear structures, is proposed in this study. The proposed model considers that time series has both linear and nonlinear components. Multiplicative and Mc Culloch–Pitts neuron structures are employed for nonlinear and linear parts of the proposed model, respectively. In addition, the modified particle swarm optimization method is used to train the proposed neural network model. In order to show the performance of the proposed approach, it is applied to three real life time series and obtained results are compared to those obtained from other approaches available in the literature. It is observed that the proposed model gives the best forecasts for these three time series.

Building neural network models for time series: a statistical approach

This paper is concerned with modelling time series by single hidden layer feedforward neural network models. A coherent modelling strategy based on statistical inference is presented. Variable selection is carried out using simple existing techniques. The problem of selecting the number of hidden units is solved by sequentially applying Lagrange multiplier type tests, with the aim of avoiding the estimation of unidentified models. Misspecification tests are derived for evaluating an estimated neural network model. All the tests are entirely based on auxiliary regressions and are easily implemented. A small-sample simulation experiment is carried out to show how the proposed modelling strategy works and how the misspecification tests behave in small samples. Two applications to real time series, one univariate and the other multivariate, are considered as well. Sets of one-step-ahead forecasts are constructed and forecast accuracy is compared with that of other nonlinear models applied to the same series. Copyright © 2006 John Wiley & Sons, Ltd.