Stock index prediction and uncertainty analysis using multi-scale nonlinear ensemble paradigm of optimal feature extraction, two-stage deep learning and Gaussian process regression

Wang, Jujie; He, Jinwei; Cheng, Feng; Liu, Feng; Li, Yang

doi:10.1016/j.asoc.2021.107898

Cited by 21 publications

(4 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Through experimental research on different types of deep reinforcement learning algorithms, Bahoo et al (2024) introduced a long short-term memory network to replace the traditional network structure of Deep Q-Network (DQN) and developed a quantitative trading system, which made the algorithm model serve users better. Wang and Liu (2024) put forward a model-free reinforcement-learning framework and introduced the reinforcement-learning model into long-term and short-term memory networks. When tested in a cryptocurrency market, it quadrupled income within 50 days, indicating that the variant of deep reinforcement learning has a powerful role in the field of quantitative financial transactions (Shi et al, 2024).…”

Section: Literature Reviewmentioning

confidence: 99%

Securities Quantitative Trading Strategy Based on Deep Learning of Industrial Internet of Things

Tang,

Wang,

Wang

2024

International Journal of Information Technology and Web Engineering

View full text Add to dashboard Cite

By combing the shortcomings of the current quantitative securities trading, a new deep reinforcement learning modeling method is proposed to improve the abstraction of state, action and reward function; on the basis of the traditional DQN algorithm, a deep reinforcement learning algorithm model of RB_DRL is proposed. By improving the network structure and connection mode, and redefining the loss function of the network, the improved model performs well in many groups of comparative experiments. A securities quantitative trading system based on deep reinforcement learning is designed, which organically combines models, strategies and data, visually displays the information to users in the form of web pages to facilitate users' use and seeks the trading rules of the financial market to provide investors with a more stable trading strategy. The research results have important practical value and research significance in the field of financial investment.

show abstract

Section: Literature Reviewmentioning

confidence: 99%

Securities Quantitative Trading Strategy Based on Deep Learning of Industrial Internet of Things

Tang,

Wang,

Wang

2024

International Journal of Information Technology and Web Engineering

View full text Add to dashboard Cite

show abstract

“…with: 𝑌 𝑖 = dependent variable 𝑚(𝑥 𝑖 ) = regression curve 𝜀 𝑖 = error In kernel regression, there are several estimators, but this study used the Nadaraya-Watson estimator with the following Equations (4) [20,21]: f. Compare the evaluation results of both kernel functions by calculating the coefficient of determination and MAPE values using Equations ( 5) and Equations (6).…”

Section: Data Processingmentioning

confidence: 99%

Comparing Gaussian and Epanechnikov Kernel of Nonparametric Regression in Forecasting Issi (Indonesia Sharia Stock Index)

Farida

Purwanti

Ulinnuha

2022

BAREKENG: J. Il. Mat. & Ter.

View full text Add to dashboard Cite

ISSI reflects the movement of sharia stock prices as a whole. It is necessary to forecast the share price to help investors determine whether the shares should be sold, bought, or retained. This study aims to predict the value of ISSI using nonparametric kernel regression. The kernel regression method is one of the nonparametric regression methods used to estimate conditional expectations using kernel functions. Kernel functions used in this study are gaussian and Epanechnikov kernel functions. The estimator used is the estimator Nadaraya-Watson. This study aims to compare the two kernel functions to predict the value of ISSI in the period from January 2016 to October 2019. The analysis results obtained the best method in predicting ISSI values, namely nonparametric kernel regression using Nadaraya-Watson estimator and Gaussian kernel function with the MAPE value of 15% and the coefficient of determination of 85%. Independent variables that significantly affect ISSI are interest rates, exchange rates, and inflation. Curve smoothing is done using bandwidth value (h) searched by the Silverman rule. The calculation result with the Silverman rule obtained a bandwidth value of 101832.7431.

show abstract

“…To predict three non-linear stock indexes -Dow Jones, S&P500, and NASDAQ, the authors have proposed an ensemble approach that employed Gaussian process regression, recurrent neural network (RNN), and LSTM [54]. In [55], for forecasting the S&P500 and CSI300, the authors have combined LSTM with complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) -a type of empirical mode decomposition technique.…”

Section: Literature Reviewmentioning

confidence: 99%

Forecasting non-linear macroeconomic indexes of India: an ensemble of MLP and Holt’s linear methods

2022

IJATEE

View full text Add to dashboard Cite

Electricity and electric-equipment play a critical role in modern living, and precise price forecasting for these items aids decision-makers in anticipating changes, planning, and budgeting ahead of time. This present research focused on the wholesale price indexes of items from the "manufacture of electrical equipment (MEEQ)" group of India's existing wholesale price index (WPI) series. This work proposed a novel, state-of-art ensemble forecast approach that used multilayer perceptron (MLP) and Holt's linear (HL) approaches for some specified WPIs from this group. The researchers selected the WPIs that manifest non-linearity as determined by the curve-fit technique. The paper applied a variance-based weighted average scheme to generate the ensemble forecast. The statistical rigor-based curve-fit aided in identifying that seventeen out of forty-eight indexes manifest non-linear fits. The proposed MLP-HL ensemble approach exhibited excellent forecast results (relative root mean squared error, i.e., RRMSE < 10%) for all seventeen WPIs. For each of these seventeen WPI's, the current work compared the proposed MLP-HL with nineteen models (eight statistical, four machine learning, and seven contemporary ensemble strategies). In the majority of the cases, it outran the nineteen models in terms of mean absolute error (MAE) and mean absolute percentage error (MAPE). As an alternative pathway for forecasting these seventeen non-linear WPIs, the present work suggests using this proposed MLP-HL approach.

show abstract

Stock index prediction and uncertainty analysis using multi-scale nonlinear ensemble paradigm of optimal feature extraction, two-stage deep learning and Gaussian process regression

Cited by 21 publications

References 43 publications

Securities Quantitative Trading Strategy Based on Deep Learning of Industrial Internet of Things

Securities Quantitative Trading Strategy Based on Deep Learning of Industrial Internet of Things

Comparing Gaussian and Epanechnikov Kernel of Nonparametric Regression in Forecasting Issi (Indonesia Sharia Stock Index)

Forecasting non-linear macroeconomic indexes of India: an ensemble of MLP and Holt’s linear methods

Contact Info

Product

Resources

About