Adjusting the window of a Moving Average

[Ryan Marapin]

Dear all,

I am currently constructing a 5-window and 7-window moving average to smooth out a time-series data I am analysing. As expected, for the first and last 2 observations of the time-series I cannot calculate a moving average as there are no previous year-quarter observations available. Therefore, I would like to make the window smaller for these two observations, i.e. a window of 3 for the 2nd observation, and a window of 1 for the first observation (basically moving average of observation= value observation)

In theory, I understand this, unfortunately I am not that experienced with Stata so I have no clue how to code this. If someone could help me, I would be very grateful.
I am aware that there are commands available in Stata that allow me to do this easier, but I am interested in writing it as a code so I am able to understand better the process behind it and improve my knowledge of Stata.

The command I am using to create the 5-window MA: gen c1 = (l2.earnings_nsa+2 (l1.earnings_nsa+earnings_nsa+f1.earnings_nsa)+f2. earnings_nsa)/8

My data looks as follows:

 * Example generated by -dataex-. To install: ssc install dataex
clear
input float(yq earnings_nsa)
-84       .49
-83       .49
-82  .4833333
-81       .51
-80       .52
-79  .5233333
-78  .5233333
-77 .54333335
-76       .56
-75  .5966667
-74  .6133333
-73       .65
-72  .6866667
-71       .72
-70       .76
-69       .79
-68       .82
-67  .8466667
-66       .87
-65  .8866667
-64        .9
-63       .91
-62  .9133334
-61  .9233333
-60       .93
-59  .9266667
-58  .8866667
-57       .85
-56  .8666667
-55  .9333333
-54       .98
-53      1.01
-52 1.0433333
-51      1.08
-50      1.11
-49 1.1433333
-48      1.16
-47      1.18
-46 1.2233334
-45      1.25
-44      1.26
-43 1.2566667
-42      1.25
-41 1.2433333
-40 1.2733333
-39       1.3
-38      1.32
-37 1.3766667
-36 1.4166666
-35      1.44
-34 1.4533334
-33 1.4833333
-32 1.5066667
-31 1.5166667
-30 1.5233333
-29 1.5766667
-28 1.6066667
-27 1.6233333
-26 1.6366667
-25      1.65
-24      1.66
-23 1.6633333
-22 1.6533333
-21      1.68
-20 1.7033334
-19 1.7266667
-18      1.75
-17      1.78
-16 1.7966666
-15 1.8233334
-14 1.8366667
-13      1.89
-12 1.9133333
-11 1.9233333
-10      1.93
 -9 1.9566667
 -8 1.9566667
 -7 1.9733334
 -6 1.9833333
 -5 2.0233333
 -4 2.0566666
 -3 2.0833333
 -2 2.0666666
 -1 2.0866666
  0      2.14
  1      2.14
  2      2.14
  3 2.1566668
  4      2.17
  5 2.1933334
  6 2.1933334
  7 2.2366667
  8 2.2633333
  9      2.27
 10 2.2633333
 11 2.2966666
 12 2.3133333
 13 2.3333333
 14 2.3333333
 15      2.37
end
format %tq yq

Thank you very much in advance!

Rajendra

[Nick Cox]

Rajendra or Ryan or both:

The recipe you want isn't difficult, although the code you cite is illegal! Presumably you used

gen c1 = (l2.earnings_nsa+2 * (l1.earnings_nsa+earnings_nsa+f1.earnings_nsa)+f2. earnings_nsa)/8

so your weights are (1 2 2 2 1) / 8. If that didn't work, we can try (say)

replace c1 = (l1.earnings_nsa+2 * earnings_nsa+ f1.earnings_nsa))/4 if missing(c1)

and if that didn't work, we can try

replace c1 = earnings_nsa  if missing(c1)

I quite like the general idea of binomial filters, so that I would use (1 4 6 4 1) / 16 for windows of length 5. More at

SJ-4-4 gr22_1 . . . . . . . . . . . . . . . . . Software update for bsmplot
(help bsmplot if installed) . . . . . . . . . . . . . . . . N. J. Cox
Q4/04 SJ 4(4):490
binomial smoothing plot program rewritten so that it now
produces Stata 8 graphs

STB-35 gr22 . . . . . . . . . . . . . . . . . . . . Binomial smoothing plot
(help bsmplot if installed) . . . . . . . . . . . . . . . . N. J. Cox
1/97 pp.7--9; STB Reprints Vol 6, pp.36--38
produce a plot of both yvar and the result of smoothing yvar
by a binomial filter against xvar

[Ryan Marapin]

Dear Dr. Cox,

Thank you once again for your clear and helpful answer! the code worked perfectly. I checked out the idea of binomial filters, interesting idea to consider indeed!

Best,

Rajendra