NO.PZ2020011101000020
问题如下:
Suppose an hourly time series has a calendar effect where the hour of the day matters. How would the dummy variable approach be implemented to capture this calendar effect? How could differencing be used instead to remove the seasonality?
解释:
Let s = 24 represent the hour of the day in military time (e.g. 13 = 1 p.m.). Then .
Differencing this series can be achieved by looking at observation 24 periods (hours) apart from each other (the following presumes that the error terms are iid and normal):
Once the deterministic time trend is removed the remaining is a covariance-stationary MA(1) process.
24小时对应23个哑变量,第24个Yt是前23个哑变量为0。那么是不是要从第25小时开始下一个循环,对应第1小时呢?