Comparing Regime Switching GARCH Models and GARCH Models
Transcripción
Comparing Regime Switching GARCH Models and GARCH Models
60 ANÁLISIS FINANCIERO Minoo Nazifi Naeini* y Shahram Fatahi** Comparing Regime Switching GARCH Models and GARCH Models in Developing Countries (Case study of IRAN) La comparación de los modelos GARCH y GARCH de cambio de régimen en los países en desarrollo (estudio del caso de Irán) ABSTRACT This study is for comparing GARCH models and Markov switching GARCH models in their ability to estimate and forecasting the volatility of Tehran stock market in some horizon of forecasting. This paper provides an analysis of regime switching in volatility and out-of-sample forecasting of the IRAN using daily data for the period 1995-2011. We first model volatility regime switching within a univariate Markov-Switching framework. Then we provide outof-sample forecasts of the TEHRAN daily returns using two competing non-linear models, the GARCH Markov Switching model and the uniregime GARCH Model. The comparison of the out-of-sample forecasts is done on the basis of forecast accuracy, using the statistical loss function. . The results, also, shows that SW-GARCH models can remove the high persistence of GARCH models and separately in each regime of volatility, the persistence are high. This shows the priority of SW-GARCH models. Another implication is that there is evidence of regime clustering. JEL classification numbers: C32, C11, C22, C52, C32, C53 Keywords: Volatility, Markov Regime Switching GARCH models¡ forecasting, out of sample, bootstrap RESUMEN Este estudio busca comparar la capacidad de estimación y previsión de la volatilidad del mercado de valores de Teherán para un horizonte de predicción dado, de los modelos GARCH y GARCH con cambio de régimen de Markov. Este trabajo ofrece un análisis de la variación del régimen en la volatilidad y de la previsión fuera de muestra de Irán a partir de datos diarios para el período 1995-2011. Primero modelamos el cambio de régimen de la volatilidad en un escenario univariante de cambio de Markov. A continuación ofrecemos previsiones fuera de la muestra de los rendimientos diarios de Teherán por medio de dos modelos no lineales, el modelo GARCH de cambio de régimen de Markov y el modelo GARCH de un solo régimen. La comparación de las previsiones fuera de muestra se realiza sobre la base de la precisión de los pronósticos, utilizando la función de pérdida estadística. Los resultados, además, muestran que los modelos SW-GARCH pueden eliminar la alta persistencia de los modelos GARCH y separadamente en cada régimen de volatilidad, las persistencia son altas. Esto demuestra la prioridad de los modelos SW-GARCH. Otra implicación es que hay evidencias de una agrupación de régimen. Clasificación JEL: C32, C11, C22, C52, C32, C53 Palabras clave: Volatilidad, Modelos GARCH de cambio de régimen de Markov, previsión, fuera de muestra, bootstrap Recibido: 18 de enero de 2012 Aceptado: 8 de junio de 2012 * Minoo Nazifi Naeini. MA in economics. ** Shahram Fatahi .Assistant prof at Razi University. Minoo Nazifi Naeini y Shahram Fatahi: Comparing Regime Switching GARCH Models and GARCH Models in Developing Countries (Case study of IRAN) La comparación de los modelos GARCH y GARCH de cambio de régimen en los países en desarrollo (estudio del caso de Irán) Análisis Financiero, n.º 119. 2012. Págs.: 60-68 COMPARING REGIME SWITCHING GARCH MODELS AND GARCH MODELS... 61 1. INTRODUCTION 2. RECENT STUDIES The volatility of financial markets has been the object of numerous developments and applications over the past two decades,. In this respect, the most widely used class of models is certainly that of GARCH .These models usually indicate a high persistence of the conditional variance (i.e. a nearly integrated GARCH process). Hence the estimates of a GARCH model may suffer from a substantial upward bias in the persistence parameter. Therefore, models in which the parameters are allowed to change over time may be more appropriate for modeling volatility. Last 20 years: lot of research on modeling volatility in financial markets using GARCH type models. Schwert (1989): model where returns switch between high or low variance states according to a two state Markov process. Hamilton and Susmel (1994) and Cai (1994): ARCH model with regime switching parameters. Gray (1994): proposes a class of regime-switching GARCH (RS-GARCH) models with time-varying probability, but estimates an approximation to the model. The quality of the approximation is not known. See also Dueker (1996), Bollen et al. (2000), Klaassen (2002), and Haas et al. (2004). Evidence of regime switching in the volatility of stock returns have been