Rolling Standard Deviation

[Chris Nes]

Hi Stata Experts,
even though I have noticed multiple threads on this issue, I am afraid I dont fully understand the instructions and, hence, am not able to reproduce them with my dataset.
My dataset consists of 1000+ Banks, with information on their return on assets (RoA) over a range of 16 quarters.
I want to calculate the 3-quarter rolling standard deviation of RoA for each bank in each quarter. How do I implement that?
Also, I do not have observations in every quarter for some banks, would I have to account for that, or does stata simply skip those banks where there is less than 3 quarters of data?

Help is much appreciated! Thank you in advance,
Chris

[Attaullah Shah]

If you provide an example data, you shall get more direct help. Anyhow, you can use asrol for rolling standard deviation, with minimum three observation. I assume that you named your time variable as quarter

ssc install asrol

bys bank: asrol RoA, stat(sd) window(quarter 3) min(3)

Please note, if your data has missing values, asrol takes care of that. If you do not want a minimum of three observations for calcualtion of the standard deviation, then remove the option min(3), so your code will look like.

bys bank: asrol RoA, stat(sd) window(quarter 3)

. For more details of asrol, type:

help asrol

[Chris Nes]

Dear Attaullah, please excuse the missing data example, I am new to statalist.
Anyhow, your solution worked perfectly! Thank you a lot!
Best,
Chris

[Nick Cox]

Here's another way to do it . The style is similar -- but also different. With rangestat (SSC) you can specify that you want several statistics. Here I keep track of the count of the number of observations used. I didn't see a precise specification of how your three quarters are to be set-up. With rangestat you can have them centred on each observation (one past, one now, one future), backwards, forwards,etc.

The example dataset here isn't panel data, but I pretend that it is and show you the syntax.

. webuse turksales, clear  

. gen whatever = 42 

. rangestat (count) sales (sd) sales, interval(t -2 0) by(whatever) 

. list in 1/12, sep(4)

     +-----------------------------------------------------+
     |      t      sales   whatever   sales_~t    sales_sd |
     |-----------------------------------------------------|
  1. | 1990q1        100         42          1           . |
  2. | 1990q2   97.84603         42          2    1.523086 |
  3. | 1990q3   98.84029         42          3    1.078043 |
  4. | 1990q4   100.8275         42          3   1.5180421 |
     |-----------------------------------------------------|
  5. | 1991q1   98.90981         42          3   1.1277854 |
  6. | 1991q2   100.9992         42          3   1.1599403 |
  7. | 1991q3   101.9653         42          3   1.5617701 |
  8. | 1991q4   104.1229         42          3   1.5992399 |
     |-----------------------------------------------------|
  9. | 1992q1   99.74297         42          3   2.1900249 |
 10. | 1992q2   102.3116         42          3   2.2008343 |
 11. | 1992q3   104.2382         42          3    2.255268 |
 12. | 1992q4   105.8506         42          3   1.7718276 |
     +-----------------------------------------------------+

[Chris Nes]

Hi Nick, great! I also tried the way you suggested and it worked out as well.
Thank you!