CEO Compensation and Returns to Public and Private Acquisitions: Testing the Shareholder Value Hypothesis

From edegan.com
Jump to navigation Jump to search
Academic Paper
Title CEO Compensation and Returns to Public and Private Acquisitions: Testing the Shareholder Value Hypothesis
Author Ed Egan, Jim Brander
Status Working paper
© edegan.com, 2016

Working Paper

The working paper is called "CEO Compensation Full (Last Version).docx". It's in the dropbox (Dropbox\coauthoredprojects\ResearchWithJim\ExecCompAcqs). This produces the pdf below, which is also available on SSRN:

Brander, James A. and Egan, Edward and Endl, Sophie, Comparing CEO Compensation Effects of Public and Private Acquisitions (January 15, 2021). Available at SSRN: https://ssrn.com/abstract=3766568

Database, dataset and scripts

The database is compacqs

The SQL script and input data is in:

Z:\compacqs\

The SQL script is:

BuildCompAcqs.sql

The STATA do file is Master.do and is in:

E:\McNair\Projects\CompAcqs\

The current version of the dataset is:

MasterV1-1.txt

There's also a CEO.log file that details results run on the old dataset, for information only.

Current Notes

The paper was submitted to the Journal of Risk and Financial Management (JRFM), an open-access journal with fast turn-around times on January 17th, 2021. On February 8th, 2021, we received an R&R with a 10-day revision window. This section provides notes on the feasibility of the R&R.

A key issue concerns requests for us to update the data, which:

  • Isn't a live dataset on my current Research Computing Hardware.
  • Requires Cumulative Abnormal Returns (CARs) that were previously calculated as a part of the Winner's Curse in Acquisitions (Academic Paper) that was accepted for publication in 2017.
    • Uses both public and private acquisitions data, that were loaded as a part of VCDB20 but may need more variables.
  • Needs ExecComp, Compustat, and CRSP data from WRDS (I have access to all until March).

I didn't notice any requests from the reviewers to handle multiple acquisitions on the same day, which doesn't appear to have been fixed in the previous version. However, I can see we are aggregating acquisitions over the year, so this is somewhat moot in the averages and totals.

Reviewer Comments

Some of Reviewer 3s comments are problematic:

  • Another major problem with this empirical work is due to the problem of endogeneity reported by the authors. How was this problem resolved? The methodology used by the authors does not solve this problem! Why didn't the authors think about using the latest dynamic panel data models? There are some models who have emerged to precisely solve the problems that this sample presents. This should be explored by the authors.
    • Fixed effect panel data should address issue of unobserved heterogeneity.
  • On the other hand, the authors need to present to us in the tables the overall results of the tests to the models used. For example, they refer us to the Hausman Test but do not show us the results of this test! Wald tests, Test F, Fixed Effects Test F and the respective R-Square value for fixed effects regression and random effects regression are missing.
    • Just generate test stats!
  • Where is the correlation matrix? It must be presented and its coefficients analysed.
  • Finally, there is another problem in the work of these authors and the main objective of the paper and the fact that the authors want to understand the difference between so-called public and private acquisitions. At the beginning of the development of the empirical work the authors must demonstrate to us that in fact the two subsamples are different in terms of the variables that will be studied. Thus, differences tests should be performed and reported to these two subsamples, only by so that it will make sense to perform all the empirical work later.

Build Notes

I started a new database called wrds for the major pulls.

Ed To Do

  • Institutional investors measures
  • Profitability: profit and net income
  • MA activity 2016 to 2020
  • Take a look at the dynamic panel

Note see also: Governance Measures

New Measures

From Thomson-Reuters 13F filings (joined using CUSIP8 and year, where available):

  • iisharestot -- Total number of shares held by institutional investors
  • iisharestop5 -- Number of shares held by the five largest institutional investors
  • iitop5prop -- Proportion of shares held by the five largest institutional investors compared to all institutional investors
  • iicount -- Number institutional investors
  • blocktot -- Number of blockholders (institutional investors with >= 5% of shares)

From COMPUSTAT Fundamentals Annual (joined using CUSIP8 and year, where available):