found by Hamilton and Susmel (1994), Hamilton and Lin (1996), Edwards and Susmel (2001), Coe (2002) and Kanas (2005). In addition, Engel (1994), Engle and Kim (2001) Kouretas (2003) are examples of studies that they have estimated Markov-Switching regime models for currency markets.. Gray (1994) presents a tractable Markov-switching GARCH model and a modification of his model is suggested by Klaassen (2002); see also Bollen, Gray, and Whaley (2000), Dueker (1996), Haas, Mittnik, and Paolella (2004), and Marcucci (2005) for related papers. Schwert (1989) consider a model in which returns can have a high or low variance, and switches between these states are determined by a two state Markov process. Hamilton and Susmel (1994) and Cai (1994) introduce an ARCH model with Markov-switching parameters in order to take into account sudden changes in the level of the conditional variance. This paper addresses three important issues with respect to the behavior of stock returns volatility of a recently established emerging of stock market the TEHRAN Stock Exchange using daily data for the period 1995-2011. The main findings of the paper are summarized as follows. First, there is strong evidence in favor of volatility regime switching modeling of nonlinearities which affects the stock returns. Second, the estimation of the MRS –GARCH model accurately describes the two regimes based on the different pattern of adjustment of the stock returns volatility; and the estimated model captures all the events that are responsible for the presence of nonlinear features in the TEHRAN stock returns during the period 1995-2011. Third, there is a high probability for regime clustering with the likelihood that a low volatility regime will be followed by a low volatility regime greater that the likelihood a high volatility regime will be followed by a high volatility regime. Finally, the comparison of the out-of-sample short-run forecasts generated by the two models suggest that on the basis of the forecast accuracy criteria, we conclude that the MRSGARCH and have priority to univariate GARCH models. The plan of this paper is as follows. First, there is some introduction and the studies about these issues in section 2 .we explore the issue of GARCH models in section 3, section 4 is for a description for MRS GARCH models. Then in section 5 the stock market data and methodology are illustrated. then we compare the forecasting performance of two competing non-linear models the Markov Regime Switching GARCH (MRS) model and the GARCH models in sectiion5 and finally in section 6 the discussion and conclusion are sketched. 3. DATA AND METHODOLOGY 3.1. Data The data set used in this study is the daily closing prices of value-weighted TEPIX index over the period 29/09/1995 through 03/04/2011. The data set is obtained from the web site of the Central Bank of Republic of Iran (www.irbours. com). The procedures are computed numerically by using MATLAB optimization routines. The data is divided into a ten year insample estimation period (2480 observations) and a subsequent one year out-of-sample forecasting period (400observations) Daily observations are converted into continuously compounded returns in a standard method as log differences: Minoo Nazifi Naeini y Shahram Fatahi: Comparing Regime Switching GARCH Models and GARCH Models in Developing Countries (Case study of IRAN) La comparación de los modelos GARCH y GARCH de cambio de régimen en los países en desarrollo (estudio del caso de Irán) Análisis Financiero, n.º 119. 2012. Págs.: 60-68 62 ANÁLISIS FINANCIERO The plot of return and price series are given in Figure 1 and Figure 2. TEPIX return index displays usual properties of financial data series. 3.2. Empirical results 3.2.1. GARCH Descriptive statistics of return series are represented in Table 1 As table shows, the index has a positive average return 0.156%. Daily standard deviation is 5.05%. The series also displays a negative skewness of 0.504 and an excess kurtosis of 22.48. These values indicate that the returns are not normally distributed, namely it has fatter tails. Also, Jarque-Bera test 2 statistic of 0.25 confirms the –normality of TEPIX returns. These findings are consistent with other financial returns’ properties. Table 2 present estimation results for uniregime GARCH models. It is clear from the table that almost all parameter estimates including µ in uniregime GARCH models are highly significant at 1%. Only the leverage effect ξ of EGARCH model with normal and GED errors are insignificant. However, the asymmetry effect term ξ in GJR-GARCH models is significantly different from zero, which indicates unexpected negative returns imply higher conditional