VC Database Rebuild

Plan

Rebuild roundbase, round, geo, ipos, mas from SDC data. Create companybase from roundbase Create round from roundbase. Build stageflags from round.

Clean companybase by putting flags for Undisclosed Company, US location. Check if key (coname, statecode, datefirstinv) is valid. Remove duplicates manually/update command from roundbase. Check if round key is valid. Remove duplicates.

Build statelookup tables and roundlookup tables.

Clean firmbase tables. Clean ipo tables. Clean mas table.

Run matcher on ipos, companybase. Matcher on mas, companybase. Fix duplicate matches.

Join ipos and companybase. Check if count is valid. Fix match as required. Pull ipo key into companybase and companybase key into ipo table first. Then join.

Join mas and companybase. Check if count is valid. Fix match as required. Pull mas key into companybase and companybase key into mas table first. Then join.

Join ipocompanybase with macompanybase to get a table of portcos, ipos and mas.

Calculate exit date based on ipo, ma, datelastinv + 5 years.

Pull in sel flag into companybase and build dead or alive flag.

Match geodata to companybase. Pull geokey into companybase table. Lookup addresses to get geo data as required using geo.py.

Clean fundbase and check valid key (fundname, statecode, firstinvdate)

Clean firmbase and check valid key (firmname, foundingdate)

Loading starting data into database

Database is named vcdb2. It is located in /bulk/VentureCapitalData/SDCVCData. Launch with psql vcdb2. Load the following tables by running the commands below. Make sure the sql scripts and data txt files are all located in the folder. Check that the line numbers copied into your new tables match the line numbers in the Load files.

\i LoadFunds.sql
\i LoadIPOs.sql
\i LoadRoundbase.sql
\i LoadFirms.sql
\i LoadGeoData.sql
\i LoadLongDescription.sql
\i LoadRound.sql

Cleaning Process

The roundbase table which is used to build the core company and round tables contains some data that we would like to remove like Undisclosed companies and duplicate entries. In order to find what to clean, build your companybase table first. You know your companybase table is clean once it contains a 1:1 relationship between keys and entries. We will then apply these changes to the roundbase table because any cleaning changes made downstream should be incorporated upstream into the base table. Otherwise when you build anything else off your roundbase table, dirty keys will infect the other areas of your database. Once the roundbase table is clean we will rename it roundbasecore so that we know it is clean and good to use for building other core tables.

Creating Base Tables

Create the base tables, companybase and round, by running the following scripts. These are the initial tables you will need to clean and join in order to get the master tables.

DROP TABLE companybase;
CREATE TABLE companybase AS
SELECT DISTINCT 
coname,updateddate,foundingdate,datelastinv,datefirstinv,investedk,city,description,msa,msacode,nationcode,statecode,addr1,addr2,indclass,indsubgroup3,indminor,url,zip 
FROM roundbase
ORDER BY coname;

DROP TABLE round;
CREATE TABLE round AS
SELECT DISTINCT coname,statecode,datefirstinv,rounddate,stage1,stage3,rndamtdisck,rndamtestk,roundnum,numinvestors
FROM roundbase
ORDER BY coname;

Cleaning the Companybase table

Every table will contain some duplicate keys and erroneous entries. We're going to clean the companybase table so that every key (coname, statecode, datefirstinv) is unique. This means that there will be a 1:1 relationship between 1 key and 1 entry. Given an entry you will be able to create a unique key and given a coname, statecode, datefirstinv key you will be able to find exactly 1 entry that the key corresponds to in the companybase table set.

So first check to see if the key is valid on the base data using the following 2 queries.

SELECT COUNT(*)
FROM (SELECT coname, statecode, datefirstinv FROM companybase)a;
--44774

SELECT COUNT(*)
FROM (SELECT DISTINCT coname, statecode, datefirstinv FROM companybase)a;
--44771

You can see that they key is not unique because the counts don't match up. There are 44,771 distinct keys but there are 44,774 keys in the companybase table. So 1 key can match to more than one entry in the table. Some of the data in the companybase table contains undisclosed company names and companies that exist in other countries outside the US. So let's build flags for these two events and check the key count again.

DROP TABLE companybase1;
CREATE TABLE companybase1 AS 
SELECT *,
CASE
 WHEN nationcode = 'US' THEN 1::int
 ELSE 0::int
 END AS alwaysusflag,
CASE
 WHEN coname = 'Undisclosed Company' THEN 1::int
 ELSE 0::int
 END AS undisclosedflag
FROM companybase;

SELECT COUNT(*)
FROM (SELECT coname, statecode, datefirstinv FROM companybase1 WHERE alwaysusflag = 1 AND undisclosedflag = 0)a;
--44771

SELECT COUNT(*)
FROM (SELECT DISTINCT coname, statecode, datefirstinv FROM companybase1 WHERE alwaysusflag = 1 AND undisclosedflag = 0)a;
--44770

By looking at the counts you can see that there is still 1 duplicate key in the table. Let's find it another way. Running the query below finds the key (coname, statecode, datefirstinv) that appears twice in the table.

SELECT *
FROM (SELECT coname, statecode, datefirstinv FROM companybase1 WHERE alwaysusflag = 1 AND undisclosedflag = 0)AS key
GROUP BY coname, statecode, datefirstinv
HAVING COUNT(key) > 1;

The output looks like this:

          coname            | statecode | datefirstinv
----------------------------+-----------+--------------
New York Digital Health LLC | NY        | 2015-08-13

We'll have to copy companybase1 out of the db and have a look on textpad for something unique about one of the entries on New York Digital Health LLC that we can use to manually delete it from the companybase1 table. Turns out the url is different so we'll use that. Manually delete this record from the roundbase table using the below command. Now we're ready to build the companybasecore table.

DELETE FROM roundbase WHERE coname = 'New York Digital Health LLC' AND statecode = 'NY' AND datefirstinv = to_date('2015-08-13', 'YYYY-MM-DD') AND url = 'www.digitalhealthaccelerator.c';

companybasecore table

The queries below build your companybasecore table. The where clause takes the place of the 2 flags on nationcode and undisclosed company we built in companybase1 table. This table has a guaranteed 1:1 relationship between coname, statecode, datefirstinv and an entry in the table. The two queries at the end verify this. We use core tables to run joins later on.

