Hi @ubshams ,    Different default settings were chosen to be used for the number of augmenting lags for the ADF stationarity tests computed by PROCs ARIMA and AUTOREG.   By default, PROC ARIMA computes the stationarity tests for 0, 1 and 2 augmenting lags.  This is noted in the initial paragraph of the STATIONARITY= option documentation shown in the excerpt below:   STATIONARITY= performs stationarity tests. Stationarity tests can be used to determine whether differencing terms should be included in the model specification. In each stationarity test, the autoregressive orders can be specified by a range, test= ar_max , or as a list of values, test=(ar1,...arn) , where test is ADF, PP, or RW. The default is (0,1,2).     PROC AUTOREG, on the other hand, assumes that the specified syntax for the STATIONARITY= option results in models for the ADF test, where the disturbances are distributed as white noise.  Therefore, a specification of STATIONARITY=(ADF) assumes that the residuals from the standard Dickey-Fuller test (without any augmenting lags) are white noise.  This assumption is noted in the following excerpt from the PROC AUTOREG documentation, however, the default behavior could be more clearly described:   STATIONARITY=(ADF) STATIONARITY=(ADF=(value …value) produces the augmented Dickey-Fuller unit root test (Dickey and Fuller 1979). As in the Phillips-Perron test, three regression models can be specified for the null hypothesis for the augmented Dickey-Fuller test (zero mean, single mean, and trend). These models assume that the disturbances are distributed as white noise. The augmented Dickey-Fuller test can account for the serial correlation between the disturbances in some way.     For more details on the stationarity tests fit in PROC AUTOREG, please see the following link:   https://go.documentation.sas.com/?docsetId=etsug&docsetTarget=etsug_autoreg_details27.htm&docsetVersion=15.2&locale=en#etsug.autoreg.stationarytsts    I hope this helps explain the difference in the results you obtained when running the two procedures.  I will request that the documentation for PROC AUTOREG be updated to further clarify the default behavior for this option.  Thank you for your feedback! Best Regards, DW ... View more

Hi @karora1 ,    In general, the METHOD=NONE option should not be used in PROC EXPAND when the TO= option is also specified.  One alternative approach is to first use PROC TIMESERIES to fill in the gaps in your data with 0 values, and then run PROC EXPAND to create your moving sum.  Another alternative is to use PROC TIMEDATA to both fill in the gaps in your data and compute the moving sum.  Both of these alternatives are shown below using the example data provided by @PeterClemmensen .   data have; input id dt :date9. var; format dt date9.; datalines; 1 01jan2020 1 1 02jan2020 2 1 04jan2020 3 2 01jan2020 4 2 02jan2020 5 2 04jan2020 6 ; /* fill gaps in the data with 0 values */ /* based on INTERVAL=DAY specification */ /* and SETMISSING=0 option. */ proc timeseries data=have out=temp; by id; id dt interval=day; var var / setmissing=0; run; /* compute moving sum on daily series */ proc expand data=temp out=want method=none; by id; id dt; convert var = movesum / transformout = (movsum 2); run; proc print data=want; run; /* fill in gaps with 0 values and compute */ /* moving sum in one step. */ proc timedata data=have out=_null_ outarrays=want2(drop=_:); by id; id dt interval=day; var var / setmissing=0; outarrays movsum; movsum[1]=var[1]; do t=2 to _length_; movsum[t]=var[t] + var[t-1]; end; run; proc print data=want2; run; I hope this helps! DW       ... View more

