Page 106 - Contributed Paper Session (CPS) - Volume 6

CPS1837 Qiguang Dong et al.

CSI300 volatility forecasting model and its MCS
test
2
Qiguang Dong , Hang Li , Lili Chen
1
3
1 PLA Military Science Academy, 100142, Beijing, China
2 Donlinks School of Economics and Management, University of Science and Technology
100083, Beijing, China
3 School of Economics and Management, Tsinghua University, 100084, Beijing, China

Abstract
In this paper, 5 min frequency observations are taken to forecast the actual
volatility of CSI300 stock index intraday returns. Both realized volatility and
logarithm-transformed realized volatility are modelled directly in the ARFIMA
model specification. Besides, GARCH family models and different distributions
are utilized to address the potential heteroscedasticity problem. Applying the
out-of sample rolling time window forecasting and Model Confidence Set
which is proved superior to SPA test, this paper compares the empirical
performance of all specified models. The empirical results show that: (1) Both
RV and LnRV series have a long memory due to both Hurst indexes are greater
than 0.5 and smaller than 1. (2) The symmetric and skewed generalized error
distributions ged and sged are employed more accurate than normal and
student-t distributions. (3) The model LnRV-sGARCH-sged is outperformed
than the rest in the long memory model as well as in the short memory model.

Keywords
Volatility forecasting; stock index; MCS test; CSI300

1. Introduction
The efficiency of the volatility forecast is crucial for option pricing, but also
in many areas of finance. Since ARCH models, GARCH models and Stochastic
Volatility models are introduced to estimate and forecast volatility in financial
market, these models gradually established their dominance and still in the
continuous development. However, these models usually use lower frequency
observations to estimate and forecast volatility of financial assets, which lost
an amount of intraday trading information and have difficulty to deal with
multidimensional problems. High frequency observations can quickly and
effectively to capture market information, which can be more accuracy to
reflect the actual situation in the financial markets than lower frequency
observations. (Andersen and Bollerslev, 1998; BarndorffNielsen and
Shephard,2002). RV is the sum of intraday squared returns, which accuracy
depends on the returns of intraday high frequency observations (Taylor and

95 | I S I W S C 2 0 1 9