Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Do Going Private Transactions Affect Plant
Productivity?∗
Sreedhar Bharath†, Amy Dittmar‡, and Jagadeesh Sivadasan§
April 2013
Abstract
We examine whether constraints on public firms such as capital market myopia and agency
problems impact firms’ efficiency by testing whether going private improves plant-level productivity
relative to peer control groups. We find no evidence that this is the case. Our key finding is that
while there is evidence for substantial within-plant increases in productivity after going private,
there is little evidence of efficiency gains relative to peer groups of plants constructed to control
for industry, age, size, past productivity and the endogeneity of the going private decision effects.
Also, we do not find evidence that myopic markets lead to under-investment at the plant level. On
the contrary, we find that after going private, firms shrink capital and employment, and sell and
close plants more quickly, relative to peer groups. Our findings cast doubt on the view that public
markets cause listed firms to make sub-optimal, productivity-decreasing choices, or under-invest at
the plant level.
Keywords: Going private, delisting, productivity, efficiency, firm performance, investment
JEL classification codes: G34, G14, G32, D24, D22
∗The research in this document was conducted while the authors were Census Bureau research associates at the
Michigan Census Research Data Centers. Research results and conclusions expressed are those of the authors, and do not
necessarily indicate concurrence by the Bureau of the Census. The results presented here have been screened to ensure
that no confidential data are revealed. We thank Clint Carter and Arnold Reznik for prompt processing of our disclosure
and data access requests, Natarajan Balasubramanian for advice and help with the data, and Xiaoyang Li for research
assistance. All remaining errors are our own.†[email protected], Arizona State University‡[email protected], University of Michigan§[email protected], University of Michigan
1
1 Introduction
The stock market oriented U.S. financial system is often criticized for providing strong incentives to
corporate managers to behave in a myopic manner. Porter (1992) argues that the U.S. system ad-
vances the goals of shareholders interested in near-term appreciation of their shares even at the expense
of long-term performance of American companies. The nexus of stock market analysts, traders and
fund managers increases the focus on quarterly earnings and other short-term metrics of performance.
Further, managers of public firms are subject to agency conflicts and may engage in empire building
(Jensen (1986) and Jensen and Meckling (1976)) or exacerbate the myopic behavior to improve their
own labor market prospects or compensation, which may be tied to the stock price (Rumelt (1987),
Campbell and Marino (1994), Narayanan (1985) and Holmstrom and Ricart i Costa (1986)).
In this paper, we match data on going private transactions to rich plant-level US Census micro-
data to examine how going private affects plant-level productivity, investment and exit (both sale and
closure of plants). We do this to determine whether changes, if any, are consistent with the view that
agency costs and market myopia impact behavior of listed firms more than private firms. If potential
agency costs and capital market myopia associated with being public affect operational performance,
we expect productivity to improve after firms go private.
Perhaps the most widely cited evidence about harmful short-term market pressures come from a
survey of corporate CFOs published in Graham, Harvey and Rajagopal (2005). In this survey, the
authors find that managers would rather take economic actions that would have negative long term
consequences and sacrifice value in order to meet short term quarterly earnings benchmarks. However,
though they can be very informative, surveys measure beliefs, which may not always coincide with
the actions of managers. Indeed, Shleifer and Vishny (1997) in their review of corporate governance
conclude that the theories and arguments in favor of the view that U.S. companies are relatively short
sighted are remarkably short of empirical support. We focus our analysis on the effect of going private
because, as a private firm, the agency conflict that results from dispersed ownership is alleviated and
the influence of a myopic capital market is removed. We therefore examine changes in the plant-level
efficiency of firms that opted out of public markets by going private and contrast these firms’ plants
against a peer group of ’matched’ plants that did not undergo a similar change from public to private.
We find no evidence that going private causes plants to improve their productivity relative to
their matched peers. Specifically, while there is evidence of substantial within-plant (establishment)
increases in productivity after going private (5.1% in the preferred Blundell-Bond TFP measure and
7.2% in labor productivity in the long-run), there is little evidence (close to zero) of difference-in-
difference efficiency gains relative to a peer groups constructed to control for industry, size, and age.1
1We use a widely used measure of labor productivity (defined as real output by employment), and two measures of total
2
Further, our productivity results hold when we exclude all plants from both the going private and
matched control samples that underwent a change in ownership after the going private date, ad-
dressing the potential concern that matched control plants may be undergoing improvements through
changes in ownership.
These results continue to hold when we address potential endogeneity concerns by creating our
matched control sample using past plant productivity or the propensity to go private. To measure the
propensity to go private, we use information for all the firms at the time of their IPO (which is on
average 13 years before the going private decision). In doing so, we build on Bharath and Dittmar
(2010), which investigates the determinants of going private and find that both the characteristics
at the time of the IPO as well as how firm characteristics change over time impact the decision to
go private. Specifically, Bharath and Dittmar show that firm characteristics at the time of the IPO
predict the ultimate decision to go private with a 71% accuracy rate.2 We therefore use their model
to determine the propensity of a firm to go private. The propensity as well as the past productivity
matched results are similar to those from using the industry-age-size matched control group. None of
our findings suggest that operational efficiency of establishments that went private are differentially
enhanced even six years after going private, thus casting doubt on the view that the U.S. stock mar-
ket’s excessive focus on short term results is affecting their long-run operating performance.
Next, we examine investment decisions as well as establishment exits. Most models of market
myopia (Stein (1989)) and also some empirical work (Bhojraj et. al.(2009)) suggest that stock mar-
ket short-termism could lead managers to boost current earnings, at the expense of forgoing longer
term investments. The key problem is that if short-run earnings are poor, the market is unable to
determine whether this is caused by poor management or by prudent long term investments. The mar-
ket partly uses current earnings to forecast future earnings and knowing this, the managers attempt
to manipulate the information available to shareholders by increasing current earnings, potentially at
the expense of long-term investments, resulting in under-investment. Even if the markets were to
assume no myopia on the part of managers, the latter have an incentive to deviate from non-myopic
behavior in order to fool the market by manipulating current earnings. Manipulation in earnings could
be accounting actions or even real actions (such as forgoing investments). To test this implication
factor productivity (defined as a residual from regressing log real output on all logged inputs) in our analysis. All measures
give similar (close to zero) results in the alternative difference-in-differences analysis, suggesting that the peer group changes
for these measures were similar to that for the going private sample. In the descriptive before-after analysis, labor productivity
shows larger increases than TFP measures because it does not adjust for the larger increases in capital (relative to labor).
The alternative measures are discussed in detail in Section 2.2.2Because Bharath and Dittmar’s propensity model examines the probability of listed firms to be taken private, the propensity
matching approach restricts the control group to establishments of publicly listed firms; so the comparison in this case is
between establishments that go from public to private vs establishments that strictly remained public.
3
of myopic behavior, we examine establishment-level capital stock, employment and plant closures. If
market myopia leads to under-investment when the firms are listed, we would expect to see a relative
increase in capital stock, commensurate increases in employment, and a greater patience with (po-
tentially short-term) under-performance and hence a relatively lower propensity to exit (close or sell)
plants, after going private.
We find no evidence that firms increase establishment-level investment after going private. In fact,
relative to the three control groups discussed above, we find that, if anything, firms shrink capital
stock by 15% and employment by 3.3% in the long-run the baseline ’Difference-in-Differences’ analysis
after going private. We further find that going private firms exit (sell or close) plants more quickly
(15.3% higher hazard) than the matched control group, particularly in the short-run after going private.
The higher exit is driven by both greater sales (33% higher hazard) and closures (6.1% higher hazard).
Taken together, the results on establishment level capital stock and exit propensity do not suggest
under-investment by going-private firms while they are public relative to after going private. Some
models of market myopia (Bebchuk and Stole (1993) and Bizjak, Brickley and Coles (1993)) as well
as those of agency problems (Jensen (1986)) predict “over-investment” by public firms. Bebchuk and
Stole (1993) and Bizjak, Brickley and Coles (1993) point out that the relationship between managerial
short-term objectives, imperfect information and firm investment behavior can result in either under-
or over- investment and depends on the observability of investment. For our purposes, the key im-
plication of these models is that the capital market’s fixation with meeting short-term expectations,
coupled with asymmetric information problems, too often hinders corporate managers from focusing
on long-term value creation. Severing the link between capital markets and firms (by going private) will
therefore solve the managerial myopia problem. We should then expect to see a (long term) increase
in productivity due to managers focussing on the value enhancing long term projects. While we do
find that firms appear to downsize after going private, the fact that we find no difference-in-differences
changes in productivity between the public and private firm establishments suggests that myopia is
not impacting the operational performance of listed firms. In other words, the productivity results rule
out both“under” and “over” investment induced by market myopia insofar as these terms are defined
in the context of optimal operational efficiency. Our study contributes to this literature by examining
if there are operational improvements to be obtained by removing the hypothesized myopic pressures
imposed by capital markets; thus, comparing firms when they are public to when they are private.3
Next, we test whether going private firms are more patient with lower performing plants after
de-listing. On the contrary, we find moderate evidence that going private firms more aggressively
3The lack of improvement in productivity also argues against one explanation for downsizing after going private – empire
building due to more severe agency problems in public firms. At the least, our results rule out a negative effect of empire
building on operational efficiency.