variance as compared to same size positive returns. The degree of volatility persistence for GARCH models can be obtained by summing ARCH and GARCH parameters estimates (α 1 + ξ 1). For EGARCH (1, 1) and GJR-GARCH (1, 1), persistence is equal to β 1 and (α 1+ξ )/2 + β 1 respectively.. All models display strong persistence in volatility ranging from 0.980 to 0.987, that is, volatility is likely to remain high over several future periods once it increases. Minoo Nazifi Naeini y Shahram Fatahi: Comparing Regime Switching GARCH Models and GARCH Models in Developing Countries (Case study of IRAN) La comparación de los modelos GARCH y GARCH de cambio de régimen en los países en desarrollo (estudio del caso de Irán) Análisis Financiero, n.º 119. 2012. Págs.: 60-68 COMPARING REGIME SWITCHING GARCH MODELS AND GARCH MODELS... Conditional mean is rt=ì+ut and conditional variance is and and respectively for GARCH, EGARCH, GJR. 63 If distribution assumptions for standardized errors are compared, it reveals that normality assumption is highly outperformed by other two fat-tailed distributions in terms of loglikelihood values. It is an anticipated result because of fat tails property of Ian Stock Market. Overall, student-t distribution yields an improvement in fitting the data over the others and the GJR GARCH model with student-t has the largest log-likelihood among uniregime GARCH models. If a Minoo Nazifi Naeini y Shahram Fatahi: Comparing Regime Switching GARCH Models and GARCH Models in Developing Countries (Case study of IRAN) La comparación de los modelos GARCH y GARCH de cambio de régimen en los países en desarrollo (estudio del caso de Irán) Análisis Financiero, n.º 119. 2012. Págs.: 60-68 64 ANÁLISIS FINANCIERO GARCH model is successful at capturing volatility clustering, squared standardized residuals should have no autocorrelation. 3.2.2. MRS-GARCH Estimation results and summary statistics of SW-GARCH models are presented in Table 3 Almost all parameter esti- mates are significantly different from zero at least 95% confidence level. The conditional mean estimates in high volatility regime of SW-GARCH with normal and GED distributions are barely significant at 90% confidence level. However, ARCH parameters α 1 in both volatility regimes of SW-GARCH with normal distribution are insignificant. In order to see existence of different volatility regimes, we compute the un-conditional variances in each volatility regime. The long term volatility level depends on the estimates Minoo Nazifi Naeini y Shahram Fatahi: Comparing Regime Switching GARCH Models and GARCH Models in Developing Countries (Case study of IRAN) La comparación de los modelos GARCH y GARCH de cambio de régimen en los países en desarrollo (estudio del caso de Irán) Análisis Financiero, n.º 119. 2012. Págs.: 60-68 COMPARING REGIME SWITCHING GARCH MODELS AND GARCH MODELS... of constant parameter á0. Results in Table 3 are consistent with this argument and display that there are huge differences between α 0 estimates of each volatility regime. The parameter estimates á0 in high volatility regimes are nearly eight times greater than parameter estimates α 0 in low volatility regimes. Moreover, short run dynamics of volatility is determined by the ARCH parameter α 1 and GARCH parameter α 1. Large estimates of α 1 suggest that effect of shocks to future volatility die out in a long time, so volatility is persistent. Large values of α 1 display reaction of volatility to the recent price changes. 65 As expected conditional mean returns in low volatility regime are higher than that of high volatility regimes for all SWGARCH models. So, the lower uncertainty in TEPIX index gives chance of higher profit to the practitioners. This shows the importance of regimes switching models to model volatility. Comparing the low and high volatility regimes in all SWGARCH models, the former volatility regimes have lower α 1 estimates but higher β1 estimates than latter volatility regimes have. So, the GARCH processes in the low volatility regimes are more reactive but less persistent than that in the high volatility regime. In addition, it is interesting to notice that degree of volatility persistence (α 1 + β1) within low volatility regime is higher compared to the high volatility regime. Persistence within each regime is calculated as α i1 + β I where i=1, 2. 