Samples with different sizes

[4545]

Hello all,

I'm trying to explain a dependent variable that assumes values only during trading days, based on variables that vary in sample size since some occur everyday, some other occur only during weekends. How can I workaround this? I can't regress this with different sample sizes, can I?

also, since some variables do not occur everyday, when I try to make them stationary (and because some days there is a "NA" for those variables), using ARMA models i get "MA estimation requires a continuous sample".

Can somebody help me?
Thanks in advance!

[trubador]

First, you need to make a decision as to what frequency are you going to work in. Variables that are observed 7 day a week will be OK in any case. The problem is to decide between trading day (5-day week) and weekend. The former will probably have more observations than the latter.

Since your dependent variable is in the former frequency, it is the determining factor and you should go with 5-day week frequency. You need to drop your "weekend" variables as there is no actual impact to measure.

Finally, you need to build a common sample and remove NAs before carrying out MA or GARCH modeling. This is not automatically done like other routines.