  • gp -- gross profit
  • ni -- net income
  • gpr -- gross profit ratio
  • npr -- net profit ratio
  • csho -- number of shares outstanding (in millions)
  • revt -- revenue (note that this replaces the old revt variable)
  • roa -- return on assets (ebitda/totalassets)
  • leverage -- total liabilities/total assets
  • totalassets -- same as the old at variable but from a new pull
  • interlock -- 1/0 CEO is on compensation committee
  • mktcap -- Market Capitalization calculated as closing price (prcc_c) x common shares outstanding (csho)

Thomson-Reuters 13F filings in conjunction with COMPUSTAT Fundamentals Annual (joined using CUSIP8 and year, where both available):

  • iifractot -- fraction of shares held by institutional investors
  • iifractop5 -- fraction of shares held by top 5 institutional investors

Old Governance Measures

The following governance measures are listed under firm variables below:

  agovgoldenparachute - 0/1 for the CEO has a golden parachute (many missing)
  agovlimitabilityactbywritconsent - 0/1 whether governance has limited shareholders ability to act by written consent (many missing)
  agovmajvoterequirement - 0/1 for whether board members are elected on majority or plurality, which matters for uncontested seats (many missing)
  agovresignrequiredonmajorityvote - 0/1 "Indicates that a director is required to submit his/her resignation upon failing to receive support from a majority of votes cast (which, typically, the board may chose to accept or reject)." (many missing) 

I tried each of them in a FE panel regression of the type found in table 3, like:

areg tdc116l car5totpub car5totpriv agovresignrequiredonmajorityvote i.year, absorb(ceofirmid) cluster(ceofirmid)

In each case, the sample dropped to ~2k (from 41k), and there was no significance for ANY variable.

Reference Papers and their data

Hartzell Starks (2000) - Institutional Investors and Executive Compensation:

  • we obtain institutional equity holdings for each year between December 1991 and December 1996 from the CDA Spectrum database. CDA Spectrum derives these holdings from institutional investors’ 13-f filings. (Institutional investors with more than $100 million in equities must report their equity ownership to the SEC in quarterly 13-f filings)
  • Measures: We measure total institutional ownership as the fraction of shares outstanding owned by institutions. This measure should primarily capture indirect institutional monitoring, although it may also reflect some direct monitoring. Our second measure of institutional investor influence, the concentration of institutional ownership, is designed to capture the direct institutional monitoring. It is calculated as the proportion of the institutional investor ownership accounted for by the top five institutional investors in the firm.

David Kochhar Levitas (1998) - The effect of institutional investors on the level and mix of CEO compensation:

  • Money Market Directory, Moody's Bank and Finance Manual, Nelson's Directory of Investment Managers
  • Percent of ownership (of II) by type
  • Blockholders (shareholders with greater than 5%) provide better governance (indicator variable)

Bebchuk Cohen Hirst (2017) - The Agency Problems of Institutional Investors:

  • References data from FactSet Ownership database

Lewellen (2011) - Institutional investors and the limits of arbitrage:

  • The CDA/Spectrum database is compiled from institutions’ 13F filings with the Securities and Exchange Commission (SEC).
  • according to quarterly 13F filings compiled by Thomson Financial... though not exclusively [see the Wharton Research Data

Services (WRDS) User Guide for details].

  • Measures are N/A (wrong perspective)

Bloom Klein (2012) - Institutional Investors and Stock Market Liquidity:

  • Any financial institution exercising discretionary management of investment portfolios over $100 million in qualified securities is required to report those holdings quarterly to the SEC using Form 13F. Qualified securities include stocks listed for trading in the US, among other securities. These filings, compiled quarterly by Thomson/CDA and available through Wharton Research Data Services (WRDS), are the source of the stock holdings used in this study for the period 1980 to 2010.
  • Two measures of institutional stock ownership – the percentage of a stock owned by institutions and the number of institutions that own the stock
Looking inside WRDS

Datasets:

I used the first one. See wrds for the preprocessing.

Restoring the old data

A backup of the old database is available as compacqs_Fc.dump in Z:/mcnair/backups and the source files all appear to be in Z:/mcnair/compacqs (the SQL file and build notes are in E:\mcnair\Projects\CompAcqs).