3.3. In samples In addition to the goodness-of-fit statistics, we consider various statistical loss functions to analyze in-sample estimation performance of the volatility models. We assume daily squared return as actual volatility. As seen Table 4, one of the SW-GARCH models obtains the highest ranking according to all statistical loss functions. Also, first three ranks are generally shared by SW-GARCH models. Thus, evaluating in sample estimation results according to loss functions, as well as goodness-of-fit statistics, we conclude that SWGARCH models perform better than uniregime GARCH models in describing Iran Stock Market volatility. Lastly, as seen in third column of Table 4, comparing persistence of uniregime GARCH models and SW-GARCH models, it is observed that the high persistence in the former specification is reduced by latter models. This result is consistent with Lamoureux and Lastrapes’s (1990) finding that is high persistent in volatility of GARCH is caused by regime shifts in the volatility process. Among all SW-GARCH models, SWGARCH with student-t2 shows the largest decline in volatility persistence. 4. FORECASTING EVALUATION In this section, we investigate ability of markov regime switching and uniregime GARCH models to forecast Iran Stock Market volatility at different forecast horizons. The forecast horizons 1, 5, 10 and 22 days are considered in this thesis. In Table 5, we present the forecast error statistics for one day ahead. The six of seven forecast error statistics suggest that SW-GARCH models provide the most accurate volatility forecasts. In terms of MSE, MAPE, R2LOG, MAE1 and MAE2, the best forecasting performance belongs to the SWGARCH model with t2; As well as short forecast horizons, we consider forecasting performance of various GARCH models at longer horizon 22 days (one month). Results are presented in Table 5 and 6. The rankings for one month horizon are quietly similar to that of the one day horizon. According to all loss functions except HMSE, SW-GARCH model with t2 is the best model in forecasting volatility while SW- GARCH model with student-t ranks second. Following markov regime switching models, standard uni-regime Minoo Nazifi Naeini y Shahram Fatahi: Comparing Regime Switching GARCH Models and GARCH Models in Developing Countries (Case study of IRAN) La comparación de los modelos GARCH y GARCH de cambio de régimen en los países en desarrollo (estudio del caso de Irán) Análisis Financiero, n.º 119. 2012. Págs.: 60-68 66 ANÁLISIS FINANCIERO Minoo Nazifi Naeini y Shahram Fatahi: Comparing Regime Switching GARCH Models and GARCH Models in Developing Countries (Case study of IRAN) La comparación de los modelos GARCH y GARCH de cambio de régimen en los países en desarrollo (estudio del caso de Irán) Análisis Financiero, n.º 119. 2012. Págs.: 60-68 COMPARING REGIME SWITCHING GARCH MODELS AND GARCH MODELS... 67 GARH models are ranked as fourth, fifth and sixth. On the other hand, HMSE suggests that top three volatility forecasters are standard uniregime GARCH models. It is important to note that there is substantial deference between results of HMSE and other statistical loss functions if model comparisons are considered. Most of time, result of HMSE are completely opposite to that of others. Marcucci (2005) has confronted with similar results and stated that HMSE loss is not particularly suitable for evaluating deferent volatility forecasts and it should be expected to give weird results. sis. First, there is strong evidence in favor of regime switching GARCH modeling of nonlinearities in the stock returns volatility of the TEHRAN. Second, the estimation of the MRS GARCH accurately describes the two regimes based on the different pattern of adjustment of the stock returns volatility; and the estimated model captures all the events that are responsible for the presence of nonlinear features in the TEHRAN stock returns. Third, there is a high probability for regime clustering with the likelihood that a low volatility regime will be followed by a low volatility regime greater that the 5. CONCLUSIONS Likelihood a high volatility regime will be followed by a high volatility regime. Fourth, we consider two competing non-linear models to conduct forecasting of the stock returns volatility. These models are the estimated MRS GARCH models and univariate GARCH model. By obtaining the 1step 5- step,10- step- 22- step ahead forecast for stock returns volatility for the out-of-sample period, we compare the outof-sample performance of the two competing models on the The purpose of this paper is twofold. First, it aims to model volatility regime switching GARCH for the stock returns volatility of TEHRAN Stock Exchange using daily data for the period 1995-2011. To conduct our study, we adopt the univariate GARCH developed by marcucci(2005). There are several important findings that stem from the present analy- Minoo Nazifi Naeini y Shahram Fatahi: Comparing Regime Switching GARCH Models and GARCH Models in Developing Countries (Case study of IRAN) La comparación de los modelos GARCH y GARCH de cambio de régimen en los países en desarrollo (estudio del caso de Irán) Análisis Financiero, n.