DROP TABLE companybasecore;
CREATE TABLE companybasecore AS
SELECT DISTINCT 
coname,updateddate,foundingdate,datelastinv,datefirstinv,investedk,city,description,msa,msacode,nationcode,statecode,addr1,addr2,indclass,indsubgroup3,indminor,url,zip
FROM roundbase WHERE nationcode = 'US' AND coname != 'Undisclosed Company';
--44740

--recheck keys
SELECT COUNT(*)
FROM (SELECT coname, statecode, datefirstinv FROM companybasecore)a;
--44740

SELECT COUNT(*)
FROM (SELECT DISTINCT coname, statecode, datefirstinv FROM companybasecore)a;
--44740

Cleaning ipos table

Check to see if the existing keys in the table are valid. We are using issuer, issuedate, statecode as the key.

SELECT COUNT(*)
FROM (SELECT issuer, issuedate, statecode FROM ipos)a;
--10440

SELECT COUNT(*)
FROM (SELECT DISTINCT issuer, issuedate, statecode FROM ipos)a;
--9491

The keys are not unique so we must remove duplicate keys first. You will need to try different methods to isolate the duplicate keys. This is where you can be creative. I first started by finding the duplicates based on issuer, issuedate and statecode which is our key. Have a look in textpad/excel for ways to filter these keys. We would like to save as much information as possible so rather than excluding all these entries which sum to 1888 rows in the ipos table maybe there's some other way we can filter out records and still have distinct keys.

DROP TABLE ipoduplicates;
CREATE TABLE ipoduplicates AS
SELECT *, COUNT(*)
FROM (SELECT issuer, issuedate, statecode FROM ipos)a
GROUP BY issuer, issuedate, statecode
HAVING COUNT(*) > 1;
--939
\COPY ipoduplicates TO 'ipoduplicates.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV;

In the file you will notice that many keys contain different principalamts. Let's keep the MAX principal amount and throw out the same key that has a lower principalamt. This query is shown below.

DROP TABLE ipoinclude;
CREATE TABLE ipoinclude AS
SELECT issuer, issuedate, statecode, MAX(principalamt) AS principalamt
FROM ipos
GROUP BY issuer, issuedate, statecode;
--9470

Now use the ipoinclude table to create a ipocore table. Then check to see if this core table has unique keys so 1 key matches with 1 record. This is the defining characteristic of a core table.

DROP TABLE ipocore;
CREATE TABLE ipocore AS
SELECT ipos.issuer, ipos.issuedate, ipos.statecode
FROM ipos INNER JOIN ipoinclude ON ipos.issuer = ipoinclude.issuer AND ipos.issuedate = ipoinclude.issuedate AND 
ipos.statecode = ipoinclude.statecode AND ipos.principalamt = ipoinclude.principalamt;
SELECT COUNT(*)
FROM (SELECT DISTINCT issuer, issuedate, statecode FROM ipocore)a;

You should notice that the ipocore table count does not match the count of DISTINCT keys. This means there are still some duplicates. So I created another duplicate table.

DROP TABLE ipoduplicates2;
CREATE TABLE ipoduplicates2 AS
SELECT *, COUNT(*)
FROM (SELECT issuer, issuedate, statecode FROM ipocore)a
GROUP BY issuer, issuedate, statecode
HAVING COUNT(*) > 1;

Then I created DELETE statements for all these entries. I deleted them from the ipoinclude table which will prevent these keys from appearing in the ipocore table when you JOIN the ipos with ipoinclude table.

--manually remove bad keys
DELETE FROM ipoinclude WHERE issuer = 'PacTel Corp' AND statecode = 'CA';
--1
DELETE FROM ipoinclude WHERE issuer = 'Templeton Dragon Fund Inc' AND statecode = 'FL';
--1
DELETE FROM ipoinclude WHERE issuer = 'Sterling Commerce' AND statecode = 'TX';
--1
DELETE FROM ipoinclude WHERE issuer = 'Sothebys Holdings Inc' AND statecode = 'NY';
--1
DELETE FROM ipoinclude WHERE issuer = 'TD Waterhouse Group Inc' AND statecode = 'NY';
--1
DELETE FROM ipoinclude WHERE issuer = 'Berlitz International Inc' AND statecode = 'NJ';
--1
DELETE FROM ipoinclude WHERE issuer = 'Spain Fund Inc' AND statecode = 'NY';
--1
DELETE FROM ipoinclude WHERE issuer = 'Ultramar Corp' AND statecode = 'CT';
--1
DELETE FROM ipoinclude WHERE issuer = 'Goldman Sachs Group Inc' AND statecode = 'NY';
--1
DELETE FROM ipoinclude WHERE issuer = 'Fidelity Advisor Korea Fund' AND statecode = 'MA';
--1
DELETE FROM ipoinclude WHERE issuer = 'Euronet Services Inc' AND statecode = 'KS';
--1
DELETE FROM ipoinclude WHERE issuer = 'Emerging Markets Tele Fund Inc' AND statecode = 'NY';
--1
DELETE FROM ipoinclude WHERE issuer = 'FirstMiss Gold Inc' AND statecode = 'NV';
--1
DELETE FROM ipoinclude WHERE issuer = 'Templeton Vietnam Opportunitie' AND statecode = 'FL';
--1
DELETE FROM ipoinclude WHERE issuer = 'Hybridon Inc' AND statecode = 'MA';
--1
DELETE FROM ipoinclude WHERE issuer = 'Indonesia Fund Inc' AND statecode = 'NY';
--1
DELETE FROM ipoinclude WHERE issuer = 'OpenTV Corp' AND statecode = 'CA';
--2
DELETE FROM ipoinclude WHERE issuer = 'Scudder New Europe Fund' AND statecode = 'NY';
--2
DELETE FROM ipoinclude WHERE issuer = 'Austria Fund Inc' AND statecode = 'NY';  
--2

Now again JOIN your ipos table with your ipoinclude table and check the key count.

DROP TABLE ipocore;
CREATE TABLE ipocore AS
SELECT ipos.*
FROM ipos INNER JOIN ipoinclude ON ipos.issuer = ipoinclude.issuer AND ipos.issuedate = ipoinclude.issuedate AND 
ipos.statecode = ipoinclude.statecode AND ipos.principalamt = ipoinclude.principalamt;
--9470
SELECT COUNT(*)
FROM (SELECT DISTINCT issuer, issuedate, statecode FROM ipocore)a;
--9470

The counts line up so now you should have a clean ipocore table!

