Financial Time Series Forecasting Research Articles

Expert Aggregation for Financial Forecasting

Machine learning algorithms dedicated to financial time series forecasting have gained a lot of interest over the last few years. One difficulty lies in the choice between several algorithms, as their estimation accuracy may be unstable through time. In this paper, we propose to apply an online aggregation-based forecasting model combining several machine learning techniques to build a portfolio which dynamically adapts itself to market conditions. We apply this aggregation technique to the construction of a long-short-portfolio of individual stocks ranked on their financial characteristics and we demonstrate how aggregation outperforms single algorithms both in terms of performances and of stability.

A Multi-Method Survey on the Use of Sentiment Analysis in Multivariate Financial Time Series Forecasting.

In practice, time series forecasting involves the creation of models that generalize data from past values and produce future predictions. Moreover, regarding financial time series forecasting, it can be assumed that the procedure involves phenomena partly shaped by the social environment. Thus, the present work is concerned with the study of the use of sentiment analysis methods in data extracted from social networks and their utilization in multivariate prediction architectures that involve financial data. Through an extensive experimental process, 22 different input setups using such extracted information were tested, over a total of 16 different datasets, under the schemes of 27 different algorithms. The comparisons were structured under two case studies. The first concerns possible improvements in the performance of the forecasts in light of the use of sentiment analysis systems in time series forecasting. The second, having as a framework all the possible versions of the above configuration, concerns the selection of the methods that perform best. The results, as presented by various illustrations, indicate, on the one hand, the conditional improvement of predictability after the use of specific sentiment setups in long-term forecasts and, on the other, a universal predominance of long short-term memory architectures.

Research on Intelligent Forecasting of Financial Market Parameter Time Series Based on Deep Learning

Objectives: Deep learning has become the most representative and potential intelligent system modeling technology in artificial intelligence. However, the complexity of financial markets goes far beyond all economic games. Methods: This paper is devoted to the feasibility and efficiency of the deep-integration neural network model as one of the main paradigms of in-depth learning in the intelligent prediction of financial time. A prediction model of stack self-coding neural network composed of bottom stack self-coding and top regression neurons is proposed. Results: Firstly, the self-encoder unsupervised learning mechanism is used to identify and learn the time series, and the layers of the neural network are learned greedy layer by layer. Then the stack self-encoder is extended to the SAEP model with supervised mechanism, and the parameters learned by SAE are used. Used to initialize the neural network, and finally use the supervised learning to fine-tune the weights. Conclusion: The research results show that the model provides effective financial planning and decision-making basis for financial forecasting, maintains the healthy development of financial markets, and maximizes the benefits of profit-making institutions.

Analysis of autocorrelation function of stochastic processes by F-transform of higher degree

The autocorrelation function of a stochastic process is one of the essential mathematical tools in the description of variability that is successfully applied in many scientific fields such as signal processing or financial time series analysis and forecasting. The aim of the paper is to provide an analysis of fuzzy transform of higher degree applied to stochastic processes with a focus on its autocorrelation function. We introduce a higher-degree fuzzy transform of a stochastic process and investigate its basic properties. Further, we introduce a higher-degree fuzzy transform of bivariate functions in the tensor product of polynomial spaces and demonstrate its approximation ability. In addition, we show that the bivariate higher-degree fuzzy transform of multiplicative separable functions is a product of univariate fuzzy transform of the respective functions, which is a desirable property in the processing of functions of higher dimensions. The obtained results are used to demonstrate an interesting identity between the autocorrelation function of the fuzzy transform of a stochastic process and the bivariate fuzzy transform of the autocorrelation function of the stochastic process. This identity provides two ways to determine the autocorrelation function of the fuzzy transform of a stochastic process, which can be useful, especially in a situation where its direct calculation becomes complex or even impossible, for example, the calculation of a sample autocorrelation function.

