Purpose

As an indicator of volatility in stock prices, the identification of model volatility is important for predicting movements in financial markets and making successful investment decisions. This study determined autoregressive conditional heteroskedasticity models to explain the volatility structure of nine deposit banks in the BIST Bank Index.

Data

Within the scope of the study, models were estimated using the stock index return values of Akbank, Denizbank, Garanti Bank, Halkbank, İşbank, QNB Finansbank, Şekerbank, Vakıfbank, and Yapı Kredi Bank. For this purpose, logarithmic returns were calculated by using daily closing prices between 02.01.2008 and 07.06.2018.

Methodology

The model that best describes the volatility in the stock return values of the banks was estimated using single and multi-regime models. The symmetrical generalized autoregressive conditional heteroskedasticity (GARCH) model was developed by Bollerslev (1986), and the asymmetric exponential generalized autoregressive conditional heteroskedasticity (EGARCH) model was introduced by Nelson (1991). Markov switching conditional variance models, which add different regime states to the model, can be formed in both symmetric and asymmetric structures. This study used the MSGARCH module developed by Ardia, Bluteau, Boudt, Catania, Peterson, and Trottier (2018). During the estimation, both single- and multiregime models were estimated using the maximum likelihood estimation method considering the different distribution characteristics specific to financial time series.

Results

The analysis revealed that the daily average return of the nine banks in the BIST Bank Index is very close to zero, and standard deviations are between 2,277182 and 3,252496. İşbank had the lowest standard deviation, and Denizbank had the highest standard deviation. Before modeling the volatility in the return series using GARCH models, an ARCH test was used to determine whether the series had heteroskedasticity. For this purpose, mean equations were estimated using an autoregressive moving average model structure, and the ARCH test was applied. The null hypothesis suggesting that there is no ARCH effect in the return series was rejected at a 1% significance level based on the ARCH-LM test results. In this case, the volatility in the series can be examined with conditional variance models. Parameter and information criteria values were also used in the estimation of the model.

Conclusion

This study determined the models that best explain volatility. These models include the ged distributed MSEGARCH model for Akbank; the skewed ged distributed MSGARCH model for Denizbank; the t distributed EGARCH (1,1,1) model for Garanti bank; the t distributed GARCH (1,1) model for Halkbank; the ged distributed MSEGARCH model for İşbank; the skewed ged distributed MSGARCH model for QNB Finansbank; the normal distributed MSEGARCH model for Yapı Kredi Bank; the ged distributed MSEGARCH model for Şekerbank; and the t distributed EGARCH (1,1,1) model for Vakıfbank.

This analysis determined that the return series of banks other than Halkbank have an asymmetric effect or regime switching. When the predicted models are examined, the series generally implies a regime switching effect. Therefore, if different regime features are ignored during estimation, the estimation results may be biased. As a result of the analysis, it has been determined that the return series of the banks have similar characteristics, although they do not have exactly the same model structures. It can be said that these different results in the modeling between banks are due to the policies implemented by the banks or their structural characteristics.