01-03-2018 12:06 PM
I have daily time series data from 2014 to today. I used it to build ARIMA models and forecast for the next days and weeks.
Now, i also have 15 minutes interval data, and would like to build model and do forecasting for the next 15 minutes periods.
I know that ARIMA model does not fit for interval forecasting. Do you have a model that fit for interval forecasting ?
01-03-2018 03:08 PM
Hi. do you have an example of your code? I am not sure why you cannot do the same forecasting with ARIMA that you did with the daily series. Could you do this by changing the seasonality of your interval?
I'm not sure what you mean by fitting for interval forecasting, isn't that what you did when you forecasted the daily data?
01-03-2018 03:20 PM
We can use ARIMA for 15 minutes period, for daily, for monthly, ...
But ARIMA is not the best model for 15 minutes period.
I will figure out the appropriate model.
01-03-2018 03:32 PM
It may help to know what is the nature of the data that you are trying to forecasts and why you think ARIMA is not adequate. SAS/ETS offers several other model classes for time series data. However, without knowledge of the data it is difficult to make suggestions.
01-03-2018 10:32 PM
As long as your time series measurements are equally spaced in time, you can use PROC ARIMA to model/forecast them. The time interval information such as daily, monthly, or fifteen-minute plays no role in the actual computation. The time interval information does play a role in what type of model you might want to fit the series. For example, monthly data usually has seasonal pattern of length 12 (differencing order of 12) whereas quarterly data has seasonal length of 4. This means that what model to fit your series depends on your knowledge about the series. You can specify very general ARIMA models using PROC ARIMA. You will need to decide what type of seasonality makes sense for your 15-minute interval series---for example, if your measurements are stock trades every fifteen minute during an 8 hour business day then your season length might be 32 (4 fifteen minute intervals per hour).
Hope this helps.
01-03-2018 10:59 PM
01-09-2018 09:24 AM
I think you probably need to include multiple seasonality in the model such as hour of the day and day of the week. You will need to create dummy vars or trig curves to represent the desired seasonality, and include them as independent variables in the model. You could using regression with some lagged independent variables or auto-regression in addition to ARIMA and UCM with the additional independent vars.