Artificial chemical reaction optimisation of recurrent functional link neural networks for efficient modelling and forecasting of financial time series

In contrast to multilayer neural networks, a functional link artificial neural network uses functional expansion units for transferring lower input space to higher dimensions. It achieves enhanced ...

Artificial chemical reaction optimisation of recurrent functional link neural networks for efficient modelling and forecasting of financial time series

Artificial chemical reaction optimisation of recurrent functional link neural networks for efficient modelling and forecasting of financial time series

Financial Time Series Forecasting: Comparison of Traditional and Spiking Neural Networks

One of the most common applications of neural networks itself is data prediction models, for example, future stock market prices, calculated based on historical data. Spiking neural networks are one of the emerging architectures showing great potential in solving complex problems in complicated information environments. However, to the best of our knowledge, the spiking neural networks have not been successfully applied in stock market data prediction. The values of exchange-traded funds (ETF), due to their flexibility and simplicity, can be a good application of such a tool. Therefore, the following article provides the results of a comprehensive experimental comparison of different spiking neural networks in predicting ETF values. The main goal was to check if the spiking neural networks obtain better or worse results of forecasting than traditional neural networks. The secondary goal was a comparison of different spiking neural network architectures between themselves to judge which one is the most applicable to the given problem

Learning to Rank for Multi-Step Ahead Time-Series Forecasting

Time-series forecasting is a fundamental problem associated with a wide range of engineering, financial, and social applications. The challenge arises from the complexity due to the time-variant property of time series and the inevitable diminishing utility of predictive models. Therefore, it is generally difficult to accurately predict values, especially in a multi-step ahead setting. However, in domains such as financial time series forecasting, an ex-ante prediction of the relative order of values in the near future is sufficient; i.e., the next 100 days can help make meaningful investment decisions. In this paper, we propose a dynamic prediction framework that makes it possible to make an ex-ante forecast of time series with a special focus on the relative ordering of the forecast within a forward-looking time horizon. Through the lens of the concordance index (CI), we compare the proposed method with conventional regression-based time-series forecasting methods, discriminative learning methods and hybrid methods. Moreover, we discuss the use of the proposed framework for different types of time series and under a variety of conditions. Extensive experimental results on financial time series across a majority of liquid asset classes show that the proposed framework outperforms the benchmark methods significantly.

ANN-Time Varying GARCH Model for Processes with Fixed and Random Periodicity

Financial Time Series Forecasting is an important tool to support both individual and organizational decisions. Periodic phenomena are very popular in econometrics. Many models have been built aiding capture of these periodic trends as a way of enhancing forecasting of future events as well as guiding business and social activities. The nature of real-world systems is characterized by many uncertain fluctuations which makes prediction difficult. In situations when randomness is mixed with periodicity, prediction is even much harder. We therefore constructed an ANN Time Varying Garch model with both linear and non-linear attributes and specific for processes with fixed and random periodicity. To eliminate the need for time series linear component filtering, we incorporated the use of Artificial Neural Networks (ANN) and constructed Time Varying GARCH model on its disturbances. We developed the estimation procedure of the ANN time varying GARCH model parameters using non parametric techniques.

Study on the application of LSTM-LightGBM Model in stock rise and fall prediction

This paper proposes a hybrid financial time series forecast model based on LSTM and LightGBM, namely LSTM_LightGBM model. Use the LightGBM model to train the processed stock historical data set, and save the training results. Then the opening price, closing price, highest price, lowest price, trading volume and adjusted closing price are separately input into the LSTM model for prediction. The prediction result of each attribute is used as the test set of the prediction after LightGBM training. Constantly adjust the parameters of each model, and finally get the optimal stock price forecast model. The model is validated with the rise and fall of AAPL stock. Through the comparison of evaluation index root mean square error RMSE, mean absolute error MAE, prediction accuracy Accuracy and f1_score. It is found that the LSTM_LightGBM model exhibits stable and better prediction performance in the stock prediction. That is to say, the LSTM_LightGBM model proposed in this paper is stable and feasible in the stock price fluctuation forecast.

