7,627
edits
Changes
Jump to navigation
Jump to search
Import Now import the matches into the dbase and generate the patent statistics using PatentMeasures.sql. Note that it is not possible to generate the patent measures for all of Compustat -- the data is just too big.
Now there appear to be two unrelated technical issues. All A quick build of the patenting measures appear to suffer from data atrophy for non-LBOs. By relyear==7, we are down to 160/173 nongraphs is in MainTTests-LBOs having coverageReprocessed. In addition there appears to be an extreme problem with the pairgood variablexlsx. . tab year if patentstock==. Fiscal year | FreqMaking a chart template is a good idea [https://www. Percent Cumexcelcampus. com/charts/chart-templates-default----------+----------------------------------- 1975 | 173 11.11 11.11 1976 | 173 11.11 22.22 1977 | 173 11.11 33.33 1978 | 173 11.11 44.44 1979 | 173 11.11 55.56 2014 | 173 11.11 66.67 2015 | 173 11.11 77.78 2016 | 173 11.11 88.89 2017 | 173 11.11 100.00 ------------+----------------------------------- Total | 1,557 100.00 The problem seems formatting/] but it won't let you save everything you'd like to be a data construction do (especially it won't put on the errorbars automatically). It isnEd't just the patent variables, all variables, except the newly added patent measures, are missing for the period 1975 to 1979 and 2014 to 2017s chart template was saved in C:\Users\ed\AppData\Roaming\Microsoft\Templates\Charts\LBOPaper. crtx
The Effect of Leverage Buyouts on Innovation (view source)
Revision as of 15:59, 31 July 2019
, 15:59, 31 July 2019→Code
*LBOregmarker, 1/0 for the last year of CS data (-1 or -2)
*Somepatenting, 1/0 for patent(s) at some point in history
*matchfilter for whether the obs is in the match set(at -1)*regfilter is a legacy marker, for whether the obs is in the regif you don't want to move accounting variables from t-2 to t-1.
*savefilter for whether the obs will be saved for the analysis
The propensity score regression is:
. logit lboregmarker lbomatchmarker lvg lvgsq tobinql revgrowthl taxtoassets ebitdatoassets revtl if regfiltermatchfilter==1, robust
Iteration 0: log pseudolikelihood = -1157.9027
With the increased sample, we now find that LBOs are associated with reduced likelihood of patenting overall. However, this results is driven by the LBOs in period 1 from 1980 to 1995. In period 2, from 1995 to 2015, there is no statistically significant association. We also tried using decades but this wasn't instructive.
The matching scripts, LBOmatchingscript.jl and LBOmatching.jl, were rewritten by Ed to run with Julia 1.1.1. They use the LBOregmarker LBOmatchmarker and regfilter matchfilter -- we are matching at t-1 (or obs with logitpreg only at t-2have it, and the variables used to create it, copied from t-2 to t-1). Note that we only export if regfiltermatchfilter==1 to MatchInput2V1.txt. All 179 patenting LBOs are in there, however, not all of them have a logitpreg:
. count if hadlbo==1 & regfilter==1
173
This is because variables just aren't available to calculate them. This was explored and is the best it reasonably can be after applying some manual fixes (see below).
su tobinql if regfilter & hadlbo //175
su revgrowthl if regfilter & hadlbo //176
...all other vars have full coverage
We also pushed logitpreg from t-2 to t-1. This allows us to then match EVERYTHING at t-1 using contemporary variables, inluding patentstock.
To run the matching script:
We also adjusted the bounds on patenting to 25% (and rounded the result to an integer) in the match script(s). This gave 173 matches from 173 candidates (i.e. we caught one unmatched at the 20% bound). See below for more information about catching the one unmatched. We also adjusted the script so that it just outputs the matches and throws out everything else. This can be uncommented if needed.
In PatentMeasures.sql
*Originality is based on citations received and generality is based on citations made. There may have been an error in originality in the previous version (hard to tell - misnamed tables)
*The Match5yrCites table is big: 170,169,703 obs -- it is the explorative/exploitive measures that are very hard to build. The citation measures could likely be built for every CS firm 1974-present.
*Ultimately creates: MatchedMeasuresPortfolioV2-12.txt Make sure that there are copies of PatentMeasures.sql and matchresults.txt in Z:/lbo. Then. to run the SQL script, ssh on to mother (192.168.2.121) as researcher, and: cd /bulk/lbo psql lbo \i PatentMeasures.sql
Then run Analysis.do. The current versions of the datasets with their origins are:
*Datasetup.do -> MatchInput2V1MatchInput2V2.txt -> LBOmatchscript.jl -> matchresults.txt -> matchresultsV2-12.dta*Datasetup.do -> MatchInput2V1MatchInput2V2.txt -> LBOmatchscript.jl -> matchresults.txt -> PatentMeasures.sql (on dbase lbo)-> MatchedMeasuresPortfolioV2-12.txt -> PatentStatisticsV2-12.dta*Datasetup.do -> PreMergeRelevantV2-12.dta
*CPI.dta
*OldData.dta
Analysis.do was updated to take into account that the matches are made at t-1. Its preparatory code was also entirely rewritten. Markers for LBOs included in the old data are loaded (as olddata and olddataflag, at the pair level) for reference. None of the old matches are refound in the new data -- these are entirely new matches, likely as the logit specification was changed.
====Catching the one unmatched====
