Hi @mkeintz,   your help is greatly appreciated!    The code seems to work in general but there are a few issues I cannot resolve. That's what my output looks like:   gvkey is my id and length corresponds to size.   1. The cells marked in yellow contain the same id, which shouldn't be the case, right?   2. I do know what the maximum length of a subseries is, but I do not know when it starts and when it ends.   3. How do I transfer the information I got in this step to my original data set, which contains all my observations with quarterly cash flows. Somehow I now need to delete all the sub-series in my original data set that are not the longest of a firm.    Once I managed this, I can proceed with the output.   Maybe you have another idea on how to tackle this. Thanks again for all the help!   Best, Waynerun           ... View more

Hi @mkeintz, thank you very much for your contribution!   The code you suggested worked perfectly 🙂   Just for better understanding. What does "^=1" mean?   The questions you have raised are also good food for thought. First, you are right. It doesn't matter if the companies have different time series. Some firms were established in 1989, others years later. So that's not a problem.   Secondly, the data are prepared in such a way that each company always starts with the first quarter of a fiscal year and ends with the fourth quarter of a fiscal year. So it should not be possible to have missing values only for Q3. If the value for Q3 is missing (e.g. in 1993), the entire year is missing. However, this could be a very strict approach and could be reconsidered.   Furthermore, I have reviewed my new data set, and indeed there is often a break in the early years of my time series. Therefore, censoring the data after the break would eliminate most of the company's observations, as you suspected. Is there any way to determine which of the two (or perhaps even three) subseries is the longest? If so, it might be a good idea to use the longest subseries for further analysis.    For a minimum series size, 12 consecuitive observations (3 years) could be suitable as this would correspond to the 12 lags of the autocorrelation function. I think that since the data set is large enough to get meaningful results, it is not necessary to create two subseries.   I look forward to hearing your and others' thoughts on this topic and how to test best for seasonality!   Best, Waynerun ... View more

Thank you very much @ballardw and apologies for my vague description. My previous approach was so flawed that it didn't make sense to post it in the chat.   I rearranged my data set according to your example. Now I have the problem that some firms do not have complete time series. For example   id          date                     Value a           01Jan1998             5 a           01Apr1998             8 a           01Jul1998             10 a           01Oct1998             3 a           01Jan2001             2 a           01Apr2001             6 a           01Jul2001             7 a           01Oct2001             9   Is there a procedure I can use to delete all observations after a break occurs in my time series?     Regarding the final output, I was thinking to test seasonality by looking at the autocorrelation function. So, I would do this for all firms in the sample individually and then calculate a mean across all firms.  Similar to the following picture.     But if there is another way to test seasonality in my cross-sectional data, this would be great as well!   Best, Waynerun     ... View more