Spatiotemporal adaptive neural network for long-term forecasting of financial time series

Optimal decision-making in social settings is often based on forecasts from time series (TS) data. Recently, several approaches using deep neural networks (DNNs) such as recurrent neural networks (RNNs) have been introduced for TS forecasting and have shown promising results. However, the applicability of these approaches is being questioned for TS settings where there is a lack of quality training data and where the TS to forecast exhibit complex behaviors. Examples of such settings include financial TS forecasting, where producing accurate and consistent long-term forecasts is notoriously difficult. In this work, we investigate whether DNN-based models can be used to forecast these TS conjointly by learning a joint representation of the series instead of computing the forecast from the raw time-series representations. To this end, we make use of the dynamic factor graph (DFG) to build a multivariate autoregressive model. We investigate a common limitation of RNNs that rely on the DFG framework and propose a novel variable-length attention-based mechanism (ACTM) to address it. With ACTM, it is possible to vary the autoregressive order of a TS model over time and model a larger set of probability distributions than with previous approaches. Using this mechanism, we propose a self-supervised DNN architecture for multivariate TS forecasting that learns and takes advantage of the relationships between them. We test our model on two datasets covering 19 years of investment fund activities. Our experimental results show that the proposed approach significantly outperforms typical DNN-based and statistical models at forecasting the 21-day price trajectory. We point out how improving forecasting accuracy and knowing which forecaster to use can improve the excess return of autonomous trading strategies.

Dynamics of Heteroscedasticity Modelling and Forecasting of Tax Revenue in a Developing Economy: A Review

Tax revenue modelling and forecasting is very crucial for revenue collection and tax administration management. The dynamics of heteroscedasticity in the financial time series (tax revenue) in the domain of technique used to model and predict tax revenue in the emerging economy threw us to this investigation. The reviews are categorized into two the tax revenue and stock exchange index. Five factors were considered in this studies modelling, forecasting, linear model, nonlinear model and heteroscedasticity, it is on this note that we syntheses over 75 studies from the literature to consider the pattern of reporting tax revenue and stock market index. Thus, from the reviewed literature, we inferred that the pattern of reporting tax revenue data and the analytical techniques employed by most of these studies are responsible for the instability (volatility) in the financial time series forecasting. Also, results revealed that linear models are mostly applied to tax revenue data with fewer non-linear models, while combination and single non-linear models were mostly used for stock exchange data. Thus, we recommend the combination of linear and nonlinear models for both tax revenue and stock exchange data which can minimize the error of heteroscedasticity in the forecasting of tax revenue in a developing economy.

A Hybrid Approach Integrating Multiple ICEEMDANs, WOA, and RVFL Networks for Economic and Financial Time Series Forecasting

The fluctuations of economic and financial time series are influenced by various kinds of factors and usually demonstrate strong nonstationary and high complexity. Therefore, accurately forecasting economic and financial time series is always a challenging research topic. In this study, a novel multidecomposition and self-optimizing hybrid approach integrating multiple improved complete ensemble empirical mode decompositions with adaptive noise (ICEEMDANs), whale optimization algorithm (WOA), and random vector functional link (RVFL) neural networks, namely, MICEEMDAN-WOA-RVFL, is developed to predict economic and financial time series. First, we employ ICEEMDAN with random parameters to separate the original time series into a group of comparatively simple subseries multiple times. Second, we construct RVFL networks to individually forecast each subseries. Considering the complex parameter settings of RVFL networks, we utilize WOA to search the optimal parameters for RVFL networks simultaneously. Then, we aggregate the prediction results of individual decomposed subseries as the prediction results of each decomposition, respectively, and finally integrate these prediction results of all the decompositions as the final ensemble prediction results. The proposed MICEEMDAN-WOA-RVFL remarkably outperforms the compared single and ensemble benchmark models in terms of forecasting accuracy and stability, as demonstrated by the experiments conducted using various economic and financial time series, including West Texas Intermediate (WTI) crude oil prices, US dollar/Euro foreign exchange rate (USD/EUR), US industrial production (IP), and Shanghai stock exchange composite index (SSEC).