4
exit (both sell and close) plants that have lower productivity. Specifically, we find that the signs on
all of the estimates suggest stronger selection to exit lower productive plants, with majority of the
estimates being both statistically and economically significant. Thus, the going private firms appear
to attempt to improve the productivity of their portfolio of plants, but this is achieved through exiting
low productivity plants rather than through improvements.
We interpret our results as suggesting that: (i) myopia or agency problems associated with cap-
ital markets do not affect operational efficiency; (ii) investors in going private transactions appear
to gain value not by improving productivity within target plants, but by reducing capital, as well as
selling and closing (less productive) plants. To explore potential alternative explanations of our results,
we examine differences across different types of acquirers involved in the going private transactions.
Specifically, we classify transactions into three sub-groups – private equity firms, management, and
private operating firms – by collecting data on the parties that drive each of these transactions. We
then compare changes in these groups relative to an industry-age-initial size matched control group
of establishments. One alternative explanation is that high leverage used in going private transactions
could lead to short-termism after delisting, and at the same time exert pressure to downsize plants as
well. Because higher leverage is more likely for management buy-outs and private equity takeovers,
examining the outcomes for operating acquisitions allows us to consider a sub-sample where high
leverage is less of a concern. Another alternative explanation is that (even in transactions without
much debt) for some investors, their time horizon for recouping investment in going private transac-
tions may be short, so that short-termism persists after going private. This is less likely to be true for
operating firm buy-outs.
Our general finding is that regardless of the mechanism used to go private, there is no improvement
in productivity relative to the matched control group.4 In particular, operating firm takeovers, which
constitute about 45% of our sample and which are less likely to be affected by leverage and short-
term investment horizons, do not show improvements in productivity. We also find that operating
firms show large (6.8% to 7.9% in the short-run, 8.3% to 11.6% in the long-run) declines in capital
but smaller and not statistically significant declines in employment. Further, operating firms show a
significantly higher propensity than matched controls to both sell and close establishments; in fact,
over a 6 year horizon, operating acquirers are more likely to close a plant (1.7%) relative to matched
controls than either private equity (-0.6%) or management acquirers (-3.3%), though the propensity
to sell is higher for management and private equity acquirers (6.4% and 7.2% relative to 3.7% for
operating acquirers). Thus, we find that qualitatively the results for the operating firm sample mirror
what we find for the overall sample in the baseline regressions, suggesting that leverage or investor
4One small exception is that using one of our three baseline measures of productivity (Blundell-Bond TFP), we find firms
that go private by being bought by an operating firm have a significant improvement once we purge sold plants from sample.
However, this result is not robust to other TFP measures and is no longer significant once we match by past-productivity.
5
short-termism is unlikely to be the main explanation for the baseline results.
We perform a number of checks and additional analysis to determine the robustness of our results.
We analyze outcomes at the acquiring firm to investigate if changes at the acquiring firm offset, and
hence affect the interpretation of, results for the going private establishments. In particular, we find
no difference-in-differences increases in capital or employment, or productivity measures at acquiring
firm establishments, and also no increase in new establishment openings at acquiring firms. Thus,
we conclude that there is no evidence for changes at the acquiring firm establishments offsetting (or
complicating interpretation of) changes documented at the going-private establishments. In addition
to this, we use profitability (instead of productivity) measures as alternative performance measures,
explore changes in outcomes over different time periods, and check robustness to using alternative
specifications.
Our paper contributes to the long literature that examines productivity changes around corporate
events. Maksimovic, Phillips and Prabhala (2011) and Maksimovic and Phillips (2001) examine pro-
ductivity changes and purchase decisions by acquirers after mergers and acquisitions to understand
how firms redraw their boundaries and evaluate the efficiency of resource reallocation. Maksimovic
and Phillips (2002 and 2008) find that plants acquired by conglomerate firms increase in productivity
and conclude that organizational forms’ comparative advantages differ across industry conditions. Our
paper has a similar approach but focuses on public to private transactions. Lichtenberg and Siegel
(1991) find that TFP (total factor productivity) increases after a leveraged buy out (LBO) using a
sample of 131 firms that conducted an LBO in 1983-1986. Kaplan (1989a, 1989b, and 1991) also ex-
amines the benefits of going private using a sample of LBOs and highlights the importance of tax and
incentive improvements due to the high leverage in these transactions. Our paper expands on these
studies and our sample of firms includes LBOs and MBOs but has a large number of private operating
firms buying public firms and taking them private. Davis et. al. (2008a and 2008b) study changes
after a private equity deal and point out that productivity and employment relationships uncovered in
earlier studies may not hold because of the tremendous changes in the private equity industry due to
increased competition for transactions. They also observe that fundraising (inflation -adjusted dollars)
by U.S. private equity groups is 100 times greater in 2006 than in 1985 and is a primary driver of
these changes. Davis et al (2008a and 2008b) focus on the entire universe of private equity firm deals,
the vast majority of which are private to private transactions. Alternatively, our study focuses only
on public-to-private transactions of manufacturing firms including those acquired by private operating
firms, and studies TFP and employment changes. Based on public 13E-3 filings with the SEC, public
to private deals account for only about 157 out of more than 5,000 deals done by private equity firms
from 1980-2005. Thus, we examine the impact of market myopia by focusing on going private deals
and they investigate the role of private equity.
6
This paper also contributes to an understanding of the impact of capital market myopia on firms
investment and productivity. Both survey and empirical evidence seem to be consistent with the view
of managerial myopia induced by capital markets. Graham, Harvey and Rajagopal (2005) report in
a survey of more than 400 financial executives, that 80 percent of the respondents indicated that
they would decrease discretionary spending on such areas as research and development, advertising,
maintenance, and hiring in order to meet short-term earnings targets and more than 55 percent said
they would delay new projects, even if it meant a small sacrifice in value creation. Assuming these
survey responses reflect the actual intent and behavior of executives, these results indicate that myopia
is a larger issue than companies simply using accounting actions to meet quarterly earnings expec-
tations from financial markets. There are also real actions such as asset sales and forgone strategic
investments that corporate managers use to meet the forecasted quarterly earnings number. In a re-
lated empirical study, Bhojraj et al (2009) document that using accruals or discretionary expenditures
(such as R&D expenditure) to meet or beat analyst forecasts results in short-term positive impact
on firm performance, but long-term under-performance relative to firms that do not manage earnings
to meet forecasts. These results confirm managerial myopia due to capital market pressures in an
empirical setting if one further assumes that firms that manage earnings are most likely to engage in
myopic behavior relative to their control sample that do not manage earnings. Although the creation
of long-term company value is widely accepted as management’s primary responsibility, these results
suggest that managing predominantly for the market’s short-term earnings expectations often impairs
a manager’s ability to deliver value. Asker, Farre-Mensa and Ljungvist (2012) compare a set of small,
private firms with data available on Sageworks with a matched sample of similarly small listed firms.
For this sample, the authors find that listed firms invest less and are less responsive to changes in
investment opportunities compared to matched private firms. Our study differs from these others in
that we examine the change in productivity and investment in labor and capital of firms that go private
relative to a matched sample. By employing the Census data, we are able to both examine a large set
of firms and calculate the impact of going private on their productivity, and we provide evidence that
is inconsistent with public firms being more subject to myopia.
The remainder of the paper is organized as follows. Section 2 describes the data and productivity
measures. In Section 3 we discuss methodology and present results on productivity changes. Section
4 discusses the analyses and results from examining other outcomes (capital, employment, and plant
exit). Section 5 discusses the results and related additional robustness checks. Section 6 concludes.
7
2 Data and productivity measures
2.1 Data sources and description
The main sources of data used in this study are the Census of Manufactures (CMF) for the years 1977,
1982, 1987, 1992, 1997 and 2002, and the Annual Survey of Manufactures (ASM) for remaining be-
tween census years from 1978 to 2004. This census data provide detailed information on individual
establishments (i.e., plants) of both public and private firms. This data has been used in previous
studies, particularly to study the effects of mergers and acquisitions on productivity, discussed earlier
in this paper. We also use the longitudinal business database (LBD) to obtain identifiers to link
establishments over time, and for data used in exit analysis.5 Technical details on the cleaning of the
data, as well as the detailed definitions of the key variables used in the study are provided in the data
appendix. Detailed descriptions of the productivity variables are provided in the next section (with
additional details provided in the data appendix).
To analyze the effect of going-private on firm productivity and other outcome measures, we use a
comprehensive sample of firms that went private as detailed in Bharath and Dittmar (2010). Bharath
and Dittmar (2010) use all forms of 13e-3 filings to identify going private transactions and require
that firms are no longer registered or traded (even over the counter). They also supplement their
sample with data from Lehn and Poulsen (1989) to ensure a complete sample in the early periods.
We then match these firms to census databases using the Compustat-SSEL bridge available at the
Census using 6-digit CUSIP identifers for the period 1981 to 2005 to identify all establishments owned
by the sample of going private firms. In the baseline analysis we create a control sample for each
establishment in the going private sample (hereafter ‘going private establishment’), by including up to
eight establishments (based on data availability) that are closest in size (measured using employment)
to the going-private establishment in the going-private year, from within the same 3-digit SIC industry,
and belonging to the same age quartile. We use this matched sample throughout but also provide
evidence using alternative matched samples by industry, past productivity, age as well as one based
on propensity score matching on the probability a firm will go private.
