scholarly journals A GARCH Model with Artificial Neural Networks

Information ◽  
2020 ◽  
Vol 11 (10) ◽  
pp. 489
Author(s):  
Wing Ki Liu ◽  
Mike K. P. So
Keyword(s):  
Neural Network ◽  
Short Term Memory ◽  
Garch Model ◽  
Network Models ◽  
Variable Importance ◽  
Neural Network Models ◽  
Out Of Sample ◽  
Importance Analysis ◽  
Artificial Neural ◽  
T Distribution

In this paper, we incorporate a GARCH model into an artificial neural network (ANN) for financial volatility modeling and estimate the parameters in Tensorflow. Our goal was to better predict stock volatility. We evaluate the performance of the models using the mean absolute errors of powers of the out-of-sample returns between 2 March 2018 and 28 February 2020. Our results show that our modeling procedure with an ANN can outperform the standard GARCH(1,1) model with standardized Student’s t distribution. Our variable importance analysis shows that Net Debt/EBITA is among the six most important predictor variables in all of the neural network models we have examined. The main contribution of this paper is that we propose a Long Short-Term Memory (LSTM) model with a GARCH framework because LSTM can systematically take into consideration potential nonlinearity in volatility structure at different time points. One of the advantages of our research is that the proposed models are easy to implement because our proposed models can be run in Tensorflow, a Python package that enables fast and automatic optimization. Another advantage is that the proposed models enable variable importance analysis.

scholarly journals Combining Housing Price Forecasts Generated Separately by Hedonic and Artificial Neural Network Models

2021 ◽  
pp. 130-148
Author(s):  
Salvatore Joseph Terregrossa ◽  
Mohammed Hussein Ibadi
Keyword(s):  
Neural Network ◽  
Artificial Neural Network ◽  
Forecast Error ◽  
Weighted Least Squares ◽  
Network Models ◽  
Housing Unit ◽  
Neural Network Models ◽  
Data Set ◽  
Out Of Sample ◽  
Artificial Neural

Aims: A) To enhance accuracy in forecasting housing unit prices by forming combinations of component forecasts generated separately by hedonic and artificial neural network models; B) To help ascertain whether a constrained or unconstrained linear combining model achieves superior forecasting performance. Place and Duration of the Study: Department of Business Administration, Istanbul Aydin University, Istanbul 34295, Turkey; from 2019 to 2020. Study Design: A cross sectional data set of housing unit prices and corresponding housing unit attributes and characteristics is formed and then randomly divided into two segments: in sample (80%) and out of sample (20%). Three different methods (hedonic, artificial neural network and combining) are then employed to process the same in sample data set, and generate out of sample forecasts. The three forecasting methods are then tested and compared. Methodology: Out of sample combination forecasts are formed with component forecast weights generated by in sample weighted least squares (WLS) regression of realized price against in sample component forecasts. Four types of regressions are run: unconstrained, with and without a constant; constrained, with and without a constant. Then the mean absolute forecast error of each forecasting method is calculated and the mean difference in absolute forecast error between all pairs of models are compared and tested with a nonparametric Wilcoxon sign rank test. Results: The combining model formed with component forecast weights generated by weighted least squares (WLS) regression with the constant term suppressed and the sum-of-the-coefficients constrained to equal one, generally performs the best, in comparison with all other forecasting models (component and combination) examined in the study. Conclusion: The findings represent further evidence regarding the benefits of applying constraints on the linear combining forecast model; and demonstrate that a constrained linear combining model can be a successful technique for enhancing the forecast accuracy of housing unit prices.

scholarly journals Prediction of Stress in Power Transformer Winding Conductors Using Artificial Neural Networks: Hyperparameter Analysis

Energies ◽  
2021 ◽  
Vol 14 (14) ◽  
pp. 4242
Author(s):  
Fausto Valencia ◽  
Hugo Arcos ◽  
Franklin Quilumba
Keyword(s):  
Neural Network ◽  
Artificial Neural Network ◽  
Finite Element ◽  
Power Transformer ◽  
Network Models ◽  
Activation Function ◽  
Neural Network Models ◽  
Transformer Winding ◽  
Artificial Neural ◽  
Femm Software