Hi @somebody ,   @Kurt_Bremser is exactly correct in his comments regarding the use of the ID statement in PROC EXPAND.     You indicated that you have gaps in the data due to weekends and holidays.  You can ignore the gaps due to weekends by using the 'weekday' interval.  By default, the weekday interval is defined as a 5-day interval from Monday-Friday.  If you have gaps in the weekday data due to holidays, then you might want to fill in those gaps with missing values prior to computing the lead variable.  You can perform your calculations using either a combination of PROC TIMESERIES (to fill in the gaps) and PROC EXPAND (to create the lead variable), or you can do this all in one step using PROC TIMEDATA.   Below is a slightly modified version of the earlier data you had provided, where some observations are omitted in each BY group to create gaps in the weekday data.  The two approaches mentioned above are illustrated in the code, along with some comments.     /* created some gaps in weekday data for each BY group */ data WORK.sample; infile datalines dsd truncover; input date:YYMMDD8. Name:$100. Holding:BEST12.; format date YYMMDDN8. Holding BEST12.; datalines4; 20130916,ORACLE CORP,34776812 20130917,ORACLE CORP,35633772 20130918,ORACLE CORP,36068432 20130919,ORACLE CORP,37077832 20130920,ORACLE CORP,36241472 20130923,ORACLE CORP,35624692 20130924,ORACLE CORP,35356424 20130925,ORACLE CORP,35038106 20130926,ORACLE CORP,34763832 20130927,ORACLE CORP,34763832 20130930,ORACLE CORP,34473542 20131001,ORACLE CORP,34473542 20131002,ORACLE CORP,34849730 20131003,ORACLE CORP,35494052 20131004,ORACLE CORP,35233922 20131008,ORACLE CORP,34601606 20131009,ORACLE CORP,33997304 20131010,ORACLE CORP,33845228 20130916,MICROSOFT CORP,71171845 20130917,MICROSOFT CORP,72924869 20130918,MICROSOFT CORP,73814023 20130919,MICROSOFT CORP,75878883 20130920,MICROSOFT CORP,74167999 20130923,MICROSOFT CORP,75703447 20130924,MICROSOFT CORP,75133009 20130925,MICROSOFT CORP,74456146 20130930,MICROSOFT CORP,73255808 20131001,MICROSOFT CORP,73255808 20131002,MICROSOFT CORP,74055748 20131003,MICROSOFT CORP,75425858 20131004,MICROSOFT CORP,74872708 20131007,MICROSOFT CORP,74102553 20131008,MICROSOFT CORP,73528128 20131009,MICROSOFT CORP,72243118 20131010,MICROSOFT CORP,71919738 20130916,T ROWE PRICE GROUP,2449245 20130917,T ROWE PRICE GROUP,2509565 20130918,T ROWE PRICE GROUP,2540160 20130920,T ROWE PRICE GROUP,2552340 20130923,T ROWE PRICE GROUP,2578730 20130924,T ROWE PRICE GROUP,2559166 20130925,T ROWE PRICE GROUP,2535952 20130926,T ROWE PRICE GROUP,2515950 20130927,T ROWE PRICE GROUP,2515950 20130930,T ROWE PRICE GROUP,2494780 20131001,T ROWE PRICE GROUP,2494780 20131002,T ROWE PRICE GROUP,2522228 20131003,T ROWE PRICE GROUP,2569240 20131004,T ROWE PRICE GROUP,2550260 20131007,T ROWE PRICE GROUP,2523834 20131008,T ROWE PRICE GROUP,2504124 20131009,T ROWE PRICE GROUP,2460032 20131010,T ROWE PRICE GROUP,2448936 ;;;; proc sort data=sample; by name date;run; /* run PROC TIMESERIES to fill in gaps in the data with missing values */ /* based on WEEKDAY interval */ proc timeseries data=sample out=fullsample; by name; id date interval=weekday; var holding / setmissing=missing; run; /* see gaps filled in with missing values */ proc print data=fullsample; run; /* run PROC EXPAND with METHOD=NONE to prevent missing values from being */ /* interpolated prior to computing the lead variable. Including the ID */ /* statement allows values of ID variable to be written to OUT= data set */ proc expand data=fullsample out=sample3 method=none; by name; id date; convert holding = lead_holding_with_id/ transformout=(lead 1); run; proc print data=sample3; var name date holding lead:; title 'Results from PROCs TIMESERIES and PROC EXPAND'; run; /* Alternatively, use PROC TIMEDATA to fill in gaps and create LEAD variable */ /* in a single PROC step. */ proc timedata data=sample out=_null_ outarrays=sample4(drop=_s: _c:); by name; id date interval=weekday; var holding / setmissing=missing; outarrays lead_holding_timedata; do t=1 to _length_ - 1; lead_holding_timedata[t]=holding[t+1]; end; run; proc print data=sample4; var name date holding lead:; title 'Results from PROC TIMEDATA'; run; title; I hope this helps! DW ... View more

Hi @itslarajean ,    Although the PROC VARMAX documentation does not include many details on the output of the ROOTS option, the SAS/IML product includes the VTSROOT call routine, which performs a similar analysis.  Please see the following documentation link to the VTSROOT subroutine for a description of the calculations and output:   https://go.documentation.sas.com/?docsetId=imlug&docsetTarget=imlug_langref_sect553.htm&docsetVersion=15.1&locale=en    Additional details can also be found in the following reference: Lütkepohl, H. (1993). Introduction to Multiple Time Series Analysis. 2nd ed. Berlin: Springer-Verlag. I hope this helps! DW ... View more

Hi @Golf ,   The difference in the results returned by the MINIC option in your two PROC ARIMA steps is due to a difference in the approximation used for the error series.  In the first PROC ARIMA step, the error series model was an AR(12) model, whereas in the second PROC ARIMA step, the error series model was an AR(15) model.  If you include the PERROR= option with the same specification in both IDENTIFY statements, then the Minimum Information Criterion values will the the same through AR lag 12.  For example:   proc arima data = ARDL; identify var = dnpl MINIC p=(1:12) q=0 perror=(10:12); identify var = dnpl MINIC p=(1:15) q=0 perror=(10:12); run; Regarding your question on how to "add intercept and trend to this procedure", I assume you mean to the underlying procedure used by the MINIC option.  The methodology used by the MINIC option automatically corrects for the mean, however, it assumes the series is stationary (ie. constant mean and variance).  If your series is non-stationary, then you can include a differencing specification in the VAR= option to detrend the series.  The MINIC option operates on the differenced series.  If your series has a deterministic trend component, then you will need to detrend the series prior to running PROC ARIMA with the MINIC option.   For more information on the MINIC option, please see the following documentation link: https://go.documentation.sas.com/?docsetId=etsug&docsetTarget=etsug_arima_details05.htm&docsetVersion=15.1&locale=en    I hope this helps! DW ... View more