deseasonalize time series

[Felipe Damasceno]

I'm using stata 14 with Windows 10 OS.

I dealing with quarterly accounts receivable and payables data. I want to deseasonalize but I have no idea how.


The data below is a pick of the mean receivables over all firms of my sample by quarter (because it shows the overall seasonal tendency). Does anyone knows how to smooth this time series?

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(qdate meanRec)
159  .7051772
160  .9537871
161 .52570915
162  .3577844
163  .4055271
164 1.5014802
165  .6501129
166 .40079415
167  .4518343
168 2.0093749
169  .8835684
170   .551052
171  .5462328
172 1.5105826
173   .778314
174  .5514978
175  .6027386
176 1.5464938
177  .9570103
178  .6769719
179  .7758076
end
format %tq qdate

[Nick Cox]

Lots of ways to do it, the best probably being a structural time series model.

Meanwhile, a naive regression approach is of a little interest:

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(qdate meanRec)
159  .7051772
160  .9537871
161 .52570915
162  .3577844
163  .4055271
164 1.5014802
165  .6501129
166 .40079415
167  .4518343
168 2.0093749
169  .8835684
170   .551052
171  .5462328
172 1.5105826
173   .778314
174  .5514978
175  .6027386
176 1.5464938
177  .9570103
178  .6769719
179  .7758076
end
format %tq qdate

gen quarter = quarter(dofq(qdate)) 

tsset qdate 
reg meanRec L1.meanRec qdate i.quarter 

predict crude

line meanRec crude qdate, xtic(`=yq(2000,1)-0.5'(4)`=yq(2005,1)-0.5', tlength(*3)) ///
xla(`=yq(2000,2)+0.5'(4)`=yq(2004,2)+0.5', format(%tqCY) tlength(0)) xtitle("")

[Felipe Damasceno]

Hey Nick.

Actually I don't to mitigate the higher spikes. I believe they are just the effects I want to measure. Nevertheless, we can see that every forth quarter we have a small spike tendency that might be questioned about driving some results. I like to deseasonalize the time series to mitigate such questions.

[Nick Cox]

As I commented in 2006, economists see seasonality as a nuisance and environmental scientists and epidemiologists see it as interesting structure.

So, just fit a model with quarterly indicators.

[Felipe Damasceno]

i see. It makes sense.

Thanks Nick