Other useful things:

  • Distance was calculated using GoDoMetar.pm which doesn't have a writeup but is in E:\mcnair\Projects\Winner's Curse\Distance
  • The Governance variables were added to the Winner's curse project later from ISS (formerly RiskMetrics). See E:\mcnair\Projects\Winner's Curse\Governance
  • The CARs appear to have been processed with E:\mcnair\Projects\Winner's Curse\Stata\EdsRegs.do, which takes Estimation5K.txt as an input.

I made compacqs folders in E:\projects and /bulk/ for use, created a new compacqs database, and restore the tables from the last backup. The pg_restore threw some errors (the new server doesn't have a plpythonu extension control file) but was otherwise fine. I put the latest version of everything I could find in an old subfolder on E.

The last version of the dataset produced by BuildCompAcqs.sql was 2-3, which was processed by CEOAug22.do to build CEOAugustv2.log. However, I don't see any outregs or other automated table generation and most of the regressions are missing.

There's a do file in the dropbox, CEOOct.do, that is identical to CEOAug22.do up until around line 100. It looks like the last version of the STATA code. However, it loads AcqCEOSep, which must have been a .dta produced by Jim.

I built the following:

  • MasterV3-0.txt from the master table of the reloaded dbase, with the code in Revision.sql
  • Revision3-0.do, based on CEOOct.do, and loading MasterV3-0.txt. It produces Revision3-0.log, which currently puts us at the last build.

I also copied them to the dropbox.

Estimating a clean new build

Variables from the Winners dataset

The biggest hurdle to updating the data is that we pulled many of the variables from the Winner's dataset:

--Builds the Acqbase table where the unit of observation is an acquirer year
DROP TABLE Acqbase;
CREATE TABLE Acqbase AS
SELECT acusip, year, sum(car5) as car5tot, avg(car5) as car5avg, sum(car5*tpublic) as car5totpub, avg(car5*tpublic) as car5avgpub, sum(car5*(1-tpublic)) as car5totpriv, avg(car5*(1-tpublic)) as car5avgpriv, avg(sharesoutstanding) as sharesoutstanding, sum(car5*sharesoutstanding) as valctot, avg(car5*sharesoutstanding) as valcav, sum(tpublic) as numacqpub, sum(1-tpublic) as numacqpriv, count(tname) as numacq,
sum(transactionvalue) as transactionvalue, sum(transactionvalue*tpublic) as transactionvaluepub, sum(transactionvalue*(1-tpublic)) as transactionvaluepriv,
sum(CASE WHEN transactionvalue IS NOT NULL THEN 1::int ELSE 0::int END) as numwtv, sum(CASE WHEN transactionvalue IS NOT NULL THEN (1::int)*tpublic ELSE (0::int)*tpublic END) as numwtvpub, sum(CASE WHEN transactionvalue IS NOT NULL THEN (1::int)*(1-tpublic) ELSE (0::int)*(1-tpublic) END) as numwtvpriv,
sum(CASE WHEN transactionvalue IS NULL THEN 1::int ELSE 0::int END) as numwotv, sum(CASE WHEN transactionvalue IS NULL THEN (1::int)*tpublic ELSE (0::int)*tpublic END) as numwotvpub, sum(CASE WHEN transactionvalue IS NULL THEN (1::int)*(1-tpublic) ELSE (0::int)*(1-tpublic) END) as numwotvpriv,
sum(coalesce(transactionvalue,enterprisevalue,equityvalue)) as targetvalue, sum(coalesce(transactionvalue,enterprisevalue,equityvalue)*tpublic) as targetvaluepub, sum(coalesce(transactionvalue,enterprisevalue,equityvalue)*(1-tpublic)) as targetvaluepriv,
min(anaic::int) as anaic,
sum(tit) as numacqit, sum(tvc) as numacqvc,
sum (stockswap) as numstockswap, sum (terminationfeeacquiror) as numterminationfeeacquiror, sum (terminationfeetarget) as numterminationfeetarget, sum (tbalancesheetprivate) as numtbalancesheetprivate, sum (tincomestatementprivate) as numtincomestatementprivate, avg (agovgoldenparachute) as agovgoldenparachute, avg (agovlimitabilityactbywritconsent) as agovlimitabilityactbywritconsent, avg (agovmajvoterequirement) as agovmajvoterequirement, avg (agovresignrequiredonmajorityvote) as agovresignrequiredonmajorityvote, avg (pcstock) as pcstock, avg (pccash) as pccash, avg (drivingdur) as drivingdur, avg (distance) as distance, 
sum (CASE WHEN pcstock=100 THEN 1::int ELSE 0::int END) as numallstock, sum (CASE WHEN pccash=100 THEN 1::int ELSE 0::int END) as numallcash
FROM Winners GROUP BY acusip, year;
--18341