Cleaning mas table

Check to see if you have bad keys in the table. The row count of the table should match up with count of distinct keys based on targetname, targetstatecode, announceddate.

SELECT COUNT(*)
FROM mas;
--114890 
SELECT COUNT(*)
FROM (SELECT DISTINCT targetname, targetstatecode, announceddate FROM mas)a;
--114825

Great! The counts don't match so we'll have to clean the mas table. There is no obvious field to filter against with mas. So I inserted an id column in mas and took the MIN id for duplicate keys.

CREATE TABLE mas1 AS
SELECT *
FROM mas;
ALTER TABLE mas1 ADD COLUMN id SERIAL PRIMARY KEY;
ALTER TABLE mas ADD COLUMN id SERIAL PRIMARY KEY;

DROP TABLE masinclude;
CREATE TABLE masinclude AS
SELECT targetname, targetstatecode, announceddate, MIN(id) as id
FROM mas1
GROUP BY targetname, targetstatecode, announceddate;
--114825

DROP TABLE mascore;
CREATE TABLE mascore AS
SELECT mas.*
FROM mas INNER JOIN masinclude ON mas.id = masinclude.id; 
--114825

SELECT COUNT(*)
FROM (SELECT DISTINCT targetname, targetstatecode, announceddate FROM mascore)a;

The mas distinct key count match the total count of the table so therefore the mascore table is clean.

Name Based Matching companybase keys to mas keys

Before attempting to match companybasecore with mascore you need a clean table or you will get many errors in the matcher output file. Luckily the core tables should already contain distinct keys if you've followed the process. However running the matcher will still yield many errors. So we will filter the mas keys some more. The first thing is to remove mas keys (targetname, announceddate, targetstatecode) where the announceddate falls within the same week. Keep the key that has the minimum announceddate and discard the higher date. Shown below:

DROP TABLE maskeys;
CREATE TABLE maskeys AS
SELECT DISTINCT targetname, targetstatecode, announceddate
FROM mascore;
--114825

DROP TABLE maskeysmindates;
CREATE TABLE maskeysmindates AS
SELECT targetname, targetstatecode, MIN(announceddate) AS announceddate
FROM mascore
GROUP BY targetname, targetstatecode;
--113236

DROP TABLE maskeysdatewindow;
CREATE TABLE maskeysdatewindow AS
SELECT maskeys.*, maskeysmindates.announceddate as minanndate,
CASE WHEN maskeys.announceddate - INTERVAL '7 day' > maskeysmindates.announceddate OR maskeys.announceddate = 
maskeysmindates.announceddate THEN 1::int
ELSE 0::int
END AS dateflag
FROM maskeys LEFT JOIN maskeysmindates ON (maskeys.targetname = maskeysmindates.targetname AND 
maskeys.targetstatecode = maskeysmindates.targetstatecode);
--114825

The dateflag is 1 when the current key's announceddate is 1 week older than the minimum announced date or it is the minimum announceddate for that targetname, targetstatecode pair. If the announceddate is less than 1 week greater than the minimum announceddate for te targetname, targetstatecode pair, then it is 0.

CREATE TABLE maskeysdatefiltered AS
SELECT targetname, targetstatecode, announceddate
FROM maskeysdatewindow
WHERE dateflag = 1; 
--114794
\COPY maskeysdatefiltered TO 'maskeysdatefiltered.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV

Grab the portco keys from the companybasecore table:

DROP TABLE portcokeys;
CREATE TABLE portcokeys AS
SELECT DISTINCT coname, statecode, datefirstinv
FROM companybasecore;
--44740
\COPY portcokeys TO 'portcokeys.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV

Put the portcokeys and maskeysdatefiltered text files into the Matcher Input folder. For more instructions on how to run the Matcher see The Matcher (Tool)

Fixing Errors in the Matcher Output for portco and mas

You will still receive multiple warnings in the output.matched file. In Excel add flags to exclude if the announceddate < datefirstinv and another exclude flag if the datefirstinv = announceddate. Also add a warning flag if the Warning column is "Hall-Warning:Multiple". Then import this back into your db by creating a matcheroutput table.

DROP TABLE matcherportcomas;
CREATE TABLE matcherportcomas (
 warning varchar(100),
 file1coname varchar(100),
 file1statecode varchar(2),
 file1datefirstinv date,
 file2targetname varchar(100),
 file2targetstatecode varchar (2),
 file2announceddate date,
 excludeflag1 int,
 excludeflag2 int,
 warningflag int
);
\COPY matcherportcomas FROM 'matcheroutputportco-mas.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV
--9645

You've imported 9,645 matches into your matcher table in vcdb2 but if you run the query below you will get the number of "good" matches. These are matches that do not contain warnings, where the datefirstinv > announceddate for a merger/acquisition and where the datefirstinv does not equal the announceddate.

SELECT COUNT(*) FROM 
(SELECT file1coname, file1statecode, file2targetname, file2targetstatecode FROM matcherportcomas WHERE excludeflag1 = 0 AND 
excludeflag2 = 0 AND warningflag = 0)a;  
--8291

As you can see we're throwing out a lot of the data in the matcher file (9645 -> 8291). So the next few queries will try and save as much of the bad matches as possible and add them back to the good matches to create our matcherportcomascore table.

Select the portco keys that are matched to the minimum announceddate for any mergers:

DROP TABLE matcherwarningmindates;
CREATE TABLE matcherwarningmindates AS
SELECT file1coname, file1statecode, file1datefirstinv, MIN(file2announceddate) 
FROM matcherportcomas 
WHERE excludeflag1 = 0 AND excludeflag2 = 0 AND warningflag = 1
GROUP BY file1coname, file1statecode, file1datefirstinv;
--364

Then using the temporary key (file1coname, file1statecode, file1datefirstinv, file2announceddate) join this back to the original matcher table to get the rest of the data we will want in the core table.

DROP TABLE matcherportcomasinclude;
CREATE TABLE matcherportcomasinclude AS
SELECT m.* FROM
matcherportcomas AS m INNER JOIN matcherwarningmindates AS mi ON m.file1coname = mi.file1coname AND m.file1statecode = 
mi.file1statecode AND m.file1datefirstinv = mi.file1datefirstinv AND m.file2announceddate = mi.min  
WHERE excludeflag1 = 0 AND excludeflag2 = 0 AND warningflag = 1;
--366

The inner join result should equal the amount in the matcherwarningmindates table but it doesn't. So to find the dirty entries we'll use the query below.