Panel A of Table 1 presents the summary statistics on the number of establishments in event time
for a period of thirteen years with year 0 being the date of going private. Overall, we have 28,518 go-
ing private establishment year and 157,391 control firm establishment year observations, representing
5,676 going private and 90,009 control sample firm years in the sample. All of our analysis examines
outcomes at the establishment level.6
5We also check if the employment and exit results for manufacturing firms hold for the full universe of firms using data
from the LBD (see point v in section 5).6The first step of the data setup process identifies a baseline sample of going private establishments and matched controls
in year -1. Because of attrition (due to closure or due to not yet being born), the number of establishments is lower in years
8
While the establishment level changes provide a detailed and disaggregated picture of the effects
of going-private, aggregating to the firm level is difficult here because the firm identifier for the estab-
lishments of the acquired firm will change to that of the acquiring entity after the going-private event,
which is not our focus here. Importantly, McGuckin and Nguyen (1995) show how aggregating to the
firm level could mask interesting establishment level changes at the target establishments. There are
two additional reasons to focus on establishments. One, many of the firms have multiple establish-
ments operating in multiple industries; thus, forming a suitable control group for a firm that matches
the firm’s industry composition is more difficult. Two, one aspect of firm behavior we specifically
want to look at is the decision to sell or shutdown particular establishments (see section 4.2). The
establishment level analysis could mask improvements that occur due to selective closure of inefficient
establishments; we address this separately in Section 4.2.3.
Table 1 Panel B breaks down the number of firms and establishments by type of acquirer, i.e. the
type of firm that took the firm private. Of the 11,255 (20,114) going private firms (establishments),
112 (819) were take private by management, 115 (958) by a private equity firm, and 188 (1,207) by a
prior operating firm, similar to the breakdown in Bharath and Dittmar (2010). If a firm is acquired by
both a private equity firm and management, then the transactions are classified under multiple groups,
so there are a small number of overlaps in these subgroups. For 48 (301) firms (establishments), we
were unable to determine the acquirer type.
Table 1 Panels C and D provide summary statistics for the going private firm (target) as well as the
acquirer, respectively. Note that acquirer is the same as target when it is management that takes the
firm private. In both Panel C and D, the labor productivity of the plants (for both target and acquirer)
is larger than the industry average. Both total factor productivity (TFP) measures are systematically
higher than the industry mean for the acquirers (panel C), but not for target, as the Blundell-Bond
TFP measure is negative for all groups except the targets of private equity takeovers. Comparing
these panels, the going private sample has fewer plants in fewer industries with lower sales than the
acquirer. Further, the going private establishments have lower productivity than the acquirers, using
all three measures of productivity in t-1, with the exception of those that are unclassified.
before and after year -1. The number of firms in columns 5, 6 and 7 report the number of firm identifiers for the identified
establishments; because some going-private establishments in year -1 were bought from other firms before year -1 or sold to
other firms after year -1, the number of going-private firms is larger before and after year -1. For the control firms, attrition
in number of establishments more than offsets increases due to sales and so number of firms is the largest in year -1.
9
2.2 Key productivity measures
In this section, we discuss in detail the three alternative measures used in our analysis of establishment-
level productivity, as well as the variables and methodology used in their definition.
• Labor productivity: Labor productivity is defined as log real value of shipments divided by
employment. Value of shipments is simply the sales value deflated using 4-digit SIC industry-
specific output deflators (obtained from Becker and Gray’s (2009) NBER-CES manufacturing
industry database). Employment is the total number of employees reported in the ASM-CMF
database.
While labor productivity adjusts for increase in labor inputs, total factor productivity (TFP) mea-
sures adjust for changes in all inputs. A standard measure of TFP is the residual in a Cobb-Douglas
production function: yit = βmmit + βkkit + βeeit + βnnit + βllit + TFPit, where yit is the log of real
value of shipments of establishment i in year t, m is log real materials, k is log of real depreciated
capital stock, e is log of real depreciated energy costs, and n is log of white-collar (non-production)
employment and l is log of blue-collar (production) employment. The residual TFP is typically recov-
ered using a regression of real output on inputs; however this regression suffers from endogeneity as
inputs are likely to be chosen based on TFP (Marschak and Andrews 1944). We use two approaches
to address endogeneity concerns, as described below.7
• OLS-FE TFP measure: The main source of endogeneity in panel context is likely to arise
from fixed TFP heterogeneity across firms, arising from unobserved differences in factors such
as labor quality, locational advantages, and entrepreneurial quality. A solution is to use panel
data transformations with fixed effects, which control for any factors that do not change over
time. The OLS-FE productivity measure is defined as the residual from an OLS establishment-
fixed-effects regression of log real value of shipments on log real materials, log real energy costs,
log blue-collar employment, log white-collar employment and log real capital.
• Blundell-Bond system-GMM TFP measure: While the use of fixed effects to solve the
endogeneity problem is simple and appealing, in practice the approach often yields unrealistically
low (and sometimes even negative) coefficients for capital, because fixed effects increases the
noise component of the capital measure.8 Further, it is plausible that some productivity shocks
are anticipated, so that inputs respond to changes in productivity as well. An alternative solution
that addresses these issues is the “system” GMM approach, which uses both first differences
and levels (Blundell and Bond 2000), where lagged first-differences are used as instruments for
7In results available in an online appendix, we verified robustness to using three additional productivity measures: Solow
residual TFP measure, Levinsohn and Petrin TFP measure, and the Translog TFP measure. For brevity, we do not present
these in the paper but note the results are similar throughout.8Intuitively, capital often has lagged effects, and sometimes investment could have short-run disruptive effects. So changes
in capital often has low or negative correlation with contemporaneous changes in output.
10
equations in levels, in addition to lagged-levels as instruments for equations in first-differences.
A brief description of the Blundell-Bond (2000) procedure used by us is provided in the data
appendix.
Because the Blundell-Bond approach addresses potentially important endogeneity issues in the
most systematic manner, we use this measure as our main baseline measure. We also report the other
two measures because these are widely used in the productivity literature as well (and because the
Blundell-Bond coefficient estimates are sometimes sensitive to choice of instruments), so robustness
to these alternative measures would provide reassurance about the resilience of our results. When
comparing the magnitudes of labor productivity to TFP measures, it should be noted that if firms
experience larger increases in capital (or other inputs) relative to labor, the magnitude of the labor
productivity measure could be significantly higher than for TFP (because labor productivity does not
adjust for changes in capital).
For TFP measures, we allow production function coefficients to vary by 2-digit SIC code; as
expected (Griliches and Mairesse 1995) we find that relative to the Blundell-Bond estimator, the
coefficient on the employment variable is upward biased in simple OLS, while the capital coefficient is
downward biased when using OLS fixed effects.9
3 Analysis of productivity changes
When examining productivity and other outcomes, we present two sets of results. The first set of
“before-after” results summarizes what happened to the key variables of interest within the estab-
lishments that belonged to firms that went private, compared to their levels prior to going private.
The second set of “difference-in-differences” results presents the changes in the variables of interest
relative to changes in a matched control group of establishments. Both of these results use fairly
standard methodologies from the literature,which we discuss in the following sub-sections.
3.1 The before-after methodology
To examine before-after changes, we retain data for as many of the 13 years surrounding the event as
available for each establishment belonging to the firms that went private. These include up to 6 years
of data before the year of going private, the year of the firm went private, and up to 6 years after the
firm went private.10
9However, despite differences in coefficient estimates, we find that the estimated TFP residuals are very highly correlated
(correlation greater than 0.85) in our sample, so it is not surprising that in the analysis below we find consistent results across
alternative approaches. This consistency across alternative TFP estimators have been found by others using US Census micro
data e.g. Greenstone, Hornbeck and Moretti (2010).10Note that there may be some establishments that were born less than 6 years before the firm went private, and some
establishments that exit less than 6 years after the going-private event.
11
We then use simple regression approaches to summarize the before-after changes in two ways.
First, we use the following regression specification:
yit = βLR PRE LR PRE + βSR PRE SR PRE + βSR POST SR POST
+ βLR POST LR POST + fi + eit (1)
where yit stands for the dependent variable (productivity or other measures), fi stands for plant fixed
effects, and the four dummy variables are defined to capture four distinct time periods as follows: (i)
The long-run before going private: LR PRE is a dummy equal to one for the 3-year period from 6
to 4 years before going private and zero otherwise; (ii) The short-run before going private: SR PRE
is a dummy equal to one for the 3-year period from 3 to 1 years before going private and zero other-
wise; (iii) The short-run after going private: SR POST is a dummy equal to one for the 4-year period
from 0 to 3 years after going private and zero otherwise; and (iv) The long-run after going private:
LR POST is a dummy equal to one for the 3-year period from 4 to 6 years after going private and
zero otherwise. The term eit stands for residual error.
The estimates of interest are not the levels of the dependent variables, but rather their changes
over time.11 Specifically, we are interested in the following changes:
(i). Short-run post- versus short-run pre- going private (βSR POST − βSR PRE): This provides an
estimate of the short-run changes after going private, relative to the period just before the
going private event. Thus, if the new owners take steps that have immediate effects on the
performance of the plant, this should be reflected in this estimate.
(ii). Long-run post- versus short-run pre- going private (βLR POST − βSR PRE): This provides an
estimate of long-run changes in the dependent variable after, relative to the period just before,
the going private event. If the actions of the new owners take some time to have an impact, we
may obtain significant estimates here, but not in (i) above.
(iii). Test for prior trend (βSR PRE−βLR PRE): This provides an estimate of trends in the dependent
variable prior to going private. If the establishment was experiencing an increasing (decreasing)
trend in the dependent variable, this would manifest as a positive (negative) estimate in this
test. Thus, any changes we document in (i) or (ii) above, should be evaluated in the context of
the pre-existing trend captured by the estimate here.
