topic How to forecast TimeSeries data dependent on same time and day of week in Statistical Procedures

How to forecast TimeSeries data dependent on same time and day of week

vasanthz — Thu, 01 Feb 2024 13:31:13 GMT

Hi Experts,

I have a SAS dataset that has transactions count by hour for 3 years.

The count of transactions at a specific hour depends on the same hour and day of the previous n weeks.

Example, The transaction count of future Tuesday 10AM, depends on all the past Tuesdays 10AM values.

The value at Thurdsay 6 PM would depend on all past Thursday 6PM.

Could you please let me a starting program to forecast this for the next day. Thank you. I have tried using PROC ARIMA with NLAG=168. But that doesn't seem to work, the forecast for 00:00 hours starts at high value and then the forecast value tends to move to zero by 23:00

PROC ARIMA DATA=PDB.UNIFIED_TS;
IDENTIFY VAR=SALES NLAG=168;
ESTIMATE P=1 Q=1;
RUN;
FORECAST LEAD=168 INTERVAL=HOUR ID=DATETIME OUT=PDB.RESULTS;
RUN;
QUIT;

I then tried using PROC UCM, but I am not sure what I am doing wrong there.

PROC UCM DATA=TIMESER_TS;
ID DATETM INTERVAL=HOUR;
MODEL MSU;
IRREGULAR;
LEVEL;
SLOPE;
SEASON LENGTH=24 TYPE=TRIG;
ESTIMATE;
FORECAST LEAD=48 OUTFOR=FORE_UCM;
RUN;
QUIT;

Thank you,

Vasanth.S

Re: How to forecast TimeSeries data dependent on same time and day of week

vasanthz — Thu, 01 Feb 2024 13:32:02 GMT

Pardon my ignorance, I am a base programmer and don't have much knowledge on statistics and learning now.

Re: How to forecast TimeSeries data dependent on same time and day of week

Ksharp — Fri, 02 Feb 2024 05:04:02 GMT

It is a question about SAS/ETS.
Better post it at Forecasting forum:
https://communities.sas.com/t5/SAS-Forecasting-and-Econometrics/bd-p/forecasting_econometrics

Also you could try PROC ESM .
And I quoted:
"The ESM procedure generates forecasts by using exponential smoothing models with optimized smoothing
weights for many time series or transactional data."