SELECT *, COUNT(*) FROM
(SELECT file1coname, file1statecode, file1datefirstinv FROM matcherportcomasinclude)a
GROUP BY file1coname, file1statecode, file1datefirstinv
HAVING COUNT(*) > 1;

file1coname       | file1statecode | file1datefirstinv | count
-------------------------+----------------+-------------------+-------
 PA Inc                  | TX             | 2007-09-25        |     2
 High Sierra Energy L.P. | CO             | 2004-12-23        |     2

Find these records in the matcherportcomas table in Excel and delete 1 entry from each manually:

DELETE FROM matcherportcomasinclude WHERE file1coname = 'PA Inc' AND file1statecode = 'TX' AND file2targetname = 'PA Corp' 
AND file2targetstatecode = 'VA';
--1
DELETE FROM matcherportcomasinclude WHERE file1coname = 'High Sierra Energy L.P.' AND file1statecode = 'CO' AND 
file2targetname = 'High Sierra Energy GP LLC' AND file2targetstatecode = 'CO';
--1

Now we should have a clean matcherportcomasinclude table. To be sure check the number of distinct matches using the query below. It should be the same as the number of records in this table.

SELECT COUNT(*) FROM
(SELECT DISTINCT file1coname, file1statecode, file1datefirstinv FROM matcherportcomasinclude)a;
--364
SELECT COUNT(*) FROM matcherportcomasinclude;
--364

Looks good so let's UNION ALL to join the matcherportcomasinclude table with the matcherportcomas with all flags set to 0 to create the core table.

CREATE TABLE matcherportcomascore AS
SELECT *
FROM matcherportcomas  WHERE excludeflag1 = 0 AND excludeflag2 = 0 AND warningflag = 0
UNION ALL
SELECT *
FROM matcherportcomasinclude;
--8655

Recheck the key counts. 1 portco key from the companybase table should match with exactly 1 mas key from the mascore table. If you have more than 1:1 you will get errors in the next phase when you join the companybase table to the mas table.

SELECT COUNT(*) FROM (
SELECT DISTINCT file1coname, file1statecode, file1datefirstinv
FROM matcherportcomascore) AS foo;
--8655

Great! Now you are ready to begin joining the companybase table to the mas table.

Joining companybasekeys with maskeys and ipokeys

Before doing this stage make sure the following is true:

1. companybasecore, mascore, ipocore are clean core tables...They should be 1:1 on themselves. That means 1 key should match to one row in each respective table. See Cleaning the Companybase table Cleaning mas table Cleaning ipos table for instructions
2. You've done name based matching on the keys in companybasecore and mascore and cleaned up the matcher output file. See Name Based Matching companybase keys to mas keys and Fixing Errors in the Matcher Output for portco and mas
3. You've done name based matching on the keys in companybasecore and ipocore and cleaned up the matcher output file. See Name Based Matching companybase keys to ipo keys and Fixing Errors in the Matcher Output for portco and ipo

We want to join the three sets of keys together before grabbing other data from their respective tables because there will be collisions with the maskeys and ipokeys. Some companies will have ipos as well as mergers/acquisitions or the data might also be miss coded by SDC platinum. The problem for us is a company that has both an ipo and ma will cause our join row counts to increase every time we join with these duplicate keys. We want a portcokey to join with only one ipokey or maskey in our master table. Running the query below creates a table that contains the three sets of keys:

DROP TABLE companybasekeysaddmaskeysaddipokeys;
CREATE TABLE companybasekeysaddmaskeysaddipokeys AS
SELECT c.coname, c.statecode, c.datefirstinv, matcherm.file2targetname AS mastargetname, matcherm.file2targetstatecode AS masstatecode, 
matcherm.file2announceddate AS announceddate, matcheri.file2issuer AS ipoissuer, matcheri.file2statecode AS ipostatecode, 
matcheri.file2issuedate AS ipoissuedate  FROM
companybasecore AS c LEFT JOIN matcherportcomascore as matcherm ON c.coname = matcherm.file1coname AND c.statecode = 
matcherm.file1statecode AND c.datefirstinv = matcherm.file1datefirstinv
LEFT JOIN matcherportcoipocore AS matcheri ON c.coname = matcheri.file1coname AND c.statecode = matcheri.file1statecode AND 
c.datefirstinv = matcheri.file1datefirstinv; 
--44740
\COPY companybasekeysaddmaskeysaddipokeys TO 'companybasekeysaddmaskeysaddipokeys.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV

Open in Excel and add a flag to see the rows with an ipokey as well as a maskey. You can use a formula like this: =IF(OR(ISBLANK(G518),ISBLANK(D518)),0,1). You'll see there are 83 portcokeys that match to an ipokey and a maskey. We'll write a query in sql to take the ipokey or maskey with the lowest date attached to it. This will be the exit date for that portco. First we create a table that has the minimum exit date. Then we add flags to indicate when the ipokey is valid and when the maskey is valid. Then we create a companybasekeymaskeyipokeycore table that contains clean matches from companybasekey (portcokey) to ipo or mas.

DROP TABLE companybasekeysaddmaskeyaddipokeysmindate; 
CREATE TABLE companybasekeysaddmaskeyaddipokeysmindate AS
SELECT *, 
CASE
 WHEN announceddate IS NOT NULL AND ipoissuedate IS NOT NULL THEN LEAST(announceddate,ipoissuedate)
 WHEN announceddate IS NOT NULL THEN announceddate
 WHEN ipoissuedate IS NOT NULL THEN ipoissuedate
 END AS masterdate
FROM companybasekeysaddmaskeysaddipokeys;
--44740
\COPY companybasekeysaddmaskeyaddipokeysmindate TO 'companybasekeysaddmaskeyaddipokeysmindate.txt' WITH DELIMITER AS E'\t' HEADER NULL 
AS  CSV

DROP TABLE companybasekeysaddmaskeyaddipokeysmindateflag;
CREATE TABLE companybasekeysaddmaskeyaddipokeysmindateflag AS
SELECT keys.*,
CASE
 WHEN announceddate = masterdate THEN 1::int
 ELSE 0::int
 END AS maskeyvalid,
CASE
 WHEN ipoissuedate = masterdate THEN 1::int
 ELSE 0::int
 END AS ipokeyvalid
FROM companybasekeysaddmaskeyaddipokeysmindate as keys;
--44740

Now create the companybaseipokeycore and companybasemaskeycore tables using the flags created above.