The first three columns of Table 2 present the results of this analysis. Panel A includes all
establishments with available data. Panel B restricts the sample to only those establishments that
were not later sold in years t+1 to t+6 to ensure that the effects are due to going private and not
due to a subsequent sale (in either the going private or the control group). For all inferences, we
11The inclusion of the plant fixed effects implies that one of the time period dummies is not identified. However, our
estimation procedure reports the mean for the omitted LR PRE as the constant term.
12
compute p-values based on standard errors clustered by establishments.12 The first column regresses
labor productivity while columns 2 and 3 measure TFP (total factor productivity) according to the
two methods described briefly in section 2.2.
We find that there is a pre-existing improving trend in all the productivity variables (between
3.4% and 10.5%) prior to the going private decision, as evidenced by the positive and significant
coefficient on short-run pre versus long-run pre comparisons. Labor productivity and TFP increase
both in the short run (by 6.3% and about 3.2%, respectively) and the long run (by 7.2% and about
5% respectively) after going private and these differences are statistically significant. Because TFP
is measured as a residual from a regression of real output on inputs, increases in TFP translate to
increases in output holding inputs fixed; converting the three years means into annual rates, our
numbers indicate about a 1 percent annual increase in real output (net of increase in inputs) for the
going private establishments in post-going private period, but this is close to the per year increase in
the pre-going private period, so the regression results do not indicate a strong change in trend.
The first three columns of Table 2 Panel B present similar results. Again, in this panel, we exclude
establishments that are subsequently sold, resulting in approximately 8000 fewer establishment-year
observations. Similar to the full sample, the establishments exhibit a significant pre-period trend and
a significant increase in productivity from before to both the short and long run after going private.
In fact, the economic significance of the results are considerably stronger in this subsample, with
the productivity increase between 7 and 19 percent in the short run and 10 to 21 percent in the
long run. These results are consistent with the acquiring firm selling establishments that do not
experience an increase in productivity. In section 4.2.3, we will directly examine the characteristics of
the establishments that are sold.
In order to allow for a more flexible examination of the year-by-year effects (and for a direct visual
check for pre- and post-trends), we examine a standard event study graph, by plotting coefficients on
the index dummies from the following regression specification.
yit =
6∑k=−6
βkDk + fi + eit (2)
where k indexes the years after the going private event, and correspondingly Dk is a dummy variable
equal to one for the year k after the going private event. Negative values of k correspond to years
before going private. All other variables are as in (1) above. We then plot the βk coefficients as well
as the corresponding confidence intervals, to illustrate the trends in the dependent variable, and the
significance of the changes in the trends. Figure 1 shows that there is a statistically significant increase
in the productivity measures for the establishments after the firm goes private, for all alternative
measures. Further, consistent with the results in Table 2, there appears to be a strong pre-existing
increasing trend in almost all of the measures, though the figures suggest a short-run slowing down of
productivity before going private and a short-run acceleration after going private.
12The significance levels were similar when we clustered by firm.
13
3.2 The difference-in-differences methodology
The before-after analysis simply summarizes the trends in the variables of interest in the plants be-
longing to the firms that went private. However, these changes could be driven by factors specific to
the industry, or age-related changes (as plants are increasing in age over the period of our analysis).
Changes may also be driven by factors related to the pre-event size of the establishment, e.g. if
going-private firms’ establishments were relatively large, and if all large establishments experienced
relatively different patterns of productivity change.
In order to rigorously address potential bias from these industry, age and initial size related factors,
we form a matched control group for each establishment in the going-private sample.13 Specifically, for
each establishment in the going-private sample, we select up to eight matched control establishments
in the following way to allow a comparison to an average matched firms rather than a more limited
comparison sample. Using the data for the closest prior-to-going-private year in which the establish-
ment is observed in the ASM-CMF sample, we classify all establishments into 3-digit industry-age
quartile groups, using a two-way sort so that we have four groups for each. Then, we sort again
by employment within each industry-age quartile group; and we select up to four non-going-private
establishments just lower and up to four non-going-private establishments just greater in size than the
going-private establishment, for each going-private establishment. There are not always 8 matched
controls in cases where the going-private establishment was too close to the largest or smallest estab-
lishment within the industry-age quartile. For a very small sample (less than 3%) of establishments,
control groups overlap. We dropped all such control group establishments from our analysis, so that
control groups are unique to each going-private establishment.
This procedure generates non-overlapping ‘cells’, with one going-private establishment and up to
eight control establishments. We then estimate the following regression specification:
yijt = β0 + βLR PRE LR PRE + βSR PRE SR PRE + βSR POST SR POST
+ βLR POST LR POST +Djt + eijt (3)
where i refers to the plant, j refers to the cell that plant i belongs to, Djt refers to cell-year fixed
effect, and the other variables are as defined in the before-after specification (1) above. Note that
period dummy variables are defined only for the going private sample – for instance, LR PRE is a
dummy defined equal to one for going private establishments in the 3-year period from 6 to 4 years
before going private (zero otherwise). Thus, the intercept term (β0) captures the overall mean value
13The importance of industry variables (such as market structure, industry demand and capacity utilization) for firm
restructuring outcomes has been demonstrated in the literature, particularly by Kovenock and Phillips (1997) and Maksimovic
and Phillips (2008). The results in these papers suggest that it is crucial to compare outcomes holding industry-level shocks
constant, as we do using industry-age-size-year fixed effects in the baseline analysis, and other industry-specific year effects
in other analysis.
14
for the control group of establishments. The inclusion of the cell-year dummies (Djt) implies that the
coefficients on the period dummy are estimates of the differences between the going private establish-
ments and the control establishments, for that period.
Accordingly, the differences between the period dummies yield difference-in-differences estimates
that control for cell-year, or equivalently, industry-age-size-year effects (where size refers to the pre-
going-private size of the establishment). As before, we are interested in the three estimates detailed
in section 4.1, but defined as a difference from the control group, thus calculating a difference in
difference. This provides an estimate of the changes in differences between the going-private and
control group in each defined period relative to the period just before the going private event. If
both the going-private as well as their matched controls experienced similar changes in the dependent
variable, there would be no changes in the difference between the treated (i.e. going-private) and
the control group. The way the control groups are constructed, controls for any effects related to
industry-wide changes, or age-related changes or initial size related changes (or any combination of
these) are accounted for. In particular, industry-age-size specific year effects are controlled for in this
estimation. We also test if the difference between the going-private and the control firms is increasing
or decreasing in the pre-going private period. Though we match on age and size characteristics just
prior to the going private event, differences between the going private and control groups could exhibit
specific trends in the prior period. One particular concern would be that, relative to this matched
control group, the efficiency levels of the going-private establishments may have been on an up-trend;
that is, the going-private establishments may have been selected based on prior trends. Then, any
post-going-private changes may simply be a reflection of these relative trends. Therefore, this test
helps to establish whether differential prior trends are a source for biasing estimated difference-in-
differences changes.
Columns 4 to 6 in both Panels A and B of Table 2 present the results of this analysis. For all infer-
ences, we compute p-values based on standard errors clustered by the industry-size-age cells. As in the
first three columns, we regress labor productivity and two measures of TFP (total factor productivity)
according to the various methods described in section 2.2. We note three important results from this
table. First, we find that there is no pre-existing improving trend in any of the productivity variables
prior to the going private decision relative to the control establishments included in the regression.
This suggests that the earlier result of a trend in productivity for the private firms is also mimicked by
the control group of establishments, perhaps mirroring industry (or age or initial size) related trends.
Second, there is no evidence of a short-run increase in productivity variables for the going private es-
tablishments in a difference-in-differences sense when compared with the control group. This indicates
that while private firm establishments do have a short-run increase in productivity after exiting the
public markets, so do the control group establishments that do not change their public-private status.
Thus, we do not see any evidence of the pressures due to the short sighted behavior by the public
15
markets, or more severe agency problems for listed firms, which were postulated to be a drag on their
productivity. Third, we do not find any evidence of a long-run productivity increase in private firm
establishments, relative to the control sample. These results question the commonly held belief that
the U.S. stock market by its excessive focus on short-term earnings imposes myopic behavior on part
of the firm to meet such expectations.
Again, as in the case of the before-after analysis, we also examine a difference-in-differences event
study graph, by plotting coefficients on the index dummies from a DID version of equation 2 (with
cell-year fixed effects that control for industry-size-age-year effects), in Figure 2. Figure 2 confirms
the results in Table 3. First, the pre-going private trend is flat, confirming that the pre-going private
productivity trends are similar for going private group and the control group. Second, there is no
statistically significant short-run or long-run improvement in relative productivity for the going-private
establishments, compared to the pre-going private productivity levels.
Going private may still have had positive productivity consequences if it was the case that these
firms were headed for a relative decline in productivity. In other words, could it be that going private
enabled these establishments to match the performance of the control establishments, whereas without
that they would have performed relatively worse? We see no evidence for this possibility in Figures 1
and 2. First, Figure 1 shows strong improving trends in all productivity measures for the going-private
establishments; so the prior absolute productivity trends do not portend future distress. Second, none
of the productivity measures in Figure 2 show any significant dip prior to going private; in fact, the
trends are remarkably flat for most of the measures from years -2 to 0. Thus, there is no hint that
without going private, the going private establishments would have suffered declines in productivity.
