A1675
Title: Volatility forecasting with empirical similarity: Japanese stock market case
Authors: Takayuki Morimoto - Kwansei Gakuin University (Japan) [presenting]
Yoshinori Kawasaki - The Institute of Statistical Mathematics (Japan)
Abstract: The forecasting ability of various volatility models is compared through within-sample and out-of-sample forecasting simulations. The considered models are heterogeneous auto-regression models (HAR), a 1/3 model where the weight coefficients are all set to 1/3 in the HAR model (ES0), and an HAR model where the weight coefficients are determined by their empirical similarity. We also test AR(1), ARCH/GARCH and their variants, and models incorporating the realized quarticity (RQ), which are referred to as ARQ, HARQ, and ESQ. For stock data, we picked six index series stocks that are listed on the Tokyo Stock Exchange as well as 24 individual stock series. All these stocks had enough liquidity in the market from April 1, 1999, to December 30, 2013, for our investigation. Minute-by-minute data were created based on high-frequency data. Forecasting evaluation depends on what kind of evaluation function we employ. We make use of Patton's error function. By changing the length of estimation period and the forecasting period and the parameter of Patton's error function, we attempt 27,000 forecasting simulations. We find that ESQ and HARQ are almost comparative in within-sample forecasting, whereas ES0 differs in out-of-sample forecasting experiments. We also tried a model comparison based on a previous pair-wise testing procedure. We found similar results, but the details are different between the index series and the individual stock series.