DROP TABLE companybasekeymaskeycore;
CREATE TABLE companybasekeymaskeycore AS
SELECT c.coname, c.statecode, c.datefirstinv, c.mastargetname, c.masstatecode, c.announceddate
FROM companybasekeysaddmaskeyaddipokeysmindateflag AS c
WHERE maskeyvalid = 1;
--8610 

DROP TABLE companybasekeyipokeycore;
CREATE TABLE companybasekeyipokeycore AS
SELECT c.coname, c.statecode, c.datefirstinv, c.ipoissuer, c.ipostatecode, c.ipoissuedate
FROM companybasekeysaddmaskeyaddipokeysmindateflag AS c
WHERE ipokeyvalid = 1;
--2312

To check if you have the correct number of ipo and mas keys add the two counts from your query above and the count from the query below and compare it to the number of keys in your companybasemaskeycore and companybaseipocore table. In my case I get 8610 + 2312 + 83 = 2350 + 8655.

SELECT COUNT(*)
FROM companybasekeysaddmaskeysaddipokeys
WHERE ipoissuedate IS NOT NULL AND announceddate IS NOT NULL;
--83

Now you can successfully join the companybasecore table to the ipocore and mascore tables through the companybasekeyipokeycore and companybasekeymaskeycore tables. With this step done you can create a master table which will contain information from companybase and ipo and mas.

Creating companybaseipomasmaster table

Before doing this stage make sure you have followed the steps in Joining companybasekeys with maskeys and ipokeys You will be joining the companybasecore table with the mascore and ipocore through the companybasekeyipokey and companybasekeymaskey tables. The output master table will have each company name and the dates and amounts if they received a ipo or ma. As discussed in Joining companybasekeys with maskeys and ipokeys the master table includes the exit deal which had the minimum date so duplicate rows should not crop up in the master table.

DROP TABLE companybaseipomasmaster;
CREATE TABLE companybaseipomasmaster AS
SELECT c.coname, c.statecode, c.datefirstinv, c.investedk, c.city, c.addr1, c.addr2, ipokey.ipoissuedate, maskey.announceddate AS 
masannounceddate, i.principalamt AS ipoprincipalamtk, m.transactionamt AS mastransactionamtk
FROM companybasecore AS c  
LEFT JOIN companybasekeyipokeycore AS ipokey ON c.coname = ipokey.coname AND c.statecode = ipokey.statecode AND c.datefirstinv = 
ipokey.datefirstinv  
LEFT JOIN companybasekeymaskeycore AS maskey ON c.coname = maskey.coname AND c.statecode = maskey.statecode AND c.datefirstinv = 
maskey.datefirstinv
LEFT JOIN ipocore AS i ON i.issuer = ipokey.ipoissuer AND i.issuedate = ipokey.ipoissuedate AND i.statecode = ipokey.ipostatecode
LEFT JOIN mascore AS m ON m.targetname = maskey.mastargetname AND m.targetstatecode = maskey.masstatecode AND m.announceddate = 
maskey.announceddate; 
--44740
\COPY companybaseipomasmaster TO 'companybaseipomasmaster.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV

You can run checks on the ipo and mas counts to make sure everything joined properly. Any duplicate keys that were not cleaned up in previous steps will make this master table a complete mess due to all the joins so make sure you've followed the process fully. Below are some of the checks I ran:

SELECT COUNT(*) FROM companybaseipomasmaster WHERE masannounceddate IS NOT NULL; 
--8610
SELECT COUNT(*) FROM companybaseipomasmaster WHERE mastransactionamtk IS NOT NULL; 
--8610
SELECT COUNT(*) FROM companybaseipomasmaster WHERE ipoissuedate IS NOT NULL; 
--2312
SELECT COUNT(*) FROM companybaseipomasmaster WHERE ipoprincipalamtk IS NOT NULL; 
--2312

Everything looks good. These counts are compared against the key tables and core tables built in the previous steps.

Name Based Matching companybase keys to ipo keys

First verify that your keys in companybasecore and ipocore are unique by using the following queries. If not following instructions in these sections Cleaning the Companybase table and Cleaning ipos table

SELECT COUNT(*) FROM companybasecore;
--44740
SELECT COUNT(*) FROM
(SELECT DISTINCT coname, statecode, datefirstinv FROM companybasecore)a;
--44740
SELECT COUNT(*) FROM ipocore;
--9470
SELECT COUNT(*) FROM
(SELECT DISTINCT issuer, issuedate,statecode FROM ipocore)a;
--9470

Next export the keys to a text file and put in the Matcher input folder. Run the matcher on these files. For instructions on how to use the Matcher check this out The Matcher (Tool)

DROP TABLE portcokeys;
CREATE TABLE portcokeys AS
SELECT DISTINCT coname, statecode, datefirstinv
FROM companybasecore;
--44740
\COPY portcokeys TO 'portcokeys.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV

DROP TABLE ipokeys;
CREATE TABLE ipokeys AS
SELECT issuer, statecode, issuedate
FROM ipocore;
--9470
\COPY ipokeys TO 'ipokeys.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV

Fixing Errors in the Matcher Output for portco and ipo

After running the Matcher on your portcokeys and ipokeys you will notice there are some errors in the matched output file. Add flags in Excel that exclude rows where the issuedate < datefirstinv and where the issuedate = datefirstinv. If the exclude flag is 1 than you would want to exclude this entry from your table i.e. the issuedate > datefirstinv. If the flags are selected to 0, then you will want to keep this row. Also add a column for a warning flag that is 1 if the warning column is "Hall-Warning:Multiple". Next copy this txt file into the db by creating a new table.

DROP TABLE matcherportcoipo;
CREATE TABLE matcherportcoipo (
 warning varchar(100),
 file1coname varchar(100),
 file1statecode varchar(2),
 file1datefirstinv date,
 file2issuer varchar(100),
 file2statecode varchar (2),
 file2issuedate date,
 excludeflag1 int,
 excludeflag2 int,
 warningflag int
);
\COPY matcherportcoipo FROM 'matcherportco-ipos.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV
--2592

You can see the "good" matches by setting all the flags to 0 as shown in the query below.

