Nonlinear Extension of Asymmetric Garch Model within Neural Network Framework

Abstract

The importance of volatility for all market participants has led to the development and application of various econometric models. The most popular models in modelling volatility are GARCH type models because they can account excess kurtosis and asymmetric effects of financial time series. Since standard GARCH(1,1) model usually indicate high persistence in the conditional variance, the empirical researches turned to GJR-GARCH model and reveal its superiority in fitting the asymmetric heteroscedasticity in the data. In order to capture both asymmetry and nonlinearity in data, the goal of this paper is to develop a parsimonious NN model as an extension to GJR-GARCH model and to determine if GJR-GARCH-NN outperforms the GJR-GARCH model.

Neural network structure identification in inflation forecasting

“…In time series analysis, many studies have used NNs on simulated data (Gençay & Liu, 1997; Gupta et al, 2005; Hwarng, 2001) or traditionally used an M3 database such as airline passenger, Canadian lynx, or sunspot data (Balkin, 1997; Ghiassi, Saidane, & Zimbra, 2005; Ho et al, 2002; Mańdziuk & Mikołajczak, 2002; Medeiros, Teräsvirta, & Rech, 2006; Zhang, 2003; Zhang & Kline, 2007). In recent studies, NNs are used in predicting highly volatile and fluctuating financial variables that are difficult to predict using standard statistical and econometric methods or models such as foreign exchange rates (Chaudhuri & Ghosh, 2016; Kuan & Liu, 1995; Rech, 2002; Zhang, 2003), interest rates (Abid & Ben Salah, 2002; Aljinović & Poklepović, 2013), stock yields, and/or stock indices (Kim, Oh, Kim, & Do, 2004; Medeiros et al, 2006; Ortega, 2012; Rech, 2002; Wang et al, 2016; Zekić‐Sušac & Kliček, 2002) and volatility (Arnerić et al, 2014; Arnerić & Poklepović, 2016; Bektipratiwi & Irawan, 2011; Bildirici & Ersin, 2012, 2014; Mantri, Gahan, & Nayak, 2010; Mantri, Mohanty, & Nayak, 2012). Some studies have applied NNs to macroeconomic variables such as growth (Aminian et al, 2006; Gonzales, 2000; Qi, 2001; Rech, 2002; Teräsvirta et al, 2005; Tkacz, 2001), industrial production (Aminian et al, 2006;Rech, 2002 ; Teräsvirta et al, 2005), unemployment (Rech, 2002; Teräsvirta, 2008; Teräsvirta et al, 2005), monetary aggregates (Rech, 2002; Teräsvirta et al, 2005), and inflation (Al‐Maqaleh, Al‐Mansoub, & Al‐Badani, 2016; Binner et al, 2005, 2004, 2006, 2007; Choudhary & Haider,…”

“…Some studies have applied NNs to macroeconomic variables such as growth (Aminian et al, 2006; Gonzales, 2000; Qi, 2001; Rech, 2002; Teräsvirta et al, 2005; Tkacz, 2001), industrial production (Aminian et al, 2006;Rech, 2002 ; Teräsvirta et al, 2005), unemployment (Rech, 2002; Teräsvirta, 2008; Teräsvirta et al, 2005), monetary aggregates (Rech, 2002; Teräsvirta et al, 2005), and inflation (Al‐Maqaleh, Al‐Mansoub, & Al‐Badani, 2016; Binner et al, 2005, 2004, 2006, 2007; Choudhary & Haider, 2008; McNelis & McAdam, 2004; Moshiri & Cameron, 2000; Moshiri, Cameron, & Scuse, 1999; Rech, 2002; Teräsvirta et al, 2005). The application of NNs in macroeconomics is in its initial stages; there is little relevant research in this field of application, although most studies show better performances of NNs compared to other econometric models (Kuan & Liu, 1995; Balkin, 1997; Gonzales, 2000; Moshiri & Cameron, 2000; Tkacz, 2001; Abid & Ben Salah, 2002; Zhang, 2003; Binner et al, 2005; Aminian et al, 2006; Medeiros et al, 2006; Teräsvirta, 2008; Choudhary & Haider, 2008; Bektipratiwi & Irawan, 2011; Mantri et al, 2012; Bildirici & Ersin, 2012, 2014; Arnerić et al, 2014; Arnerić & Poklepović, 2016), and only a few studies point to the different performances of NNs in predicting in the long and short term. Short‐term forecasts using NNs are more accurate than those using other linear or nonlinear models (Binner et al, 2004; Binner et al, 2006; Ho et al, 2002; Medeiros et al, 2006; Moshiri et al, 1999; Moshiri & Cameron, 2000; Teräsvirta et al, …”

“…This knowledge can be confirmed in numerous empirical studies. In empirical research, it is common to use only 1 hidden neuron (Arnerić et al, 2014; Arnerić & Poklepović, 2016; Teräsvirta, 2008), up to 5 hidden neurons (Al‐Maqaleh et al, 2016; Balkin, 1997; Binner et al, 2005; Claveria & Torra, 2014; Gonzales, 2000; Gupta et al, 2005; Kim et al, 2004; McNelis & McAdam, 2004; Qi, 2001; Rech, 2002), up to 10 hidden neurons (Bildirici & Ersin, 2012; Binner et al, 2006; Gençay & Liu, 1997; Ghiassi et al, 2005; Ho et al, 2002; Kuan & Liu, 1995; Moshiri et al, 1999; Moshiri & Cameron, 2000; Zhang & Kline, 2007), and even over 10 hidden neurons (Aminian et al, 2006; Chaudhuri & Ghosh, 2016; Choudhary & Haider, 2008; Ortega, 2012; Wang et al, 2016). Some research does not even give information about the number of hidden neurons (Binner et al, 2007; Hwarng, 2001), while some studies use more hidden layers (1, 2, or 3) with up to 5 hidden neurons (Abid & Ben Salah, 2002).…”

A Stock Index Futures Price Prediction Approach Based on the MULTI-GARCH-LSTM Mixed Model

“…For the past few years, traditional econometric and statistical models such as Autoregressive Integrated Moving Average (ARIMA) [4] and GARCH [5,6] have been used to predict future prices. Lim revealed that symmetric and asymmetric GARCH models have different performances in different time frames [7].…”

Forecasting volatility with a stacked model based on a hybridized Artificial Neural Network

“…In addition to the above-mentioned researches, this type of hybrid models has been broadly used in other papers. Bildirici and Ersin (2014) proposed a MS-GARCH with an ANN to improve the forecasting accuracy, Bektipratiwi and Irawan (2011) combined a radial basis function with an EGARCH to model stocks returns of an Indonesian bank and Arneric and Poklepovic (2016) developed an ANN model as an extension of a GJR-GARCH to forecast the market returns of six European emerging markets. GARCH-based models have been also combined with ANNs to predict the volatility in commodity markets, such as gold (Kristjanpoller and Minutolo 2015) or oil (Kristjanpoller and Minutolo 2016).…”