The variables, along with their new sources and comments, are as follows:

  • acusip, year, tname, tpublic, anaic, transactionvalue, enterprisevalue, pcstock, pccash -- A part of the VCDB20 build
  • tit, tvc -- Added with lookup tables based on NAICS (Est. 30 mins)
  • car5, sharesoutstanding -- Would need to be calculated using a GVKEY based data pulled from CRSP, some pre-processing, and a STATA script, then a load. (Est. 6hrs, assuming I can find the old code.)
  • equityvalue, stockswap, terminationfeeacquiror, terminationfeetarget, agovgoldenparachute, agovlimitabilityactbywritconsen, agovmajvoterequirement, agovresignrequiredonmajorityvote -- Could be pulled from SDC but I don't think they are used
  • tbalancesheetprivate, tincomestatementprivate -- Built from a COMPUSTAT pull (Add 30 mins)
  • drivingdur, distance -- Determined using a Google Maps API script. (Est 2hrs + $100 and 5hrs to run).

Other things

On top of this. we'd need to do the following:

  • The ExecComp pull (1hr)
  • The COMPUSTAT pull (1hr)
  • Rework the SQL (2 hrs)

So, my best guess is: 1/2 + 6 + 1/2 + 2 + 1 + 1 + 2 = 13hrs...

Old Notes

Same Day Acquisitions

We need to solve the multiple acquisitions on a single day problem. The result needs the variables listed for acquisitions below (crucially including pub/priv indicator, CAR5, sameday indicator, sharesoutstanding, with a cusip-year key).

An outline of a process to do this:

  • SDCv2 is the import with some simple exclusions and basic processing 49031 - but it was likely overwritten and became 40216. A rebuild from SDCBase appears the replacement but this seems to have thrown out same day acqs.
  • We therefore need to go back to the raw import of SDC_MA_Oct10v4-fixed-normal.txt into table "SDC".
  • Load table prc from backup: SHOUT appears to have come from prc, which in turn came from CRSPRelNOInc. *Load table CRSPRelNOInc: this a base table for both estimation and event window builds.
  • Recode some queries from the previous work:
    • We would need to calc day number again for SDC, which would mean bringing in the dayno table.
    • Then we'd need to create new estimation data for the missing records: acq_cusip, date_announced, acqkey, dayno, avg(ret) as ret, primex, vret, eqret, reldayno, retdayno (see EstimationBase) - but maybe this time take SHOUT with you!
  • Run the estimation (using script ???.do) and build a new estimationtotal file and import it in estimationoutput
  • Recode some queries from the previous work:
    • Build a new EventBase, and everything in CRSP_TABLE.sql through to Flatter
    • Join this to EstimationOutput, calculate the abnormal returns and then CARs.
  • Finally join all of this mess back to the data that we already have. We may need to recompute targetit and targetvc.

Variables

The unit of observation is an executive's compensation year, or firm and (year-1). Each observation must have a CEO identifier, a year, and CEO compensation info. Each observation also must have lagged acquisition information, though many of these variables will be zero, and should have lagged firm level control variables.