3.3 Addressing endogeneity of the going private decision and related
selection-bias
One potential concern in our study is that the decision to go private is not random and thus it is im-
portant that we control for the endogeneity of the going private decision as it may impact productivity
changes. In particular, if the choice of firms to go private were based on some characteristics that
predict future improvements in productivity, then the before-after results in section 3.1 are biased by
this endogenous selection of going private firms. The difference-in-differences approach in section 3.2
controls for this issue, but only if the key drivers of future productivity changes are related to one (or
a combination of) industry, age, or plant size related factors.
We therefore employ two alternatives approaches to constructing the control group. First, we
utilize the results in Bharath and Dittmar (2010) to construct a propensity score matched control
sample. By matching on the propensity score, we test whether the establishments that went private
16
show an improvement over-and-above the improvement exhibited by firms that had a similar prob-
ability of being selected into going private treatment (Rosenbaum and Rubin 1985). Bharath and
Dittmar (2010) find that despite the fact that, on average, the private sample firms remain in the
public market for over thirteen years, firms that ultimately go private are very different and discernible
in information and liquidity considerations, relative to firms that remain public, throughout their public
life and even at the time of the IPO. They estimate a logit model using explanatory variables only
at the year following the IPO to predict if a firm will ultimately go private. They find that firms
that are more likely to ultimately go private have less analyst coverage, less institutional holdings,
more concentrated ownership, and more mutual fund ownership at the time of the IPO compared
to firms that remain public, supporting the importance of information considerations in the choice
between remaining public or private. They also find that firms that go private are more illiquid and
have less share turnover, supporting the importance of liquidity issues. Using a Receiver Operating
Characteristics (ROC) analysis, they show that the logit model has a 71% accuracy compared to a
bench mark of 50% accuracy with a random guess, reflecting substantial improvement over a naive
model. Bharath and Dittmar also show that the evolution of these firm characteristics impacts the
decision to go public. In so much as myopic behavior or agency problems impacts the changes in these
variables over time, myopia may influence the decision to go private; however, Bharath and Dittmar
do not directly examine myopia as a motivation for going private.
Motivated by these results, we construct a sample of the firms with the closest propensity to the
going private score of each of the firms in the going private sample from a pool of firms that did
not go private. We then use their establishments as the matched sample controls. In calculating the
propensity to go private, we use the firm specific control variables at the time of the IPO as in Bharath
and Dittmar (2010). Specifically, we use log sales, analyst coverage, R&D, capital expenditures, a
dividend dummy, turnover, market to book, free cash flow, leverage, cash, net fixed assets, and the
number of past mergers to estimate the probability a firm will go private, as in column 2 of Table 7
of Bharath and Dittmar (2010) to estimate the propensity to go private. Since these control variables
are not available for all firms in our sample, the number of establishments of firms that went private
drops from 28,518 in Table 2 to 22,488 in these estimations.14 Similar to the approach in section
3.2, we include industry-propensity cell-year fixed effects in each regression. Here a ‘cell’ refers to the
unique (i.e., non-overlapping) group of establishments comprising one going private establishment and
the control group of establishments matched (based on propensity score and industry) to this going
private establishment. Then for each of the cells, we include cell-year fixed effects in the regressions.
Accordingly, as in section 3.2, the estimated effects are the mean of the relative difference between
14One noteworthy attribute of this control group (and another reason for the drop in the number of observations) is that
all of the firms in the control group are also listed firms, as the propensity model in Bharath and Dittmar (2010) uses stock
market related variables. In section 3.2, the control group was establishments of all non-going private firms, which include
both listed and unlisted firms.
17
each going-private establishment and its matched control group. For all inferences we compute p-
values based on standard errors clustered by industry-propensity score cells in the table.
In addition to using the propensity match, we also employ a third matching criteria to address the
potential bias from selection on past productivity. Kovenock and Phillips (1997) show that productivity
is (negatively) related to restructuring decisions such as plant closures. In our context, if relatively more
productive firms went private, then there could be less scope for improvements for these plants relative
to matched plants that may have lower initial productivity. While the lack of differential prior trends
in our baseline DID analysis assuages concerns on this count to some extent, one direct approach
to addressing this concern is to include ex-ante productivity as one of the matching variables.15
Specifically, instead of establishment size used in our baseline DID approach, we use the Blundell-
Bond TFP measure, and match each going private plant to up to eight establishments closest in this
productivity measure, from within the same 3 digit industry and age quartile.
The results from the difference in difference analysis using the two alternative approaches are
presented in Table 3 Panels A and B. Similar to Table 2, Panel A presents the full sample and Panel B
presents the subset of firms that are not subsequently sold. The first three columns in both panels show
that relative to firms with a similar propensity to go private, the sample firms exhibit no pre-existing
trend in productivity. More importantly, consistent with the results in Table 2, the firms exhibit no
significant improvement in either the short-run or long-run period after going private. Columns 4
through 6 of both panels of Table 3 present the results from difference in difference estimation with
the sample matched by past TFP. Here there is evidence of a pre-existing relative trend, with two of
the three measures showing an increase in productivity from the long run period prior to going private
to the short run period prior to going private. Thus, relative to other establishments with a similar
TFP level in the year before going private, the going private establishments’ productivity is on an
upward trajectory. However, similar to the results using the other control groups, the going private
firms do not show significant relative (DID) improvement in productivity between the pre and post
periods. Specifically, in all specifications, there is not significant DID increases in productivity, in both
the short- and long-run. On the OLS TFP measure, establishments have a statistically significant
decrease in productivity after going private relative to the control group, but the economic magnitude
is small: -2.7% over the six year long-run period, translating to less than -0.5% per year. Overall, this
evidence confirms that though firms that go private improve their productivity in absolute terms, this
improvement is not better than other similar firms.
15We thank one of the referees for suggesting this method.
18
4 Analysis of other outcomes
The previous results establish the fact that while labor and TFP productivity improves for the private
firm establishments, it does not improve differentially compared to the public firm establishments.
By implication, this analysis strongly suggests that there is no evidence for either ‘over’ or ‘under’
investment in listed firms, as sub-optimal investments should result in a negative effect on operational
efficiency, and hence lead to improvements in productivity post-going private.
Nevertheless, it is interesting and informative to examine investment directly and thus in this sec-
tion we examine the change in investment by studying two other sets of outcomes. First, we look
at capital and employment changes in establishments belonging to going-private firms. If the capital
market’s short-term outlook forces listed firms to sacrifice long term growth by reducing investment
and limiting expansions, we should expect to see greater investment and corresponding employment
expansion at the establishment level after firm’s go private.
Second, we examine a firm’s propensity to sell or close plants. Again, if market short-termism
causes firms to be impatient with short-term poor performance of some new projects or establish-
ments, we would expect to see a greater nurturing of investment in plants after going private, and
correspondingly we could expect to see a lower probability of selling off or shutting down plants after
going private.
4.1 Analysis of capital and employment
In this section, we examine the capital and employment choices around the going private decision. The
regression specifications are identical to that in section 3, except for the change in dependent variables.
Table 4 presents the regression results and Figure 3 summarizes the coefficients in the regression with
the confidence intervals. We present four specifications: the before and after, the difference in differ-
ence matching as we did in Table 2, difference in difference matching as we did in Table 3 including
past TFP, and difference in difference as we did in Table 3 matching by the propensity to go private.
We find that while log deflated capital for the private establishments increases by 2.5% (specification
1a) in the short run before-after comparison, they actually decline by 5 to 19% relative to the control
groups over this same period. Also, while in absolute terms capital increases by 6.4% in the long-run
after going private, again relative to the alternative control groups they show a large and significant
decline of about 15 to 36%. We also find that there was a statistically significant upward trend in
capital in the going private establishments in absolute terms (1.7% in Column 1a) and relative to the
control sample matched by ex ante-TFP, but there was no statistically significant prior trend relative
to the other control groups. Log employment shows a decline in absolute terms both in the short run
(5.8%) and the long term (8.9%) in specification 2a. This seems in line with a prior trend decline of
3.9%. This pattern of declines in employment is also seen relative to two of the three control groups
19
in specifications 1b and 1c. In specification 2d (propensity matched sample), while there is a decline,
the change is not significantly different from zero.
In summary, we find that establishment productivity demonstrates no significant difference after
going private relative to several control samples. And, establishments decrease investment in cap-
ital and employment after going private. Taken together, these results provide no support for an
acceleration of investment after going private; thus, public markets do not lead to under-investment.
If anything, the DID results (relative to other listed, propensity score matched firms) suggest that
a positive relative trend in capital and employment is reversed in a significant way after the going
private event.16 These results suggest that public markets lead firms to invest more rather than less
compared to private firms. Thus, these results contradict the commonly held view that market myopia
leads to under-investment. Taken alone, the results for capital and employment could suggest over-
investment, either because of myopia (such as in Bebchuk and Stole (1993), Bizjak et al (1993)), or
agency problems leading to empire-building in the going private firms when they were listed; however,
this interpretation is not consistent with difference-in-differences results showing that firms produc-
tivity does not change after going private, detailed in the previous section. In other words if firms
had been overinvesting when they were public, we would have expected to see an improvement in
productivity but we find no evidence of such a change relative to control groups.