SELECT COUNT(*)
FROM matcherportcoipo WHERE excludeflag1 = 0 AND excludeflag2 = 0 AND warningflag = 0;
--2313

We would like to add back all the data we can so let's have a look at the rows with multiple matches.

SELECT COUNT(*)
FROM matcherportcoipo WHERE excludeflag1 = 0 AND excludeflag2 = 0 AND warningflag = 1;
--66

Many of the duplicates have different issuedates so we'll just select the minimum issuedate for entries where the portcokey is matched twice.

DROP TABLE matcherportcoipomindate;
CREATE TABLE matcherportcoipomindate AS
SELECT file1coname, file1statecode, file1datefirstinv, MIN(file2issuedate)
FROM matcherportcoipo
WHERE excludeflag1 = 0 AND excludeflag2 = 0 AND warningflag = 1
GROUP BY file1coname, file1statecode, file1datefirstinv;
--37

Then we can create an include table and union this with the good matches to create a matcher core file for portco and ipos.

CREATE TABLE matcherportcoipoinclude AS
SELECT m.* FROM
matcherportcoipo AS m JOIN matcherportcoipomindate AS mi ON m.file1coname = mi.file1coname AND m.file1statecode = mi.file1statecode AND 
m.file1datefirstinv = mi.file1datefirstinv AND m.file2issuedate = mi.min; 
--37

And create a matcherportcoipocore table by combining the good matches with the fixed mismatches.

CREATE TABLE matcherportcoipocore AS
SELECT *
FROM matcherportcoipo WHERE excludeflag1 = 0 AND excludeflag2 = 0 AND warningflag = 0
UNION ALL
SELECT *
FROM matcherportcoipoinclude;
--2350

Now verify that the key counts are correct. The number of distinct portco keys should equal the number of rows in the core table.

SELECT COUNT(*) FROM (
 SELECT DISTINCT file1coname, file1statecode, file1datefirstinv FROM matcherportcoipocore)a;
--2350

Boom the matcherportcoipocore table is clean and good for use.

Cleaning geo table

The geo table contains duplicate keys. The key for the geo table is (coname, city, startyear). Look at the different counts for all keys and distinct keys from the table:

SELECT COUNT(*) FROM (SELECT DISTINCT city, coname, startyear FROM geo)a; 
--43651
SELECT COUNT(*) FROM geo;
--43724
SELECT *, COUNT(*) 
FROM (SELECT city, coname, startyear FROM geo)a
GROUP BY city, coname, startyear
HAVING COUNT(*) > 1;

If you look at the rows with duplicate keys you can see they are simply complete duplicates so let's create a table with just distinct rows.

DROP TABLE geo1;
CREATE TABLE geo1 AS
SELECT DISTINCT * 
FROM geo; 
--43662 

We still have 11 keys that are not distinct. We'll need to clean those up.

SELECT *, COUNT(*) 
FROM (SELECT city, coname, startyear FROM geo1)a
GROUP BY city, coname, startyear
HAVING COUNT(*) > 1;
--8
    city      |           coname            | startyear | count
--------------+-----------------------------+-----------+-------
New York      | New York Digital Health LLC |      2015 |     2
Portland      | Undisclosed Company         |      2016 |     2
Hauppauge     | Mdeverywhere Inc            |      1999 |     2
North Mankato | Angie's Artisan Treats LLC  |      2011 |     2
Cincinnati    | Undisclosed Company         |      2016 |     4
New York      | Undisclosed Company         |      2015 |     2
San Francisco | Undisclosed Company         |      2016 |     2
San Francisco | Undisclosed Company         |      2015 |     3

Modify geo1 table query to get rid of Undisclosed Companies:

DROP TABLE geo1;
CREATE TABLE geo1 AS
SELECT DISTINCT * 
FROM geo
WHERE coname NOT LIKE '%Undisc%';
--43631
SELECT *, COUNT(*) 
FROM (SELECT city, coname, startyear FROM geo1)a
GROUP BY city, coname, startyear
HAVING COUNT(*) > 1;
--3

Now manually check the longitude and latitude of each of these rows and delete one of each of them. Then create your core table and verify that all the keys are distinct.

DELETE FROM geo1 WHERE coname = 'New York Digital Health LLC' AND city = 'New York' AND startyear = 2015 AND lattitude = 44.933143::real AND longitude = 7.540121::real;
--1
DELETE FROM geo1 WHERE coname = 'Mdeverywhere Inc' AND city = 'Hauppauge' AND endyear = 2011;
--1
DELETE FROM geo1 WHERE city = 'North Mankato' AND lattitude = 44.19030721::real AND longitude = -94.052706::real;
--1
CREATE TABLE geocore AS
SELECT *
FROM geo1;
--43628
SELECT COUNT(*) FROM (SELECT DISTINCT city, coname, startyear FROM geocore)a; 
--43628

Name Based Matching geo keys to companybase keys

Get a list of geokeys and companybasekeys and run them through the The Matcher. The key is (coname, city, startyear) so you'll need to extract the year from the datefirstinv from the companybasecore table. See below.

DROP TABLE geokeys;
CREATE TABLE geokeys AS
SELECT coname, city, startyear
FROM geocore
WHERE noaddress = 0::boolean;
--33628
\COPY geokeys TO 'geokeys.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV 

CREATE TABLE portcokeysforgeo AS
SELECT coname, city, EXTRACT(YEAR FROM datefirstinv)
FROM companybasecore;
--44740
\COPY portcokeysforgeo TO 'portcokeysforgeo.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV

After you run the matcher you will notice there are a ton of matching errors as usual. If you simply import this into your vcdb2 and try joining companybasecore with geocore your tables will start to explode. Notice how the line count jumps from 44,740 to 45,018.

