Trend function in SAS

[Christobal]

Hello all,

I have the following problem (SAS 9.1.3 user):

I have a table containing values like this:

01/2013          2.200.225 units

02/2013          2.250.000 units

03/2013          2.376.999 units

....                 ....

02/2014          3.000.100 units

I am looking for a sas function that generates a trend/forecast for these values (for e.g.: 12/2014    trend: 3.255.000). While searching the web I found a function called "proc timeseries", but unfortunately this function is not available in SAS 9.1.3.

Does anyone know a function or solution for my problem?

Thanks in advance and best regards

Cristobal

[Murray_Court]

A linear regression would give you the answer based on a linear trendline..

Proc reg data=have;

     model units = date;

  run;

This will give you gradient and intercept coefficients that you can apply to the date (in sas date form) to give you a linear estimate.

This code and be emended to include polynomial trend-line fitting as shown to account for higher order relationships:

Proc reg data=have;

     model units = date date*date;

  run;

[Christobal]

Hello,

thanks for your replay. Unfortunately the reg procedure cannot be found. I get the message "ERROR: Procedure reg not found".

It seems, that this procedure is not available in SAS9.1.3?

Best regards

[SteveDenham]

Do you have SAS/STAT installed?  You can run PROC SETINIT to see what modules are installed.

It is an unusual installation to have SAS/ETS and not SAS/STAT, but that is possible.  You can use PROC MODEL as a way to do simple regression if only SAS/ETS is installed.

Steve Denham

[cau83]

Lag the units, calculate % change, take average (crude way but I know it would work with Base SAS and any version):

01/2013          2.200.225 units        .

02/2013          2.250.000 units        2.200.225      0.2%   

03/2013          2.376.999 units        2.250.000      5.6%

....                 ....

02/2014          3.000.100 units

So the trend after the first 3 months would be 2.9%.

Then put the average of pctchange (ie trend) into a macro variable and bring it back into the data set to do forecast:

proc sql;

          select mean(PctChange) into :Trend from...;

quit;

Data Forecast;

         set ...;

          PredictedUnits = LagUnits * (1+&Trend);

run;

[Christobal]

Hello Steve, hello Chris,

thanks for your answers. Due to weekend I couldn't answer earlier 🙂

It seems that I don't have SAS/STAT installed. I will check whether this is available for me.

Your hint Chris seems to be a good solution for me. I will try whether this works fine.

Best regards

[Christobal]

Ok, now I calculated my trend and my table looks like this:

Month	Units	Prediction	pctChange	trendPct	Trend
2013/06	200000			0,38%
2013/07	210000	204000	5,00%	0,38%
2013/08	198000	208000	-5,71%	0,38%
2013/09	202000	212000	2,02%	0,38%
2013/10	210000	216000	3,96%	0,38%
2013/11	179000	220000	-14,76%	0,38%
2013/12	195000	224000	8,94%	0,38%
2014/01	198000	228000	1,54%	0,38%
2014/02	202000	232000	2,02%	0,38%
2014/03		236000			202767,6
2014/04		240000			203538,117
2014/05		244000			204311,562
2014/06		248000			205087,946
2014/07		252000			205867,28
2014/08		256000			206649,576
2014/09		260000			207434,844
2014/10		264000			208223,096
2014/11		268000			209014,344
2014/12		272000			209808,599

Now I want to print this table in a pdf document.

Till today I only printed the "units per month" and the "prediction per month":

proc gplot data=work.unitsvsprediction;

;

plot units*monat=4  / vaxis=axis1 haxis=axis3 grid;

plot2 prediction*monat=2  / vaxis=axis2;

run;

I tried to extend this statement by plot3 trend*month.... but unfortunately proc gplot doesn't support 3 plots. Is there any other possibility to solve this?

[cau83 · Accepted solution]

I'm not a SAS GRAPH expert, but I do have EG so it generated this code for me when I plotted 4 different variables on the same vertical axis. It looks like you were using 2 axes, but you should still be able to adapt this and use the asterisk for the multiple series on one of the vertical axes:

PROC GPLOT DATA = WORK.SORTTempTableSorted

96         ;

97         PLOT de2_clips * date manual_clip_effort * date autokey_effort * date manual_fullpage_effort * date  /

98          OVERLAY

99         VAXIS=AXIS1

100       

101         HAXIS=AXIS2

102       

103        FRAME LEGEND=LEGEND1

104        ;

105        /* -------------------------------------------------------------------

106           End of task code.

107           ------------------------------------------------------------------- */

108        RUN;

[udo_sas]

Hello -

Since your data features 2 different dimensions (% and absolute values) for Y values, you may want to look into using the SGPLOT procedure.

Here is an example (I converted your input data to an CSV file).

Thanks,

Udo

data WORK.HAVE    ;

    infile 'yourpath\have.csv' delimiter = ',' MISSOVER DSD firstobs=2 ;

    informat Month anydtdte7. ;

    informat Units best.;

    informat Prediction best. ;

    informat pctChange best. ;

    informat trendPct best. ;

    informat Trend best. ;

    format Month date8. ;

    format Units best. ;

    format Prediction best. ;

    format pctChange best. ;

    format trendPct best. ;

    format Trend best. ;

    input Month Units Prediction pctChange trendPct Trend $;

run;

proc sgplot data=have;

series x=month y=units;

series x=month y=prediction;

series x=month y=trend;

series x=month y=pctchange /y2axis;

series x=month y=trendpct / y2axis;

run;

Results in:

[Christobal]

Hello Chris,

thanks for your answer. The "overlay" option was the keyword. Now it works 🙂

Hello Udo,

I will also try your option with proc sgplot, as it looks very interesting and seems to help for the future 🙂

Best regards