moving skewness

[AmirSari]

Hi All,

I am trying to find a way to calculate a moving skewness by group for my sample. The window is 5 years from t-4 to t. I tried to use proc expand but it does not seem to have an option for skewness. Any help is greatly appreciated. 

here is a sample dataset.

id year x
1 2000 5
1 2001 6
1 2002 .
1 2003 9
1 2004 4.6
1 2005 7
1 2006 3.9
1 2007 4
2 2000 6
2 2001 3
2 2002 .
2 2003 2
2 2004 .
2 2005 7
2 2006 6
2 2007 8

here is my desired output:

id year x skew
1 2000 5 0
1 2001 6 0
1 2002 . 0
1 2003 9 1.29334278
1 2004 4.6 1.5163234
1 2005 7 0.43480062
1 2006 3.9 0.51928908
1 2007 4 0.97698034
2 2000 6 0
2 2001 3 0
2 2002 . 0
2 2003 2 1.29334278
2 2004 . 1.29334278
2 2005 7 1.45786297
2 2006 6 -1.457863
2 2007 8 -1.4430588

 Thank you!

[ballardw]

Your results really look like a T-3 to T window.

This is my attempt however you may need to share exactly how you are getting values for some of your "results". I get a matching value, barring rounding, for Id =1 and year=2004 but not for year 2005. I do not believe that your Id=2 and year=2003 and year=2004 could have the same result. 2003 would be skewness of 6, 3 and 2 but 2004 would be skewness of 3 and 2.

data have;
   input id year x;
datalines;
1 2000 5
1 2001 6
1 2002 .
1 2003 9
1 2004 4.6
1 2005 7
1 2006 3.9
1 2007 4
2 2000 6
2 2001 3
2 2002 .
2 2003 2
2 2004 .
2 2005 7
2 2006 6
2 2007 8
;

data want;
   set have;
   by id;
   lx1=lag1(x);
   lx2=lag2(x);
   lx3=lag3(x);

   if first.id then yrcount=1;
   else yrcount+1;
   if yrcount ge 4 then skew = skewness(x, lx1, lx2, lx3);
   else skew = 0;
   drop lx: ;
run;

[AmirSari]

Thank you for your reply.

The process is similar to moving average and standard deviation in proc expand. It is a 5-year moving skewness that requires at least three values. the values for id=1 are calculated as follows:

id year x skew
1 2000 5 0 = skew(5)
1 2001 6 0 = skew(5,6)
1 2002 . 0 = skew (5,6) because 2002 is missing
1 2003 9 1.29334278 =skew(5,6,9)
1 2004 4.6 1.5163234 =skew(5,6,9,4.6)
1 2005 7 0.43480062 = skew(6,9,4.6,7)
1 2006 3.9 0.51928908 = skew(9,4.6,7,3.9)
1 2007 4 0.97698034 = skew(9,4.6,7,3.9,4)

[ballardw · Accepted solution]

NOT a 5 year moving because of

1 2003 9 1.29334278 =skewness (5,6,., 9) That is 4 year calculation.

So you need to explain if that is a 5 year where the 5th value comes from to use as a 5 year.

This duplicates your output but note that it forces a 4-year value.

data want;
   set have;
   by id;
   lx1=lag1(x);
   lx2=lag2(x);
   lx3=lag3(x);
   lx4=lag4(x);

   if first.id then yrcount=1;
   else yrcount+1;
   if yrcount = 4 then  skew = skewness(x, lx1, lx2, lx3);
   else if yrcount ge 5 then skew = skewness(x, lx1, lx2, lx3, lx4);
   else skew = 0;
   drop lx: ;
run;

So you may need to consider what T-4 to T means.