DROP TABLE matcherportcogeo;
CREATE TABLE matcherportcogeo (
 portcoconame varchar(255),
 portcocity varchar(100),
 portcostartyear integer,
 geoconame varchar(255),
 geocity varchar(100),
 geodatefirstyear integer
);
\COPY matcherportcogeo FROM 'matcheroutputportcogeo.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV
--33608    

--try matching companybase to geo through the matcherportcogeo
CREATE TABLE companybasecorejoingeo AS
SELECT c.coname, c.statecode, c.datefirstinv, c.investedk, c.city, c.addr1, c.addr2, g.lattitude, g.longitude 
FROM companybasecore c
LEFT JOIN matcherportcogeo AS m ON m.portcoconame = c.coname AND m.portcocity = c.city AND m.portcostartyear = EXTRACT(YEAR FROM 
c.datefirstinv)
LEFT JOIN geocore AS g ON g.coname = m.geoconame AND m.geocity = g.city AND m.geodatefirstyear = g.startyear; 
--45018

Okay so we need to fix this. Luckily I already had this data in another database so I copied it out and imported it into vcdb2. The raw data can be found in a text file in the folder on the Z drive called geolookupold.txt.

CREATE TABLE geoimport (
 coname varchar(100),
 statecode varchar(2),
 datefirstinv date,
 latitude real,
 longitude real
);
\COPY geoimport FROM 'geolookupold.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV
--42678  

SELECT COUNT(*) FROM
(SELECT DISTINCT coname, statecode, datefirstinv FROM geoimport)a;
--42678

CREATE TABLE companybasegeomaster AS
SELECT c.coname, c.statecode, c.datefirstinv, c.investedk, c.city, c.addr1, c.addr2, g.latitude, g.longitude
FROM companybasecore AS c 
LEFT JOIN geoimport AS g ON c.coname = g.coname AND c.statecode = g.statecode AND c.datefirstinv = g.datefirstinv;
--44740
\COPY companybasegeomaster TO 'companybasegeomaster.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV

Gathering geo data from company addresses

If you do not already have a file with all the geo data in it you can lookup the latitude, longitude data from google using the company address. A link on how to use the Geocode.py is found here. When you copy the addresses out of the database be sure to include a distinct key that will allow you to join the geo data back with the portco key.

Build dead/alive flags

First find the deaddate for each company. Make sure that you have companybasecore, ipocore, mascore tables. Then calculate a deaddate. If there is no exit then the deaddate is datelastinv + 5 years. Take a look at the queries below.

CREATE TABLE deaddate AS
SELECT c.coname, c.statecode, c.datefirstinv, c.datelastinv, i.issuedate, m.announceddate
FROM companybasecore AS c
LEFT JOIN companybasekeyipokeycore AS ipokey ON c.coname = ipokey.coname AND c.statecode = ipokey.statecode AND c.datefirstinv = 
ipokey.datefirstinv  
LEFT JOIN companybasekeymaskeycore AS maskey ON c.coname = maskey.coname AND c.statecode = maskey.statecode AND c.datefirstinv = 
maskey.datefirstinv
LEFT JOIN ipocore AS i ON i.issuer = ipokey.ipoissuer AND i.issuedate = ipokey.ipoissuedate AND i.statecode = ipokey.ipostatecode
LEFT JOIN mascore AS m ON m.targetname = maskey.mastargetname AND m.targetstatecode = maskey.masstatecode AND m.announceddate = 
maskey.announceddate; 
--44740

CREATE TABLE deaddate1 AS
SELECT *,
CASE 
WHEN issuedate IS NULL AND announceddate IS NULL THEN datelastinv + INTERVAL '5 YEAR'
WHEN issuedate IS NOT NULL THEN issuedate
WHEN announceddate IS NOT NULL THEN announceddate 
END AS deaddate
FROM deaddate;
--44740

You will need to run the queries below to build out a master table that has dead and alive flags and the company counts for each year in the database by datefirstinv.

CREATE TABLE stageflagscore AS
SELECT *,
CASE WHEN seedflag = 1 OR earlyflag = 1 OR laterflag = 1 THEN 1::int ELSE 0::int
END AS selflag
FROM stageflags;
--143347

DROP TABLE selcos;
CREATE TABLE selcos AS
SELECT DISTINCT coname, statecode, datefirstinv, selflag
FROM stageflagscore
WHERE excludeflag = 0 AND selflag = 1;
--32597 

DROP TABLE deadalive;
CREATE TABLE deadalive AS
SELECT deaddate1.*, sel.selflag
FROM deaddate1 LEFT JOIN selcos AS sel ON deaddate1.coname = sel.coname AND deaddate1.statecode = sel.statecode AND 
deaddate1.datefirstinv = sel.datefirstinv;
--44740

--match to sel flag
DROP TABLE deadalivesel;
CREATE TABLE deadalivesel AS
SELECT da.*, flags.stage3, flags.seedflag, flags.earlyflag, flags.laterflag, flags.growthflag, flags.transactionflag, flags.excludeflag
FROM deadalive AS da LEFT JOIN stageflags AS flags ON da.coname = flags.coname AND da.statecode = flags.statecode AND da.datefirstinv = 
flags.datefirstinv;
--143310

CREATE TABLE deadalive1 as
SELECT coname, city, statecode, datefirstinv, datelastinv, deaddate, extract(year from datefirstinv) as aliveyear,
extract(year from deaddate) AS deadyear
FROM deadalive WHERE selflag=1;
--32575

DROP TABLE tempbase;
CREATE TABLE tempbase As
SELECT DISTINCT year, coname, city, statecode
FROM allyears
JOIN deadalive1 ON year>=extract(year from datefirstinv) AND year<=deadyear;
--239446

DROP TABLE alivecount;
CREATE TABLE alivecount AS
SELECT city, statecode, year, count(coname) as numalive FROM tempbase
GROUP BY city, statecode, year ORDER by count(coname) DESC;
--42296
\COPY alivecount TO 'alivecount.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV

Creating coleveloutput

One of the output tables required by the other researchers is the coleveloutput table. It contains company, geo and ipo/ma details in the form of aliveyear, deadyear. Here's how you build it:

DROP TABLE SelFlagBase;
CREATE TABLE SelFlagBase AS
SELECT DISTINCT coname, statecode, datefirstinv from stageflags where growthflag=1;
--32597

DROP TABLE companybasecore2;
CREATE TABLE companybasecore2 AS
SELECT companybasecore.*,
CASE WHEN SELFlagbase.coname IS NOT NULL THEN 1::int ELSE 0::int END AS hadgrowthvc
FROM companybasecore
LEFT JOIN SelFlagBase ON SelFlagBase.coname=companybasecore.coname AND SelFlagBase.statecode=companybasecore.statecode AND 
SelFlagBase.datefirstinv=companybasecore.datefirstinv;
--44740

