以下是引用irvingy在2006-10-1 17:32:00的发言： It's still a single number. As I said before, volatility is standard deviation, no matter which model you're using. Unde GARCH, you're forecasting variance. It's a forecast for one day. Take square root, you can get volatility forecast for that day. To get average volatility forecast for a period, you add up variance and take square root. Taking geometric mean of volatility doesn't make sense. As I said before, jamgromin is confusing annualized volatility and actual/forecasted one-year volatility.
To jamgromin, GARCH is not used to get in-sample variance/volatility. For your specific problem, you have data from January to July. You can use GARCH to forecast volatility of August 1st, say 1%, which is a daily volatility. It is equivalent to annualized volatility 1%*sqrt(252). To forecast the one-year volatility, i.e. volatility for the period from August 1st to July 31st next year, you have to use GARCH to get variance for each day in that period, add them up and take square root, then you get the one year volatility forecast.
了解你说的single number的意思了。假设使用EWMA模型，使用1月1日－8月31日的日数据（假设有n天），可以得到n－1个日波动率。但实际上n－1个波动率是通过n－1个样本得到的。
jamgromin的问题我也曾经遇到过。事实上，如果想得到n－1个波动率，那只能是年度化的日波动率（annualized daily volatility），而不可能是年波动率。
扫码加好友，拉您进群
全部版块
我的主页
收藏