4.2 Analysis of plant exit (shut-down and sales) decisions
In this section, we examine if firms are more nurturing of or patient with establishments after going
private. Because all of our analysis is based on examining going-private establishments that were
operational in the year before going private, the analysis here will essentially examine the exit rate for
gone-private establishments, relative to the industry, size, age control group. In other words, we will
be looking at differenced means, and we will not be doing a before-after or difference-in-differences
analysis, as essentially the sample conditions on no exit in the pre-going private period (so that there
is no “before” period, and hence no “difference-in-differences” analysis possible.) In this analysis, we
analyze all exits as well as separately examine sales and closings of plants. Because the exit data
derives from the Longitudinal Business database, it is not limited to manufacturing firms. Thus, in
our initial analysis of exits, we include a much larger sample than that used in the earlier tables.
However, in our primary analysis of exit rates, we will control for productivity which is only available
for manufacturing establishments, thus again limiting our sample the manufacturing firms.
16In the context of the productivity results, one relevant question is why the relative downsizing on the input side in capital
and employment did not translate into productivity gains; in untabulated results we find that sales declined in line with the
decreases in capital and employment, so that the input declines were not TFP enhancing.
20
4.2.1 Exit summary statistics
We begin with summary statistics of exit rates by industry (analogous to Table 2 in Kovenock and
Phillips 1997), various industry classifications (used in Maksimovic and Phillips 2008), time periods,
and acquirer type. This analysis is similar to presented in Maksimovic, Phillips and Prabhala (2011),
allowing us to compare our summary statistics for the going private transactions to their much broader
analysis of mergers and full-firm acquisitions. Table 5 Panel A presents the percent of establishments
in the going private sample that are sold or closed in the 3 to 6 years following the going private
transaction. As in Kovenock and Phillips (1997), the statistics show wide variance in exit rates (as
well as sale and closure rates) across industries: within these 16 industries the exit rates range from
27 to 58% over the first three years and 50 to 69% over six years, with the paper industry having the
highest exit rate and apparel and textiles having the lowest.
Panel B illustrates how the exit rates vary over time and by type of acquirer and industry classi-
fications. In addition to this, we provide summary statistics from Maksimovic, Phillips and Prabhala
(2011) labeled “MPP” for comparison. The top section of Table 5 panel B shows that the going
private firms exit plants (both through sale and closure) at a higher rate than the control sample over
the 3 and 6 year horizon. This pattern relative to controls is the same as in MPP as they find greater
exit and closure for target firm establishments as well. The exit rate in our sample over the three year
horizon (36.5%) is lower than that in the much broader MPP M&A sample; interestingly the closure
rate is higher in our sample (27.9% vs 18.6%), while the sale rate is much lower (8.6% vs 27.0%).
Thus, it appears that going private firm establishments exit strategy is more reliant on closure than
for the typical M&A target establishments. The patterns (and hence comparison to MPP) are roughly
consistent across acquirer types (in top row) and industry classifications (in bottom row). In the mid-
dle row, the split by the 1990s and 1980s show greater exit (both by sale and closure) in the 1990s,
which is consistent with the relative patterns found by MPP. In the bottom row, not surprisingly the
largest amount of exits (both sales and closures) are found in declining industries, both for the going
private sample and the control firms.17
4.2.2 Exit hazard and propensity analysis
To examine how changes in establishment characteristics and the private firm status impact the
probability of establishment exits, we first use a hazard model to investigate if and when a plant is
closed down or sold. Specifically, we are interested in the length of time it takes for a plant to be shut
17One difference between our sample statistics and MPP is that the exit rates for our control group is much closer to the
going private target sample than is the case for MPP, where the control group has starkly smaller exit rates. We speculate
that this could be because we use a more specific industry-size-age matched control group, while MPP look at all other firms
within the industry. Exit patterns are likely to be more similar once we condition on age and size – as is well documented in
the literature, and the hazard models in Table 6A confirm, that plant exit is strongly correlated with plant size and age.
21
down or sold from the date of going private, and the influence of different variables on that duration,
controlling for the fact that our comparison plants may also have exit decisions at some unobservable
time. In the baseline case, we use the exponential proportional hazard model. The model to be
estimated is:
h(t,X) = h(t, 0)exp(β′X)) = exp(β′X)) (4)
where h(t,X) is the hazard rate at time t for a firm with covariates X. A positive coefficient on variable
x in the hazard model implies that a higher x is linked to higher hazard rate and thus a lower expected
duration. The hazard ratio which is simply exp(β) tells us how much the hazard (i.e., instantaneous
risk) of exit increases for a unit change in the independent variable.18
In each estimation, the sample includes going private establishments and matched controls. As
explained in section 3.2 and 3.3, each going private establishment is matched to a unique set control
establishments (depending on data availability); each of the matched controls are assigned the same
“start/birth” year as the matched going private establishment. We use time invariant characteristics
(defined as at the year before going private) to explain duration, with the main variable of interest
being the going-private dummy variable. We include establishment log employment, age, 2-digit in-
dustry and year fixed effects as controls.19 That is, our goal is to understand whether the going private
establishments have a higher hazard for exit relative to the control group, controlling for size, age,
industry, and common year shocks.
Here, each plant we examine at the year before going private has two competing event outcomes
– sale or closure. We record only the earlier of the two outcomes; thus if a plant is sold, then it is no
longer at risk of being closed (as we are not interested in decisions made by the new owner), and vice
versa (i.e., after a plant is closed there is no possibility of it being sold). In the baseline analysis of sale
(closure) in Columns 2 and 3 of Panel A of Table 6A, exit into closure (sale) is treated as censoring. As
textbook discussions of competing risks (e.g. Pintilie 2006) point out, treating the competing event
as censoring can be misleading, and ideally we need a framework that adequately counts subjects that
experience the competing event as not having any chance of experiencing alternative event. We use
the competing risk model proposed by Fine and Gray (1999), which allows for assessing the effect of
covariates on the subhazard for each of the individual event.20
The results of this analysis are presented in Table 6, Panel A. The first three columns present the
18In untabulated results, we checked and found our results to be very similar when using the Cox proportional hazard model,
where the baseline hazard h(t, 0) is allowed to be arbitrary (instead of set to 1 in the exponential model); the Cox partial
likelihood estimator provides a way of estimating β without estimating h(t,0).19To clarify, for each establishment, there is only one observation, with the age and log employment defined as of the
year before going private. (For controls, the relevant year is the year before going private for the going private firm they are
matched to.)20This is implemented using the stata command stccreg. We thank the editor for suggesting this approach.
22
hazard rate of exiting, selling, or closing, respectively. Columns 4a and 4b present the subhazards from
the competing risk model. Compared to the industry-age-size matched control group, we find that
going private firms have statistically higher hazard rate of both closing and selling a plant. Specifically,
these establishments have a 28 (8)% higher hazard rate of selling (closing) a plant which translates
into a hazard rate of 1.3 (1.06). Using the competing risk model, going private establishments have
a close to 30% significantly higher hazard rate of closing or being sold; the competing risk analysis
suggests that the lower hazard for closure is downward biased by treating sales events as equivalent
to censoring. Overall these hazard rate results show no evidence of a lower exit rate for going private
firms’ plants and after controlling for the propensity to go private the change is not significant.
The main strength of the hazard rate model is that it explicitly accounts for the sizeable right
censoring that occurs in analysis such as ours (i.e., a sizeable number of establishments survive and
are neither sold nor closed till the last year of the dataset). However, we recognize that it is not
computationally feasible to use cell-year fixed effects in a hazard model, similar to how we did in the
previous section. Thus, in order to check robustness to using higher dimensional fixed effects, we
repeat this analysis using a linear propensity model in Panel B.21 To do this, we first define two exit
variables: 3- (6-) year exit dummy is a variable that equals one if the plant exited (was either sold or
closed) in the three (six) years after the going private event and zero otherwise. Similarly we define
two sale and closure dummy variables. The results are presented in Table 6 Panel B. All columns
include industry-size-age-year effects, so that the going private effect is evaluated relative to matched
control establishments in the same cell. We find that establishments of the going-private firms have
a 3.1% (5.1%) higher propensity than the industry-age-size matched control group to shut down a
plant with in the next 3 (6) years, immediately after going private. As in the hazard analysis, the
same patterns hold for both sale and closure: sale (closure) rate is 1.5% (1.5%) higher over the 3 year
window, and 4.3% (0.8%) higher over the six year window, with all estimates significant at the 1% level.
The analysis in Panel A and Panel B control for fixed and time varying industry effects, which
subsumes the effect of specific industry characteristics. In order to understand the impact of specific
industry characteristics and relate to previous literature that examined closures and sales, in Panels C
and D of Table 6, we repeat the propensity analysis including only year fixed effects, and covariates
for ex-ante (i.e., in year prior to going private) productivity, employment, number of plants, industry
capital utilization, industry concentration, change in industry aggregate output demand, and industry
classifications. Capacity utilization measure is taken from Gorodnichenko and Shapiro (2011); they
use the US Census Bureau’s survey of capacity utilization to construct industry specific capacity
utilization measures. Industry concentration is measured as in Kovenock and Phillips (1997) (KP1997)
21This approach has the drawback that it does not address the right censoring issue; however, our checks dropping plants
that survive to the end of the data period yielded very similar results to those reported in Panel B, so we are confident that
the qualitative conclusions are not biased by the right-censoring of the data.
23
as the market share of the top four plants in the industry. Change in output demand is defined as
aggregate change in industry output. Industry classifications used in Panel D are based on definitions
in Maksimovic and Phillips (2008) (MP2008).22 KP and MP in their work find these factors to be
important determinants plant exits.
