Kit Baum’s -tscollap- is another gem from the unsecured treasure vault of user-written Stata commands. It compacts timeseries (and panel) data, creating a new dataset of means, sums, end-of-period values, and more (see -help tscollap-). For example, you have a monthly trade data that you want to add up to generate annual data or monthly data on broad money from which you only want to pick end-of-year observation.

Why not just use

-collapse-, a built-in command that also converts data into means, sums, etc.? Sure, you may use -collapse-. But for timeseries data -tscollap- offers convenience. For one, a new tsset variable

*freq_freq* (e.g., q_q, y_y, or whatever you have indicated in the generate option if you did) is generated, i.e., you need not -tsset- or -xtset- your data again before using time series commands or functions.

Let us compare -collapse- and -tscollap- below using the quarterly macro data

* lutkepohl2.dta* from Stata website. First, we use -collapse- to calculate annual data for investment, income, and consumption. To use -collapse-, we need a grouping variable—in this case a variable that indicates year—over which the sum will be calculated.

**webuse** *lutkepohl2*, clear // load data from Stata website

**gen** *year* = year(dofq(qtr)) // extract the year component from variable

*qtr ***collapse** (sum)

*inv inc consump*, by(

*year*)

Notice that variable

*year* is not read as a time variable (type -tsset- to display the current tsset settings). Thus, we must -tsset- the data again before we can use functions and commands for timeseries analysis. The following illustrates the use of -tscollap-:

**webuse** *lutkepohl2*, clear

**tscollap** (sum) inv inc consump, to(y)

-tscollap- gives the same annual data but with the bonus of automatically generating a new time variable. In the example above, variable y_y is the time variable.

-tscollap- is available from SSC. Type -ssc install tscollap- to install.

I have one question though. Why was it named -tscollap- and not -tscollapse-? Brevity? Just curious.

Filed under: Time series | Tagged: tscollap | 5 Comments »