Abstract
In this chapter, the VaR of the MSCI emerging market index (MSCI-EMI) developed by Morgan Stanley Capital International (MSCI) is estimated using linear, nonlinear time series and ANN. In this context, the aim of the study is to estimate the VaR exceedance of the MSCI-EMI as a global financial risk indicator compared with traditional time series methods and ANN. In addition, the most effective method on this index is determined by statistical information criteria, and the comparative evaluation of the model selection criteria is carried out. The period of analysis is between December 1987-April 2020 with monthly frequency and VaR exceedance obtained with ARMA-GARCH, TGARCH, EGARCH, GJR, and ANN models. Confidence levels of models, VaR exceedance, and Kupeic statistics are obtained. VaR exceedances are examined through the superior model.
TopIntroduction
Emerging markets (EM) countries are in the process of rapid growth and development. However, the household incomes and capital markets of these countries are less mature than those of developed countries. Although EMs’ are characterized by rapid economic growth, their infrastructure and household incomes have failed to catch up with developed countries. The general characteristic of EMs’ is that they have a high level of economic development with rapid industrialization. The difference from developed and developing countries is that they are not only primarily based on agriculture, but are also active in infrastructure and industrial growth.
EM countries are growing faster than developed and developing countries. The main reason for this is that the domestic effective demand is lively and foreign direct investment is high. Therefore, they were not significantly affected by the 2008 mortgage crisis. In addition, EMs’ have integrated their national financial markets into the international financial network by adopting financial liberalization and de-regulation policies. In summary, EMs’ are very active both in the real and financial sectors. For this reason, they serve as the locomotive of the world economy.
EMs’, which have become an important market for investors, are markets where financial transactions are intensive and rapidly developing due to a specific risk. As experienced during the Mexican crisis of 1994, especially the rapid outflows of portfolio liabilities can lead EM countries into crisis in a short period of time. It can also be attractive to investors with high growth data. As a result of this situation, a financial crisis in EMs’ could spread throughout the world in a short time. Different models are available to predict whether the country is at risk. One of these is the concept of value at risk (VaR). VaR is a method developed for monetary expression of the expected maximum loss over a given time period and at a certain probability level. Different approaches exist in calculating VaR. In addition to econometric models, methods such as computer-based simulation are frequently used in literature to calculate VaR.
In this context, rapidly-developing information technologies and the speed of data transfer and analysis in financial markets are important in evaluating markets. Emerging information technologies also introduce new approaches to data storage and analysis. Artificial intelligence algorithms are also used in economic analysis. Artificial intelligence originated as a concept based on modelling the human system of thought. Artificial intelligence includes different modelling techniques such as expert systems, fuzzy logic, artificial neural networks, deep learning and decision trees. These techniques are used for classification, forecasting and clustering purposes and represent an approach to establishing models. In this context, artificial neural networks (ANN), which are used most frequently among artificial intelligence techniques, in addition to being used in fields such as engineering, biomedical, biostatistics, also contribute reliable analysis in the financial sector. ANN, in particular, are frequently used in financial analysis studies today.
In ANN, it functions as a nonlinear function due to modelling based on the human nervous system. In the ANN approach, the model is established using the training set and the information obtained from the training set is stored throughout the model. This established model is based on optimizing weights among neurons. With the help of the established model, information can be estimated for a new data. This approach can also be used for perception-oriented concepts. They are also used to complete any pattern. One of the major advantages of this approach is that it can work with incomplete information and has the ability to learn on its own.
Key Terms in this Chapter
Autoregressive Moving Average Technique (ARMA): It is a type of traditional linear time series modelling technique.
Adaptive Moment Estimation (ADAM): Adaptive moment estimation (adam) is a kind of optimization method. It is effective and uses little memory.
Autoregressive Conditional Heteroscedasticity (ARCH): It is a kind of statistical time series model. If the error variance in a time series acts as autoregressive moving average model (ARMA), the model is called as ARCH.
Recurrent Neural Network (RNN): It is a type of neural network which is used to model sequence or time series data. It has a recursive structure. And, the neural network use previous data to understand future data.
Exponential Autoregressive Conditional Heteroscedasticity (EGARCH): This is another form of GARCH.
Long-Short-Term Memory (LSTM) Network: It is a kind of recurrent neural network technique. LSTM aims to solve vanishing gradient problems.
Generalized Autoregressive Conditional Heteroscedasticity (GARCH): It is a kind of statistical time series model. If the error variance in a time series acts as autoregressive model (AR), the model is named as GARCH.