Comparative Analysis of Deep Learning Models for Multi-Step Prediction of Financial Time Series

Financial time series prediction has been a key topic of interest among researchers considering the complexity of the domain and also due to its significant impact on a wide range of applications. In contrast to one-step ahead prediction, multi-step forecasting is more desirable in the industry but the task is more challenging. In recent days, advancement in deep learning has shown impressive accomplishments across various tasks including sequence learning and time series forecasting. Although most previous studies are focused on applications of deep learning models for single-step ahead prediction, multi-step financial time series forecasting has not been explored exhaustively. This paper aims at extensively evaluating the performance of various state-of-the-art deep learning models for multiple multi-steps ahead prediction horizons on real-world stock and forex markets dataset. Specifically, we focus on Long-Short Term Memory (LSTM) network and its variations, Encoder-Decoder based sequence to sequence models, Temporal Convolution Network (TCN), hybrid Exponential Smoothing- Recurrent Neural Networks (ES-RNN) and Neural Basis Expansion Analysis for interpretable Time Series forecasting (N-BEATS). Experimental results show that the latest deep learning models such as N-BEATS, ES-LSTM and TCN produced better results for all stock market related datasets by obtaining around 50% less Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) scores for each prediction horizon as compared to other models. However, the conventional LSTM-based models still prove to be dominant in the forex domain by comparatively achieving around 2% less error values.

A two-step machine learning approach to predict S&P 500 bubbles

ABSTRACT In this paper, we are interested in predicting the bubbles in the S&P 500 stock market with a two-step machine learning approach that employs a real-time bubble detection test and support vector machine (SVM). SVM as a nonparametric binary classification technique is already a widely used method in financial time series forecasting. In the literature, a bubble is often defined as a situation where the asset price exceeds its fundamental value. As one of the early warning signals, prediction of bubbles is vital for policymakers and regulators who are responsible to take preemptive measures against the future crises. Therefore, many attempts have been made to understand the main factors in bubble formation and to predict them in their earlier phases. Our analysis consists of two steps. The first step is to identify the bubbles in the S&P 500 index using a widely recognized right-tailed unit root test. Then, SVM is employed to predict the bubbles by macroeconomic indicators. Also, we compare SVM with different supervised learning algorithms by using k-fold cross-validation. The experimental results show that the proposed approach with high predictive power could be a favourable alternative in bubble prediction.

Asymptotic properties of M-estimator for GARCH(1, 1) model parameters

GARCH(1, 1) model is used for analysis and forecasting of financial and economic time series. In the classical version, the maximum likelihood method is used to estimate the model parameters. However, this method is not convenient for analysis of models with residuals distribution different from normal. In this paper, we consider M-estimator for the GARCH(1, 1) model parameters, which is a generalization of the maximum likelihood method. An algorithm for constructing an M-estimator is described and its asymptotic properties are studied. A set of conditions is formulated under which the estimator is strictly consistent and has an asymptotically normal distribution. This method allows to analyze models with different residuals distributions; in particular, models with stable and tempered stable distributions that allow to take into account the features of real financial data: volatility clustering, heavy tails, asymmetry.

