Let me clarify I believe PMEAN computes the average for each cross-section unit, which is a block of stock prices and returns in my case.
Instead I want to compute the average across all cross section units for each time period identifier, and also I need to use a weight-- market value of the stock in my case. Maybe there is a 'unit' manipulation that would turn pmean into a function that averages over the cross-section for a time period, but that does not solve the weighting problem. I tried wmean(ret,mktval_lag) but it does not give what I expected. The weighted mean in this case is for an observation, not across observations. I could make the WLS trick a function but I am not sure how to loop over all dates (time identifiers) is an un-failable manner -- consistent with the smpl setting -- not just for my simple example, but in general. SAS's PROC SUMMARY and PROC MEANS does this, for example, by allowing a weight statement and a 'by' (group) statement. > 1. Re: Panel data time series means with weighting (Sven Schreiber) > 2. Re: Manipulating a database (Allin Cottrell) > 3. Re: Manipulating a database (Allin Cottrell) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Mon, 05 Apr 2010 14:57:22 +0200 > From: Sven Schreiber <svetosch(a)gmx.net> > Subject: Re: [Gretl-users] Panel data time series means with weighting > To: Gretl list <gretl-users(a)lists.wfu.edu> > Message-ID: <4BB9DE32.8020702(a)gmx.net> > Content-Type: text/plain; charset=ISO-8859-1 > > Hi, > first if your script does what you want, why try to "change a winning team"? > > Apart from that, for a weighted mean maybe the wmean() function in > combination with clever sample restrictions could do the job for you, > too. But what exactly do you mean with "time series means" -- I would > have thought the per-unit means of pmean() are just that. > > cheers, > sven > > MICHAEL BOLDIN schrieb: >> I would like to calculate time series means in a panel data set as >> easily as pmean() and wonder if there is a built in function or >> easier way than my script below. I am computing stock market returns >> for a portfolio where px=1 was set to select the included >> cross-sectional identifiers, and note I also need to weight (by market >> value in this case). >> >> smpl --full >> series retp= NA >> ## Loop over dates >> loop foreach dx 38748 38776 38807 >> smpl --full >> ## Set sample dates and selected ids >> smpl DATE= $dx --restrict >> smpl px=1 --restrict >> ## Compute weighted average returns as wls intercept coefficient >> wls mktval_lag ret const --quiet >> series retp=$coeff(0) >> endloop >> >> Second and third thing: I imported the data from EXCEL and the dates >> are numbers (1/1/1900=0), hence my loop over 38748 38776 38807 that >> are month end days. Is there a easy way to get all date ids to loop >> over and format as dates. >> _______________________________________________
