Financial Time Series Prediction Based on XGBoost and Generative Adversarial Networks

Considering the problems of the model collapse and the low forecast precision in predicting the financial time series of the generative adversarial networks (GAN), we apply the WGAN-GP model to solve the gradient collapse. Extreme gradient boosting (XGBoost) is used for feature extraction to improve prediction accuracy. Alibaba stock is taken as the research object, using XGBoost to optimize its characteristic factors, and training the optimized characteristic variables with WGAN-GP. We compare the prediction results of WGAN-GP model and classical time series prediction models, long short term memory (LSTM) and gate recurrent unit (GRU). In the experimental stage, root mean square error (RMSE) is chosen as the evaluation index. The results of different models show that the RMSE of WGAN-GP model is the smallest, which are 61.94% and 47.42%, lower than that of LSTM model and GRU model respectively. At the same time, the stock price data of Google and Amazon confirm the stability of WGAN-GP model. WGAN-GP model can obtain higher prediction accuracy than the classical time series prediction model.

Research on the Improvement of Big Data Feature Investment Analysis Algorithm for Abnormal Trading in the Financial Securities Market

Traditional investment analysis algorithms usually only analyze the similarity between financial time series and financial data, which leads to inaccurate and inefficient analysis of investment characteristics. In addition, the trading volume of financial securities market is huge, the amount of investment data is also very large, and the detection of abnormal transactions is difficult. The aim of feature extraction is to obtain mathematical features that can be recognized by machine. Different from the traditional methods, this paper studies and improves the big data investment analysis algorithm of abnormal transactions in financial securities market. After processing the captured trading data of financial securities market, the big data feature of abnormal trading is extracted. Combined with the abnormal trading and the financial securities market, the investment strategy is determined. The optimization objective function is set and the genetic algorithm is used to improve the investment analysis algorithm. The simulation experiment verifies the improved investment analysis algorithm, and the average Accuracy of investment analysis is increased by at least 11.24%, the ROI is significantly improved, and the efficiency is higher, which indicates that the proposed algorithm has ideal application performance.

Estimating the memory parameter for potentially non-linear and non-Gaussian time series with wavelets

The asymptotic theory for the memory-parameter estimator constructed from the log-regression with wavelets is incomplete for 1/$f$ processes that are not necessarily Gaussian or linear. Having a complete version of this theory is necessary because of the importance of non-Gaussian and non-linear long-memory models in describing financial time series. To bridge this gap, we prove that, under some mild assumptions, a newly designed memory estimator, named LRMW in this paper, is asymptotically consistent. The performances of LRMW in three simulated long-memory processes indicate the efficiency of this new estimator.

Investigating the Effect of Noise Elimination on LSTM Models for Financial Markets Prediction Using Kalman Filter and Wavelet Transform

Predicting financial markets is of particular importance for investors who intend to make the most profit. Analysing reasonable and precise strategies for predicting financial markets has a long history. Deep learning techniques include analyses and predictions that can assist scientists in discovering unknown patterns of data. In this project, application of noise elimination techniques such as Wavelet transform and Kalman filter in combination of deep learning methods were discussed for predicting financial time series. The results show employing noise elimination techniques such as Wavelet transform and Kalman filter, have considerable effect on performance of LSTM neural network in extracting hidden patterns in the financial time series and can precisely predict future actions in these markets.

Development and Performance Analysis of Fireworks Algorithm-Trained Artificial Neural Network (FWANN)

Financial time series are highly nonlinear and their movement is quite unpredictable. Artificial neural networks (ANN) have ample applications in financial forecasting. Performance of ANN models mainly depends upon its training. Though gradient descent-based methods are common for ANN training, they have several limitations. Fireworks algorithm (FWA) is a recently developed metaheuristic inspired from the phenomenon of fireworks explosion at night, which poses characteristics such as faster convergence, parallelism, and finding the global optima. This chapter intends to develop a hybrid model comprising FWA and ANN (FWANN) used to forecast closing prices series, exchange series, and crude oil prices time series. The appropriateness of FWANN is compared with models such as PSO-based ANN, GA-based ANN, DE-based ANN, and MLP model trained similarly. Four performance metrics, MAPE, NMSE, ARV, and R2, are considered as the barometer for evaluation. Performance analysis is carried out to show the suitability and superiority of FWANN.

A Review of the Fractal Market Hypothesis for Trading and Market Price Prediction

This paper provides a review of the Fractal Market Hypothesis (FMH) focusing on financial times series analysis. In order to put the FMH into a broader perspective, the Random Walk and Efficient Market Hypotheses are considered together with the basic principles of fractal geometry. After exploring the historical developments associated with different financial hypotheses, an overview of the basic mathematical modelling is provided. The principal goal of this paper is to consider the intrinsic scaling properties that are characteristic for each hypothesis. In regard to the FMH, it is explained why a financial time series can be taken to be characterised by a 1/t1−1/γ scaling law, where γ>0 is the Lévy index, which is able to quantify the likelihood of extreme changes in price differences occurring (or otherwise). In this context, the paper explores how the Lévy index, coupled with other metrics, such as the Lyapunov Exponent and the Volatility, can be combined to provide long-term forecasts. Using these forecasts as a quantification for risk assessment, short-term price predictions are considered using a machine learning approach to evolve a nonlinear formula that simulates price values. A short case study is presented which reports on the use of this approach to forecast Bitcoin exchange rate values.

Modeling and Forecasting Cryptocurrency Closing Prices with Rao Algorithm-Based Artificial Neural Networks: A Machine Learning Approach

Artificial neural networks (ANNs) are suitable procedures for predicting financial time series (FTS). Cryptocurrencies are good investment assets; therefore, the effective prediction of cryptocurrencies has become a trending area of research. Capturing inherent uncertainties associated with cryptocurrency FTS with conventional methods is difficult. Though ANNs are the better alternative, fixing the optimal parameters of ANNs is a tedious job. This article develops a hybrid ANN through Rao algorithm (RA + ANN) for the effective prediction of six popular cryptocurrencies such as Bitcoin, Litecoin, Ethereum, CMC 200, Tether, and Ripple. Six comparative models such as GA + ANN, PSO + ANN, MLP, SVM, LSE, and ARIMA are developed and trained in a similar way. All these models are evaluated through the mean absolute percentage of error (MAPE) and average relative variance (ARV) metrics. It is found that the proposed RA + ANN generated the lowest MAPE and ARV values, statistically different as compared with existing methods mentioned above, and hence can be recommended as a potential financial instrument for predicting cryptocurrencies.

Online Learning with Radial Basis Function Networks

We investigate the benefits of feature selection, nonlinear modelling and online learning when forecasting financial time series. We combine sequential updating with continual learning, specifically transfer learning. We perform feature representation transfer through clustering algorithms that determine the analytical structure of radial basis function networks we construct. These networks achieve lower mean-square prediction errors than kernel ridge regression models, which arbitrarily use all training vectors as prototypes. We also demonstrate quantitative procedures to determine the very structure of the networks. Finally, we conduct experiments on the log-returns of financial time series and show that these online transfer learning models outperform a random-walk baseline. In contrast, the offline learning models struggle to do so.