GARCH based artificial neural networks in forecasting conditional variance of stock returns

Abstract

Abstract. Portfolio managers, option traders and market makers are all interested in volatility forecasting in order to get higher profits or less risky positions. Based on the fact that volatility is time varying in high frequency data and that periods of high volatility tend to cluster, the most popular models in modelling volatility are GARCH type models because they can account excess kurtosis and asymmetric effects of financial time series. A standard GARCH(1,1) model usually indicates high persistence in the conditional variance, which may originate from structural changes. The first objective of this paper is to develop a parsimonious neural networks (NN) model, which can capture the nonlinear relationship between past return innovations and conditional variance. Therefore, the goal is to develop a neural network with an appropriate recurrent connection in the context of nonlinear ARMA models, i.e., the Jordan neural network (JNN). The second objective of this paper is to determine if JNN outperforms the standard GARCH model. Out-of-sample forecasts of the JNN and the GARCH model will be compared to determine their predictive accuracy. The data set consists of returns of the CROBEX index daily closing prices obtained from the Zagreb Stock Exchange. The results indicate that the selected JNN(1,1,1) model has superior performances compared to the standard GARCH(1,1) model. The contribution of this paper can be seen in determining the appropriate NN that is comparable to the standard GARCH(1,1) model and its application in forecasting conditional variance of stock returns. Moreover, from the econometric perspective, NN models are used as a semiparametric method that combines flexibility of nonparametric methods and the interpretability of parameters of parametric methods.

Jordan neural network for inflation forecasting

“…Here, it is addressed by careful weight initialization, where recurrent weights are predefined and not learned. Research on economic and financial time series [3,6,10,12,37] gives preference to RNNs for predicting time series versus linear, nonlinear models and/or FNNs. [34] give preference to RNNs when predicting in the long term, while predicting in the short term FNNs and linear models give better results.…”

“…they vary from 1 to 5, defined from the initial model. The weight of the context unit λ, representing the long-term memory, is varied with values 0.1, 0.3, 0.5, 0.7 and 0.9 according to [3]. It is not learned or trained but determined in advance to deal with vanishing gradient problem.…”

Fitting Conventional Neural Network Time Series Models on Sand Price Indices Dataset

“…For examination purposes, this exploration utilized information slacks from 5 to 40 and concealed hubs from 5 to 45 for BPNN-NAR show, while for the BPNN-NARMA display, this exploration utilized the mistake slacks going from 5 to 40, together with input slacks from 5 to 40 and concealed hubs from 5 to 45. The most widely recognized way to deal with choose this property is experimentation or experimentation [14,15,16], though extraordinary systems (dependable guideline) and calculations (pruning and developing) are also available [17,18,19]. It is to be seen that BPNN-NAR is a feedforward neural system write demonstrate, while BPNN-NARMA is an intermittent neural system compose display [18].…”

“…The most widely recognized way to deal with choose this property is experimentation or experimentation [14,15,16], though extraordinary systems (dependable guideline) and calculations (pruning and developing) are also available [17,18,19]. It is to be seen that BPNN-NAR is a feedforward neural system write demonstrate, while BPNN-NARMA is an intermittent neural system compose display [18]. This investigation reports the blunder measures on the test dataset, which is the most key trademark, mirroring ANN's speculation capacity [19,20].…”

Advantage of make-to-stock strategy based on linear mixed-effect model: a comparison with regression, autoregressive, times series, and exponential smoothing models

“…Comparing forecasting methods is crucial in methodological studies [41,34,13,19,4]. The model proposed in this study was compared with commonly used statistical forecasting methods, and we incorporate the two main covariates, time and product types, into the following models.…”