The purpose of this research is the evaluation of artificial neural network models in the prediction of stresses in a 400 MVA power transformer winding conductor caused by the circulation of fault currents. The models were compared considering the training, validation, and test data errors’ behavior. Different combinations of hyperparameters were analyzed based on the variation of architectures, optimizers, and activation functions. The data for the process was created from finite element simulations performed in the FEMM software. The design of the Artificial Neural Network was performed using the Keras framework. As a result, a model with one hidden layer was the best suited architecture for the problem at hand, with the optimizer Adam and the activation function ReLU. The final Artificial Neural Network model predictions were compared with the Finite Element Method results, showing good agreement but with a much shorter solution time.

scholarly journals PSIII-5 Comparison of Bivariate Machine Learning and Linear Model for Genomic Prediction with Different Heritability, QTL and SNP Panel Scenarios

2020 ◽  
Vol 98 (Supplement_4) ◽  
pp. 230-231
Author(s):  
Sunday O Peters ◽  
Mahmut Sinecan ◽  
Kadir Kizilkaya ◽  
Milt Thomas
Keyword(s):  
Neural Network ◽  
Artificial Neural Network ◽  
Normal Distribution ◽  
Genomic Prediction ◽  
Network Models ◽  
Bivariate Normal Distribution ◽  
Snp Markers ◽  
Neural Network Models ◽  
Bivariate Normal ◽  
Artificial Neural

Abstract This simulation study used actual SNP genotypes on the first chromosome of Brangus beef cattle to simulate 0.50 genetically correlated two traits with heritabilities of 0.25 and 0.50 determined either by 50, 100, 250 or 500 QTL and then aimed to compare the accuracies of genomic prediction from bivariate linear and artificial neural network with 1 to 10 neurons models based on G genomic relationship matrix. QTL effects of 50, 100, 250 and 500 SNPs from the 3361 SNPs of 719 animals were sampled from a bivariate normal distribution. In each QTL scenario, the breeding values (Σgijβj) of animal i for two traits were generated by using genotype (gij) of animal i at QTL j and the effects (βj) of QTL j from a bivariate normal distribution. Phenotypic values of animal i for traits were generated by adding residuals from a bivariate normal distribution to the breeding values of animal i. Genomic predictions for traits were carried out by bivariate Feed Forward MultiLayer Perceptron ANN-1–10 neurons and linear (GBLUP) models. Three sets of SNP panels were used for genomic prediction: only QTL genotypes (Panel1), all SNP markers, including the QTL (Panel2), and all SNP markers, excluding the QTL (Panel3). Correlations from 10-fold cross validation for traits were used to assess predictive ability of bivariate linear (GBLUP) and artificial neural network models based on 4 QTL scenarios with 3 Panels of SNP panels. Table 1 shows that the trait with high heritability (0.50) resulted in higher correlation than the trait with low heritability (0.25) in bivariate linear (GBLUP) and artificial neural network models. However, bivariate linear (GBLUP) model produced higher correlation than bivariate neural network. Panel1 performed the best correlations for all QTL scenarios, then Panel2 including QTL and SNP markers resulted in better prediction than Panel3.

Automatic Electrocardiographic Analysis Using Artificial Neural Network Models

2011 ◽  
Vol 403-408 ◽  
pp. 3587-3593
Author(s):  
T.V.K. Hanumantha Rao ◽  
Saurabh Mishra ◽  
Sudhir Kumar Singh
Keyword(s):  
Neural Network ◽  
Artificial Neural Network ◽  
Radial Basis Function ◽  
Basis Function ◽  
Network Models ◽  
Neural Network Models ◽  
Radial Basis ◽  
Artificial Neural ◽  
Artificial Neural Network Models ◽  
Self Organizing

In this paper, the artificial neural network method was used for Electrocardiogram (ECG) pattern analysis. The analysis of the ECG can benefit from the wide availability of computing technology as far as features and performances as well. This paper presents some results achieved by carrying out the classification tasks by integrating the most common features of ECG analysis. Four types of ECG patterns were chosen from the MIT-BIH database to be recognized, including normal sinus rhythm, long term atrial fibrillation, sudden cardiac death and congestive heart failure. The R-R interval features were performed as the characteristic representation of the original ECG signals to be fed into the neural network models. Two types of artificial neural network models, SOM (Self- Organizing maps) and RBF (Radial Basis Function) networks were separately trained and tested for ECG pattern recognition and experimental results of the different models have been compared. The trade-off between the time consuming training of artificial neural networks and their performance is also explored. The Radial Basis Function network exhibited the best performance and reached an overall accuracy of 93% and the Kohonen Self- Organizing map network reached an overall accuracy of 87.5%.

Sign in / Sign up

Export Citation Format

Share Document