Because we control for variables only available for manufacturing firms, the results in Panel C and
D also provide evidence of the robustness of the exit analysis for this subsample. We find that in
each specification and across both the three and six year horizons going private establishments have a
significantly higher rate of being sold or closed. Specifically, going private establishments have between
a 4.7 to 7.1% higher probability of being sold and a 3.9 to 5.3% higher probability of being closed than
the matched sample. Consistent with the findings in Table 3 of KP1997, plant closure (Column 3 in
panel C and D) is generally negatively correlated with plant productivity, size of firm (measured as firm
employment), and change in output demand. Interestingly, and consistent again with KP1997, number
of plants owned by the firm is positively correlated with closure; thus conditional on firm employment,
a plant belonging to a firm with more plants is more likely to be closed. As in KP1997, we find
negative effects of capacity utilization and lagged industry concentration (as in KP1997) but these are
weaker (not statistically significant). The pattern is similar for sales (Column 2) with respect to firm
employment and number of plants owned by the firm, but labor productivity is not strongly correlated
with sale propensity (so more or less productive plants are not systematically targeted for sale), and
change in industry demand and lagged industry concentration are weakly positively correlated with
sales over both the three and six year horizons. In panel D, we find the strongest closure rates for
declining industries, and the largest sales rates for consolidating industries, which is consistent with
what could be expected from the definitions of these industries in MP2008.
The results from the hazard and propensity models suggest that the going private firms accelerate
plant sales and shut downs after going private. Again, this evidence does not support a myopia-related
hypothesis that the elimination of stock market’s short-term focus make it more likely for firms to
nurture plants after going private.
4.2.3 Selection of plants for closure
In this subsection, we examine if the going private firms differentially target the poorly performing
plants (in a labor productivity and TFP sense) for closure. There are two motivations for doing this
analysis. First, this analysis could help understand the motivations for going private transactions, given
22Specifically, as discussed in MP (pp674): (1) Growth industries In Growth industries long-run industry shipments and
the long-run (which we define over the period 1981 to 2001 number of firms are increasing, and changes for each of these
factors are above the median industry change. (2) Consolidating industries In Consolidating industries the change in long-run
shipments is above the median industry change but the change in the number of firms is below the median. (3) Technological
Change industriesIn Technological Change industries, the change in long-run demand is below the median industry change
but the change in the number of firms is above the median. (4) Declining industriesIn Declining industries, the change in
long-run demand and the change in long-run number of firms are both below the median industry change.
24
that our earlier results show no gains in productivity at the establishment-level. If firms differentially
target lower productivity establishments for closure, this could help them to retain higher productive
plants in their portfolio.23 Second, arguably if the stock markets are indeed myopic, we would expect
to see market pressures leading to stronger targeting of worse performing plants by public firms (which
may need to be nurtured to achieve greater productivity levels). Thus, we predict a less negative effect
of productivity on shutdown decision after the firm goes private, as evidence of market myopia.
To test for differential targeting, we add the productivity measures as well as an interaction term
between the going private dummy and productivity, to the Hazard and OLS model specification in
Table 6 (panels A and B). Results are in Table 7. The control group is composed of establishments
matched on industry, age and initial size (as explained in section 3.2). The hazard model in Panel A
of Table 7 includes controls for employment and age (not reported for brevity), and industry and year
fixed effects, while the OLS model panel B includes much more detailed industry-size-age cell-year
fixed effects.
Across all the specifications, the coefficient estimates on the interaction term are always negative
when we look at the summary exit variable (Column 1) and sales decisions (Column 2). The same
is the case for closure with one exception (OLS FE TFP in Panel A). The effects are statistically
significant about half the cases, with magnitudes on the interaction term for exit in Panel B for labor
productivity (Blundell-Bond TFP) suggesting that plants with a 10% lower than mean productivity
would have a 0.4% (1%) higher propensity to be exited (closed or sold). The magnitudes suggest a
stronger targeting for sales in Panel B (as magnitudes for cross-terms for sales are higher), but this is
true only for the Blundell-Bond TFP measure in Panel B. Interestingly, across all the measures, the
average effect of productivity for sales (Column 2) is positive (and significant) in Panel A and close to
zero on Panel B, so that low productivity establishments are not targeted for sale on average, but the
negative interaction term shows that going private firms more aggressively target lower productivity
plants for sale, consistent with trying to improve the productivity of the portfolio of remaining plants,
and inconsistent with myopia.
Column 3 examines closure decisions, and two points are noteworthy here. First, we find that
the coefficient on productivity measures is always negative, consistent with Kovenock and Phillips
1997, and results in Table 6. This suggests that productivity (as measured here) is indeed informative
and guides the plant shut down decisions of all firms. Second, and more importantly for our study,
the coefficient on the interaction term of the productivity variables with the going private dummy
is generally negative (though not statistically significant in most cases). This suggests some weak
evidence that going private firms also more aggressively target less productive establishments for
23Also, while we do not have the data to verify this, firms may be selling the plants or assets of closed plants above the
pricing implicit in the going private transaction to gain value. Consistent with this, in untabulated results we find some
evidence that going private firms targeted plants in higher income areas for closure (presumably land and building in these
areas would be the most valuable).
25
closure, again consistent with attempting to improve productivity of their remaining portfolio and
contradicting the myopia story.
Taken together, these findings indicate that public firms do not seem to be unduly affected by
capital market pressures to be less patient with poorly performing plants; if anything, after going
private firms appear quicker to close and sell poorly performing plants.
4.3 Results by acquirer type
There are multiple ways a firm can go private, and our sample includes transactions that are driven by
private equity firms, management, and private operating firms. In the main analysis, we treat these
deals uniformly. However, it is possible that the productivity dynamics, as well as capital/investment
and shutdown choices differ by the parties involved in the transaction. We classify the sample firms
that went private into three categories: a buyout by a private operating firm, a buy out by a private
equity firm, a buy out by the management. We source the classifications for these deals using news
paper reports from Factiva. Specifically, we read news articles regarding each deal and classify each
into categories based on parties involved in the transactions. When it is unclear, we classify the deals
as unclassified. The category-types are non-exclusive, so that some deals may involve deals classified
under more than one type.
As discussed in the introduction, this analysis helps evaluate two alternative explanations for the
results documented thus far. One potential explanation is that going private transactions are highly
leveraged, and this leads to short term pressures even after going private, which in turn prevents
the firm from adopting long-term strategies to improve productivity, and also pressures the firm to
downsize and close/sell plants. Because leverage is more likely to be higher for private equity and
management buy-outs, the operating firm takeover subsample will allow us to examine results less
likely to be confounded by effects of high leverage. Second, independent of the extent of leverage,
investors driving going private transactions may have short horizons. Again, this is more likely to hold
for private equity transactions, than for operating firm buyouts (and arguably managers in MBOs also
have long-term horizons).
Table 1 Panel B shows that there are 819 establishments associated with a management buyout,
1,207 with a operating firm and 958 with a private equity acquirer, similar to the relative breakdown in
Bharath and Dittmar (2010). Thus 45% of our sample consists of operating firm acquisitions, which
are less likely to be affected by both leverage effects and private equity myopia effects.24
24To confirm the assumption that operating firm acquisitions do not have significant increases in leverage, we manually
collected leverage data for a random sub-sample of 30 such events for the year 1999 from the Capital IQ database. For
comparison, Guo, Hotchkiss and Song (2011) have 21 buyout firms in 1999 with post buy out data in their sample, which
26
In Table 8 Panel A, we present comparisons of analysis of productivity and TFP that is similar to
Table 2, and Panel B presents analysis of capital and labor similar to Table 4. Both are relative to
a control group composed of establishments matched on industry, age and initial size (as explained
in Section 3.2). The same analysis using establishments within 3-digit industry propensity (to go
private) score matched control group (described in Section 3.3), yield qualitatively similar results but
are omitted here for brevity (and are available on request from the authors).
Table 2 shows that on average going private establishments did not improve based on any measure
of productivity relative to a sample matched by Industry, size, age and year. In general, Table 8 Panel
A confirms this finding across each acquirer type. Specifically, no acquirer type shows any significant
difference in labor productivity or TFP using the OLS methods. There is weak evidence that unclassi-
fied and management-led buyouts had a decrease in the short run period after going private. But, this
result is inconsistent across different productivity measures. The only positive and significant change
in productivity are for those taken private by an operating firm using the Blundell Bond TFP measure
and only in the longer run. Again, the result is not consistent across TFP measures.25 Overall, the
evidence in the subgroup analysis is consistent with the full sample for each of the subgroups. Most
importantly, there is no evidence that operating firm takeovers, which are least susceptible to pressures
from high debt or impatient investors, show any improvements in productivity.
Table 4 showed evidence of a decrease in both labor and capital investment for the full sample. In
Table 8 Panel B, we find similar evidence of a decrease in capital investment across all acquirer types.
Thus, the decrease in investment is not driven by one subgroup, and again operating firm takeovers
show significant capital decline as well. The decrease in employment is consistent in sign, but not
significant for operating firms, and interestingly also not significant in the short-run for private equity
firms.