º 119. 2012. Págs.: 60-68 68 ANÁLISIS FINANCIERO basis of forecasting accuracy by applying 7 statistical loss function. The results suggest that on the basis of the forecast accuracy and the, we could conclude that MRS-GARCH model have priority in univariate GARCH models in their forecasting performance. ACKNOWLEDGEMENTS We thanks Marcucci for providing his MATLAB source codes which estimate SWGARCH models’ parameters and forecast volatility. REFERENCES [1] A. Kanas, & G. Kouretas. «A cointegration approach to the lead-lag effect among size-sorted equity portfolios,» International Review of Economics & Finance, Elsevier, vol. 14(2), pages 181-201.( 2005). [2] Bollen, Nicolas P.B., Gray, Stephen F., Whaley, Robert E., Regime switching in foreign exchange rates: Evidence from currency option prices, Journal of Econometrics,( 2000), 94, 239-276. [3] C. Brunetti , C. Scotti, Mariano, S. Roberto & Tan, H.H. Augustine, Markov switching GARCH models of currency turmoil in Southeast Asia, Emerging Markets Review, Elsevier, vol. 9(2), June. (2008), 104-128. [4] C. J. Kim and C. R. Nelson , State-space models with regime switching Classical and Gibbs-sampling approaches with applications, MIT Press, Cambridge MA and London,1999. [5] Chao, Hui-Ping, 1998. «Regime Switching In Us Livestock Cycles, 1998 Annual meeting, , American Agricultural Economics Association,(1998). [6] Charles Engel & James D. Hamilton, «Long Swings in the Exchange Rate: Are they in the Data and Do Markets Know It?,» NBER Working Papers 3165, National Bureau of Economic Research, Inc.(1989). [7] [8] [9] Coe, P. “Power Issues When Testing the Markov Switching Model with the Sup. Likelihood Ratio Test Using U.S. Output,” Empirical Economics, 27,(2002). Engel, Charles & Kim, Chang-Jin. «The Long-Run U.S./U.K. Real Exchange Rate,» Journal of Money, Credit and Banking, Blackwell Publishing, vol. 31(3), pages 335-56, August, (1999). F. Klaassen, Improving GARCH volatility forecasts], Empirical Economics, 27(2), (2002), 363-394. [10] G. Schwarz, Estimating the dimension of a model, The Annals of Statistics,6(2) ,(1989),461-464. [11] G. Lamoureux & W. Lastrapes. « Heteroscedasticity in Stock Return Data: Volume versus GARCH Effects,» Journal of Finance, American Finance Association, vol. 45(1), pages 221-29, March, (1990). [12] H. Daouk and J. Q. Guo, Switching asymmetric GARCH and options on a volatility index, Journal of Futures Markets, 24(3), (2004), 251-282. [13] J. Cai, A markov model of unconditional variance in ARCH, Journal of Business and Economic Statistics, 12(3), (1994), 309-316. [14] J. D. Hamilton, , A new approach to the economic analysis of nonstationary time series and the business cycle, Econometrica, 57, (1989), 357.384. [15] J. D. Hamilton, Time Series Analysis, Princeton University Press, Princeton, New Jersey, 1994. [16] J. Marcucci, Forecasting Stock Market Volatility with RegimeSwitching GARCH Models, Studies in Nonlinear Dynamics & Econometrics, Berkeley Electronic Press, vol. 9(4), (2005), pages 6. [17] J.D. Hamilton, Regime Switching Models, Palgrave Dictionary of Economics, 2005. [18] J.D. Hamilton, and R. Susmel, Autoregressive conditional heteroskedasticity and changes in regime, Journal of Econometrics, 64, (1994), 307-333 [19] Kim, Jin Chang , Unobserved-Component Time Series Models with Markov-Switching Heteroscedasticity: Changes in Regime and the Link between Inflation Rates and Inflation Uncertainty , Journal of Business & Economic Statistics, vol. 11, issue 3, (1993), 341-49. [20] M. Haas, S. Mittnik, and M. S. Paollela, A new approach in markov- switching GARCH models , Journal Of Financial Econometrics, 2(4) , (2004), 493-530. [21] M. J. Dueker, Markov switching in GARCH processes and mean-reverting stock market volatility, Journal of Business and Economic Statistics, 15(1), (1996), 26-34. [22] R. F. Engle and V. K. Ng, Measuring and testing the impact of news on volatility, Journal of Finance, 48, (1993), 1749-78. [23] S. Gray, Modelling the conditional distribution of interest rates as a regime- switching process, Journal of Financial Economics, 42(1), (1994), 27-62. [24] Sebastian Edwards & Raul Susmel, «Volatility Dependence and Contagion in Emerging Equity Markets,» NBER Journal of Development Economics, Elsevier, (2001). Minoo Nazifi Naeini y Shahram Fatahi: Comparing Regime Switching GARCH Models and GARCH Models in Developing Countries (Case study of IRAN) La comparación de los modelos GARCH y GARCH de cambio de régimen en los países en desarrollo (estudio del caso de Irán) Análisis Financiero, n.º 119. 2012. Págs.: 60-68