This study uses exchange-traded fund (ETF) data to investigate the ability of the time-series volatility model, the implied volatility model, and the intraday return volatility model to forecast return volatility. Among various ETFs, we adopt NASDAQ 100 Index Tracking Stock (QQQ) as the sample because it has corresponding volatility index (VIX) issued which is necessary. The results show that all volatility models applied in this study can reliably forecast volatility. The Glosten-Jagannathan-Runkle GARCH model is superior to the GARCH model, implying that the return volatility of QQQ is asymmetric. Among the added incremental information, QQQ Volatility Index (QQV) of the American Stock Exchange has better ability in forecasting the return volatility of QQQ, followed by the NASDAQ Volatility Index (VXN) of the Chicago Board Options Exchange, and then by the intraday return volatility. The probable reason is that the turnover of QQQ options is higher than that of the NASDAQ 100 Index Options (NDX) and causes QQV to contain substantially more information than VXN and to predict volatility better. We also find the predictive power of the time-series GARCH model is weaker than that of the volatility model with QQV embedded as incremental information. Since QQQ, as an ETF, has diversified its non-systematic risks, the GARCH model using non-systematic risk information to predict volatility is inevitably worse than that using implied volatility. Identical results are achieved when examining out-of-sample forecasting performance.
Relation:
Investment Management and Financial Innovations 11(2), p.100-110