ID variable:
   ceofirmid - a unique id number for each ceo-firm pair. CEOs may serve as CEOs of more than one firm. This addresses this issue.
CEO variables:
   Year - the year of the CEO's compensation package
   Execid - belonging to CEO in that year
   age - of ceo in year
   tenurefirm - tenure with firm in years (int) calc'd from joined firm date
   tenureceofromdate - tenure as CEO in years (int) cal'd from becameceo date
   tenureceo - tenure since first appeared as ceo (for firm)
   tenurecomp - tenure since first appeared in execcomp (for firm) to first appeared as ceo 
   tenurecfo - tenure since first appeared in comp as cfo (for firm) to first appeared as ceo 
   prevcfo - 0/1 (with firm)
   ceobefore - 0/1 (with any firm in execcomp)
   gender - 1 for male, 0 for female
   salary - from execcomp
   bonus - from execcomp
   optionsfv - options fairvalue
   optionsrpt - options reported value
   optionsblk - options black-scholes values
   tdc1 - total compensation from execcomp
   histyears - the number of years the CEO has an acquisition record with the firm
   prevacqpub - the number of previous acquisitions made by the CEO (for the firm) of public targets
   prevacqpriv - the number of previous acquisitions made by the CEO (for the firm) of private targets
   prevacq - the number of previous acquisitions made by the CEO (for the firm)
   prevacqit - the number of previous acquisitions made by the CEO (for the firm) of IT targets
   prevacqvc - the number of previous acquisitions made by the CEO (for the firm) of IT 
   ceoacqswformerfirm - 1/0 to indicate whether the CEO has an acquisition history from a previous firm
   numtimesceo - the number of times a CEO has been a CEO to date in the data
   ceototdur - the total number of years of the CEO has been a CEO to date
   ceoprevacqpub - the CEO's total acquisitions of public firms whilst CEO with previous firms
   ceoprevacqpriv - the CEO's total acquisitions of private firms whilst CEO with previous firms
   ceoprevacq - the CEO's total acquisitions whilst CEO with previous firms
   ceoprevacqit - the CEO's total acquisitions of IT firms whilst CEO with previous firms
Acquisitions (joined on cusip6, ayear s.t. ayear+1=year): 
   numacq - number of acquisitions for the firm year
   numacqpriv - number of private acquisitions for the firm year
   numacqpub - number of private acquisitions for the firm year
   numacqit - number of IT acquisitions for the firm year
   numacqvc - number of VC-backed acquisitions for the firm year
   car5tot - the total 5-day Cummulative Abnormal Return (0 when no acqs)
   car5av - the average 5-day CAR(0 when no acqs)
   valctot - sharesoutstanding (in millions?) x car5tot
   valcav - sharesoutstanding (in millions?) x car5tot
   sharesoutstanding - only available when their was >=1 acquisition
   hasacqs - 0/1 of there were any acqs within year
   transactionvalue - the total reported transaction value of acqs
   targetvalue - the total reported value (transaction, enterprise, or equity in that order of appearance) of the acqs
   numstockswap - number of acquisitions that involved a stock swap
   anystockswap - 1/0 for if acquisitions involved a stock swap that firm-year
   numterminationfeeacquiror - number of acquisitions that involved an acquirer termination fee
   numterminationfeetarget - number of acquisitions that involved an acquirer target fee
   numtbalancesheetprivate - number of acquisitions where the target's balance sheet is private
   numtincomestatementprivate - number of acquisitions where the target's income statement is private
   pcstock - the average percentage of stock in the payment for acquisition targets that year
   pccash - the average percentage of cash in the payment for acquisition targets that year
   drivingdur - the average driving duration to acquisition targets that year
   distance - the average driving distance to acquisition targets that year