A hybrid stock price index forecasting model based on variational mode decomposition and LSTM network

Changes in the composite stock price index are a barometer of social and economic development. To improve the accuracy of stock price index prediction, this paper introduces a new hybrid model, VMD-LSTM, that combines variational mode decomposition (VMD) and a long short-term memory (LSTM) network. The proposed model is based on decomposition-and-ensemble framework. VMD is a data-processing technique through which the original complex series can be decomposed into a limited number of subseries with relatively simple modes of fluctuations. It can effectively overcome the shortcomings of mode mixing that sometimes exist in the empirical mode decomposition (EMD) method. LSTM is an improved version of recurrent neural networks (RNNs) that introduces a “gate” mechanism, and can effectively filter out the critical previous information, making it suitable for the financial time series forecasting. The capability of VMD-LSTM in stock price index forecasting is verified comprehensively by comparing with some single models and the EMD-based and other VMD-based hybrid models. Evaluated by level and directional prediction criteria, as well as a newly introduced statistic called the complexity-invariant distance (CID), the VMD-LSTM model shows an outstanding performance in stock price index forecasting. The hybrid models perform significantly better than the single models, and the forecasting accuracy of the VMD-based models is generally higher than that of the EMD-based models.

A new financial data forecasting model using genetic algorithm and long short-term memory network

Financial data forecasting is conducive to get a better understanding of the future economic situation. Recently, variational mode decomposition (VMD) is introduced into the field of financial data forecasting. However, the prediction accuracy of the current methods is low. We propose a new VMD-based financial data prediction model. In this model, the genetic algorithm is utilized to optimize the parameters of VMD. Then VMD decomposes the data sequence into long-term and short-term trends. Finally, we employ the long short-term memory (LSTM) network to predict the future data with inputs generated by VMD. The contributions are: (1) We propose an improved VMD and LSTM based financial data forecasting model; (2) A guideline on the parameter selection of VMD to process financial data is designed; (3) A prediction-error reducing method which reduces the inherent error caused by VMD insensitivity to fluctuation is proposed. Experimental results indicate our model is accuracy-promising and superior to the baseline models in one-step-ahead forecasting of financial time series.

A fast conformal predictive system with regularized extreme learning machine

A conformal predictive system(CPS) is based on the learning framework of conformal prediction, which outputs cumulative distribution functions(CDFs) for labels in regression problems. The CDFs output by a CPS provide useful information for users, as they not only provide probability for the events related to the test labels, but also can be transformed to prediction intervals with the corresponding quantiles. Moreover, CPSs have the property of validity since the distributions and intervals they output have statistical compatibility with the realizations. This property is very useful for many risk-sensitive applications such as financial time series forecast and weather forecast. However, as based on conformal predictors, CPSs inherit the computational issue. To build a fast CPS, in this paper, we propose a CPS with regularized extreme learning machine as the underlying algorithm. To be specific, we combine the leave-one-out cross-conformal predictive system(Leave-One-Out CCPS), a variant of the original CPS, with regularized extreme learning machine(RELM), which is named as LOO-CCPS-RELM. We analyse the computational complexity of it and prove its asymptotic validity based on some regularity assumptions. We also prove that the error rate of the prediction interval output by LOO-CCPS-RELM is under control in the asymptotic setting. Experiments with 20 public data sets were conducted to test LOO-CCPS-RELM and the results showed that LOO-CCPS-RELM is empirically valid and compared favourably with the other CPSs.

Assimilation of Principal Component Analysis and Wavelet with Kernel Support Vector Regression for Medium-Term Financial Time Series Forecasting

Entities and institutional financiers have gained a lot of growth from financial time series forecasting in recent times. But the major challenges of financial time series data are the high noise and complexity of its nature. Researchers in recent times have successfully engaged the application of support vector regression (SVR) to conquer this challenge. In this study principal component analysis (PCA) is applied to extract the low dimensionality and efficient feature information, while wavelet is used to pre-process the extracted features in other to nu1llify the influence of the noise in the features with a KSVR based forecasting model. The analysis is carried out based on the quarterly tax revenue data of 39 years from the first quarter of 1981 to the last quarter of 2016. The forecasting is made for ten quarters ahead. The initial empirical result shows that the multicollinearity has been reduced to zero (0), and the analytic result reveals that the proposed model PCA-W-KSVR outperforms KSVR, PCA-KSVR, and W-KSVR in terms of MAE, MAPE, MSE and RMSE.