SELECT COUNT(*) FROM companybasecore2 WHERE hadgrowthvc=1;
--32575

DROP TABLE coleveloutput;
CREATE TABLE coleveloutput AS
SELECT companybasecore2.coname, companybasecore2.statecode, companybasecore2.datefirstinv, companybasecore2.city, 
companybasecore2.addr1, companybasecore2.addr2, companybasecore2.zip, g.latitude, g.longitude, d.deaddate, d.aliveyear, d.deadyear
FROM companybasecore2  
LEFT JOIN deadalive1 AS d ON d.coname=companybasecore2.coname AND d.statecode=companybasecore2.statecode AND 
d.datefirstinv=companybasecore2.datefirstinv
LEFT JOIN geoimport AS g ON g.coname = companybasecore2.coname AND g.statecode = companybasecore2.statecode AND g.datefirstinv = 
companybasecore2.datefirstinv
WHERE hadgrowthvc=1;
--32575
\COPY coleveloutput TO 'coleveloutput.txt' WITH DELIMITER AS E'\t' HEADER NULL AS  CSV

Creating roundplus

DROP TABLE roundplus;
CREATE TABLE roundplus AS
SELECT roundcore.*, c.city, seedflag, earlyflag, laterflag, growthflag, transactionflag, excludeflag,
CASE WHEN roundcore.datefirstinv=roundcore.rounddate THEN 1::int ELSE 0::int END as dealflag,
CASE WHEN SELFlagbase.coname IS NOT NULL THEN 1::int ELSE 0::int END AS hadgrowthvc,
extract(year from roundcore.rounddate) as roundyear,
CASE WHEN rndamtdisck IS NOT NULL THEN rndamtdisck/1000 WHEN rndamtdisck IS NULL AND rndamtestk IS NOT NULL THEN rndamtestk/1000 ELSE 
NULL::real END as roundamtm
FROM roundcore
LEFT JOIN SelFlagBase ON SelFlagBase.coname=roundcore.coname AND SelFlagBase.statecode=roundcore.statecode AND 
SelFlagBase.datefirstinv=roundcore.datefirstinv
LEFT JOIN stageflags ON  stageflags.coname=roundcore.coname AND stageflags.statecode=roundcore.statecode AND 
stageflags.datefirstinv=roundcore.datefirstinv AND stageflags.rounddate=roundcore.rounddate
LEFT JOIN companybasecore AS c ON c.coname = roundcore.coname AND c.statecode = roundcore.statecode AND c.datefirstinv = 
roundcore.datefirstinv;
--143001

SELECT coname, rounddate FROM (SELECT coname, rounddate FROM roundplus)a
GROUP BY coname, rounddate
HAVING COUNT(*) > 1;

DELETE FROM roundplus WHERE coname = 'New York Digital Health LLC';
--2

Creating round level outputs

roundplus is used to build the two table outputs below at the round level.

DROP TABLE roundleveloutput;
CREATE TABLE roundleveloutput AS
SELECT city, statecode, roundyear as year, 
sum(roundamtm*seedflag) AS seedamtm,
sum(roundamtm*earlyflag) AS earlyamtm,
sum(roundamtm*laterflag) AS lateramtm,
sum(roundamtm*growthflag) AS selamtm,
sum(seedflag) AS numseeds,
sum(earlyflag) AS numearly,
sum(laterflag) AS numlater,
sum(growthflag) AS numsel,
sum(dealflag) AS numdeals
FROM roundplus WHERE hadgrowthvc=1 GROUP BY city, statecode, roundyear ORDER BY city, statecode, roundyear;
--22266  

DROP TABLE roundleveloutput2;
CREATE TABLE roundleveloutput2 AS 
SELECT roundleveloutput.*, numalive
FROM roundleveloutput
LEFT JOIN alivecount ON alivecount.city=roundleveloutput.city AND alivecount.statecode=roundleveloutput.statecode AND 
alivecount.year=roundleveloutput.year;
--22266

Cleaning round table

Use coname, rounddate as the key for this table. Exclude all keys that occur more than once.

CREATE TABLE roundexclude AS
SELECT * FROM (
SELECT coname, rounddate FROM round) t
GROUP BY coname, rounddate
HAVING COUNT(*) > 1;
--154

CREATE TABLE roundcore AS
SELECT * FROM round
WHERE NOT EXISTS (SELECT * FROM roundexclude AS re WHERE re.coname = round.coname AND re.rounddate = round.rounddate); 
--143000

Creating Stage Flags Table

Stage flags will be used to later on to determine if a company received seed, early or later stage financing. The growthflag is '1' if either the seed, early or later flags is '1'. The exclude flag is used to exclude all companies that received financing for activities we are not interested in and thus should be excluded from our dataset. Entries like 'Open Market Purchase', 'PIPE', etc are the things that the exclude flag filters out. It is built off the round table.

DROP TABLE stageflags;
CREATE TABLE stageflags AS
SELECT coname, statecode, datefirstinv, rounddate, stage3,
CASE
 WHEN stage3 = 'Seed' THEN 1::int
 ELSE 0::int
END AS seedflag,
CASE
 WHEN stage3 = 'Early Stage' THEN 1::int
 ELSE 0::int
 END AS earlyflag,
CASE
 WHEN stage3 = 'Later Stage' THEN 1::int
 ELSE 0::int
 END AS laterflag,
CASE
 WHEN stage3 = 'Seed' OR stage3 = 'Later Stage' OR stage3 = 'Early Stage' THEN 1::int
 ELSE 0::int
 END AS growthflag,
CASE
 WHEN stage3 = 'Acq. for Expansion' OR stage3 = 'Acquisition' OR stage3 = 'Bridge Loan' OR stage3 = 'Expansion' OR stage3 = 'Pending Acq' OR stage3 = 'Recap or Turnaround' OR stage3 = 'Mezzanine' THEN 1::int
 ELSE 0::int
 END AS transactionflag,
CASE
 WHEN stage3 = 'LBO' OR stage3 = 'MBO' OR stage3 = 'Open Market Purchase' OR stage3 = 'PIPE' OR stage3 = 'Secondary Buyout' 
OR stage3 = 'Other' OR stage3 = 'VC Partnership' OR stage3 = 'Secondary Purchase' THEN 1::int
 ELSE 0::int
 END AS excludeflag
FROM round;

Cleaning firmbase

The key I chose after trying a few different combinations is firmname, statecode, addr1.