Firm (acquirer, joined on GVKEY, fyear s.t. fyear+1=year):
   fyear - fiscal year from CSTAT
   naics - as a string (for redundancy)
   naics2 - 2dg NAICS
   naics3 - 3dg NAICS
   state - name of state
   statenum - number assigned to states [0-61] as some non-states
   at -  total assets
   ppent - plant, property equipment net total
   ppegt - plant, property equipment gross total
   revt - sames as sales
   ni - net income
   acqi - current income from acquisitions in CSTAT
   acq - cost of acquisitions from CSTAT
   mkvalt- market value
   agovgoldenparachute - 0/1 for the CEO has a golden parachute (many missing)
   agovlimitabilityactbywritconsent - 0/1 whether governance has limited shareholders ability to act by written consent (many missing)
   agovmajvoterequirement - 0/1 for whether board members are elected on majority or plurality, which matters for uncontested seats (many missing)
   agovresignrequiredonmajorityvote - 0/1 "Indicates that a director is required to submit his/her resignation upon failing to receive support from a majority of votes cast (which, typically, the board may chose to accept or reject)." (many missing) 
CPI:
   deflator - for exec comp with 2010 as base year
   cpiavg - the average CPI for the compensation year
   deflatorlag - for fiscal/acq year with 2010 as base year
   cpiavglag - the average CPI for the fiscal/acq year

Notes from the draft

Execucomp data

Execuomp provides data on various characteristics of each CEO that can be used as explanatory variables for compensation, including age, sex, and experience.

Various compensation measures are provided, including salary, bonuses, and the value of option awards. We focus primarily on total compensation (variable TDC1 in Execucomp), which includes salary, bonuses, the value of option awards and other types of compensation.

Compustat data

We use Compustat data (also obtained through WRDS) for data on various firm-level explanatory variables, particularly firm size measures (employment and real plant and equipment) and industry identifiers (NAICS codes at the 2 digit and 3 digit level).

CPI data for deflating

We use annual data, converting compensation over time to real values using the CPI index as a deflator.

Old distribution from paper

After a small amount of data cleaning (consisting mainly of dropping observations for which data on important variables is missing) we are left with 6447 panels and 34,956 observations, implying an average tenure of about 5.4 years per CEO in our data.

For most observations (29,775), the number of acquisitions is zero. Exactly one acquisition occurs in 4,228 observations, exactly two occur in 1,122 observations, three occur in 391 observations, and in 310 observations there are four or more acquisitions. The data set contains 9,452 acquisitions overall, of which 8,211 are acquisitions of privately held companies and 1,241 are acquisitions of publicly traded companies.

In our primary analysis we therefore include observations (firm-CEO-year combinations) with more than acquisition. In such cases we consider two alternative measures of acquisition performance – the average abnormal return over all acquisitions in the year and the total abnormal return over all acquisitions. To see if using multiple acquisitions has an effect we can of course consider a subsample consisting only of observations with no more than a single acquisition.

Data Source Draws

COMPUSTAT

Query URL: https://wrds-web.wharton.upenn.edu/wrds/ds/comp/funda/index.cfm?navId=80

Execucomp

Query URL https://wrds-web.wharton.upenn.edu/wrds/ds/comp/execcomp/anncomp/index.cfm?navId=72

Saved Query: ExecCompForAcqsPaperV

Query spec:

 Data Request ID e0a8b1599f177e67
 Libraries/Data Sets comp/anncomp  /
 Frequency/Date Range ann / 1992 - 2017
 Search Variable TICKER
 Input Codes 
 all item(s) 
 -all-
 Conditional Statements n/a
 Output format/Compression tab /
 Variables Selected CONAME CUSIP GVKEY NAICS STATE EXECID EXEC_FULLNAME GENDER PAGE BECAMECEO CO_PER_ROL GVKEY JOINED_CO AGE BONUS CEOANN CFOANN OPTION_AWARDS_BLK_VALUE OPTION_AWARDS_FV OPTION_AWARDS_RPT_VALUE SALARY TDC1 TDC2 TITLEANN TOTAL_CURR YEAR

Note: We need GVKEY, CUSIP, and year to do the joins. NAICS and STATE can be supplemented by the COMPUSTAT and EXECUCOMP data.

Acquisition Data

This is currently taken directly from the last version of the Winner's Curse Dataset (see Z:\compacqs).

We need to rebuild this to include all of the times when we had multiple acquisitions on the same day. As we use various SDC dataset measure (like a public indicator, state, and naics) it isn't sufficient to just take the CAR5 and sharesoutstanding from the event regressions - we need to join it back to the SDC data first. We will then calculate car5tot and car5av and count private and public acquisitions by blowing out the data (so that the unit of obs is an acquisition, not an announcement day) and recalculating the aggregate variables.