It might be possible to do with syntax, if you understand the calculations needed, but there are apparently no built-in functions for it: http://www-01.ibm.com/support/docview.wss?uid=swg21478484
If you search on Google you will find pages describing how to do it in other environments, like R.
2. Split the data into two halves. First half includes the first c data values while the second half includes values c+1, c+2, ..., n. If sample size is odd, remove the middle observation.
3. Match up the observations from the first half with the values in the second half.
1 c+1
2 c+2
.
.
c n
4. Use a sign test for a matched (or paired) sample. For testing for an upward trend, you would assign a + if X(i) < X(c+i).
This can be programmed, as Stephen has said above. The sign test does not have much statistical power until the sample size is quite large. Why are you planning on using a Cox and Stuart Test for Trend? Do you have discrete data or maybe you have continuous data but you suspect that the data are non-normal? Another nonparametric test for trend is by Jonkheere-Terpstra (see Coonver's text on nonparametric statistics). SAS has this test under proc freq. I am not sure if SPSS has it too or not. The JT test may have more power than the Cox and Stuart test. Google it.
Raid has given a really good summary of how you would go about this. I think it's worth working out if it is worth using R for (I would suggest yes), but also, as Raid asks, is this definitely the way you want to go?