Table 8 panel C shows that management acquirers are in fact less likely to exit, which is driven
by their much lower propensity to close plants, though they do have a higher than control group
propensity to sell plants. Operating and private equity acquirers show a greater propensity to both
sell and close plants compared to their control groups. These results are consistent across the hazard
analysis in Panel C-A and the linear probability analysis in C-B. (One small exception is that private
equity firms seem less likely to close over the long term window (Column 3 of 6 year outcome dummy
they used to conclude that leverage increases dramatically after buy out events. We then examined average Long Term Debt
to Equity for the period before and after the acquisitions. We found that mean (median) leverage ratios for this sub-sample
actually drops by -5%(-1.25%). This confirms that operating firm acquisitions were not accompanied by large increases in
debt, so effects found for this sample of operating firm acquisitions are unlikely to be influenced by effects of excess leverage.25Further in Supplementary Appendix Table A.2, we find that this positive result is not robust to matching on ex-ante TFP.
27
in Panel C-B), which may be because they are most aggressive about exiting through sales.)
Finally, in Table 8, Panel D we replicate the OLS analysis in panel B of Table 7, examining whether
particular types of acquirers were more aggressive in targeting lower performing plants for exit (we got
similar results using the hazard model of panel A, Table 7). We find that the operating acquirers more
aggressively targeted low labor productivity plants for both sales and closures; the results for sales are
significant with the other two TFP measures as well, but the closure results are small and insignificant
(and positive with Blundell-Bond TFP). The interaction term is also systematically negative (though
not always statistically significant) for private equity acquirers, suggesting weak evidence that these
acquirers were also more aggressive in exiting low productivity plants. Interestingly, the estimates are
smaller and always insignificant for management acquirers; they do not seem to be more aggressive in
selling or closing low productivity plants. The results suggest that both operating and private equity
acquirers may be partly motivated by potential gains (say from high selling prices) through more
aggressive exit of poorly performing plants, while management does not seem to use this strategy.
Across Panel C and D, there is some evidence that Private equity acquirers are aggressive about
exiting (either through sale or closure), which could be consistent with “private equity myopia”;
however, the magnitudes of effects are similarly large for operating acquirers as well, so private equity
myopia is not the sole explanation for more aggressive exit found in the full sample.
5 Other robustness checks and discussion of results
In this section, we discuss a number of robustness checks and other tests that help to validate and
explain the conclusions from the baseline analyses. Results are available on request from the authors,
unless otherwise stated that they are included in a supplementary appendix available online.
(i) Examining outcomes at the acquirer firm: The above analyses document that there was no
change in productivity, an increased probability of sales or shutdown, and relative declines in employ-
ment and capital, at the going private establishments, relative to control groups. We interpret this as
evidence against capital market myopia or that myopia in the capital market is not worse than that
in the private market, as myopia would be expected to induce under- (or over-) investment relative
to what is optimal for productivity. It is possible that outcomes at the acquiring entity may be dif-
ferent in a way that affects this interpretation. To investigate this possibility, we use firm ownership
identification data in the Census LBD dataset to identify the acquirer firm (and its establishments at
the time of the going private event).26 We investigate three sets of outcomes. First, it could be the
26Firm ownership identifiers are not updated every year in the LBD, as documented by Jarmin and Miranda (2002).
Accordingly, all of the owner firm identifiers do not switch to that of the acquirer firm in the year after the going private
event. To overcome this limitation within the constraints of the data, for each going private establishment we track the firm
identifier and identify the first time it changes after the going private event. (Ideally, the firm identifier would change in the
year of the going private event; we allow for the possibility of delayed reporting by tracking later years and identifying the
28
case that the acquirer firm expands operations in its establishments, offsetting the declines in inputs
documented in the going private establishments. We analyzed changes in capital and employment
(and sales) at the acquirer firm establishments following the same approach as used for analyzing the
going private establishments (using a control group matched on industry, age and size at time of the
going private event). Contrary to what would be expected if there was a countervailing expansion
at acquirer firm plants, we found no difference-in-differences increases in capital or employment (or
sales) at the acquirer firm plants (both overall, as well as in a sample restricted to be in the industry
of the target going private firm) – in fact in most cases we found significant DID declines in inputs.
Second, it could be the case that the acquiring entity opens new establishments in the same industry
as the plants shutdown at the target going private firm. We analyzed the propensity to open a new
establishment and found that this propensity declines significantly in absolute terms at the acquirer
entities, and shows no differential change relative to a control group. Third, we checked whether
total factor productivity went up at acquirer establishment (say from a transfer of clients or markets
that the acquirer’s plants were able to serve without increasing inputs proportionately). We found
no significant DID changes in any of the three baseline productivity measures at the acquirer firms’
establishments, either in the short-term or the long-term. Thus, we conclude that there is limited
evidence for changes at the acquirer firm establishments offsetting (or complicating interpretation of)
changes documented at the going-private establishments.
(ii) Using operating profit measures as an alternative to productivity: As a further check to rule
out noise in productivity measures as an explanation for the baseline results, we examine two operating
profit measures: (i) gross operating profits, defined as sales less sum of materials cost, energy costs,
blue collar wage bill and white collar wage bill; and (ii) the ratio of gross operating profits to sales. We
find results (presented in Supplementary Appendix A.3) that are very similar to those in the baseline
productivity analysis; neither profitability measures show significant improvement after establishments
go private, relative to the two control groups. We also examine the ratio of different cost components
(materials, energy, blue collar wage bill and white collar wage bill) to sales, and find no significant
DID changes in any of these components.
(iii) Splitting the sample over time: It is possible that the responsiveness of managers to the stock
market’s short-term focus may have been exacerbated by the increasing use of stock options in exec-
utive compensation. Similarly, if leverage changes were driving any of our results, the use of leverage
was much higher in private equity deals in the 1980s than in 1990 and beyond. To test whether the
productivity and investment responses to going private have changed over time, we repeat the analysis
in Tables 2, 3, 4 and 5 separately for the going private transactions that occurred before and after
1992. In untabulated results, we find no notable differences between these samples; the qualitative
first change.) We then define as the acquirer firm the modal new firm identifier across all the acquired plants within a going
private firm.
29
conclusions of the analysis are the same for these samples separately as it was for the overall sample.
(iv) Relation to findings in Bharath and Dittmar (2010): Bharath and Dittmar examine why
firms go private, tracking firms from their IPO to the time they go private. They employ both a
hazard model using the full panel of data as well as a logit model using explanatory variables only
at the year following the IPO to predict if and when a firm will ultimately go private. As mentioned
earlier, they find that despite the fact that, on average, the private sample firms remain in the public
market for over thirteen years, firms that ultimately go private are very different and discernible in
information and liquidity considerations, relative to firms that remain public, throughout their public
life and even at the time of the IPO. However, the changes in the characteristics over time are also
important. Bharath and Dittmar show this by implementing a hazard model utilizing a control sample
that is matched by firm characteristics at the time of the IPO. They find that firms that are more
likely to ultimately go private have less analyst coverage, less institutional holdings, more concentrated
ownership, and more mutual fund ownership both at the time of the IPO and that these characteristics
change over time to increase the hazard rate of going private. Further, in the hazard model compared
to the matched sample, firms have a higher hazard rate of going private when they have a lower
market to book ratio. Though Bharath and Dittmar do not directly test for the importance of myopia
in the decision to go private, their results that the decline in market to book, turnover, and other
measures leave open the possibility that capital market myopia or agency conflicts could impact the
decision to go private. In this paper, we extend their findings to directly test for the impact of capi-
tal market myopia and agency problems on firms when they are private relative to when they are public.
(v) Other checks: We perform a number of other checks of the baseline results. These include (a)
checking our results to using plant fixed effects and industry-year effects (instead of cell-year) effects
in our DID specifications; (b) checking the split-by-acquirer-type results in Table 8 (Panel B and C)
using the event time figures; (c) redoing the baseline results adjusting for sampling weights (the base-
line analysis treats the ASM-CMF sample as an unbalanced panel) to check if sampling systematically
affects the going private sample differentially relative to the control groups. We find our results robust
to these checks.
6 Conclusion
An important critique of the stock market oriented U.S. financial system is that its excessive focus
on short term quarterly earnings forces public firms to behave in a myopic manner and that agency
conflicts due to disperse ownership leads managers to invest in ways that are inconsistent with share-
holders’ interests . We hypothesize that if U.S. firms are myopic in a manner that affects operational
efficiency and agency conflict impacts investing and operating decisions, then instances of going private
30
(when these constraints are eliminated) should cause U.S. firms to improve their establishment level
productivity (by focusing on long-term decisions) relative to peers. We find no evidence that this is the
case. Our key finding is that while there is evidence for substantial within establishment increases in
productivity (about 3% to 6%) after going private, there is little evidence of difference-in-differences
efficiency gains relative to a peer group of establishments constructed to control for industry, age,
initial size (at the time of going private) and the endogeneity of the going private decision effects. In
further evidence that is contrary to the standard myopia story, we find that going private firms contract
(decrease capital and employment), and exit (sell and close) plants more quickly than peer groups.
Our findings cast doubt on the view that stock markets force publicly listed firms to be short-sighted.
Our results provides an interesting contrast to Kaplan (1989) and Long and Ravenscraft (1993),
who find improvements in investment and accounting profits after a LBO. Similar to their studies, and
in our before and after analysis, we find that there is an increase in productivity and capital (though
not employment) after going private. The contribution of our study is that by employing census data
and a careful matching procedure, we are able to show that going private did not likely cause this
increase in productivity as the change is not significant once compared to an ex-ante similar sample.
However, the data, sample and comparison firms in both of these papers differ from that employed
here, so comparisons between our results and theirs must be done with caution. With this caveat
in mind, our findings suggest that using matching techniques to control for the industry, size, age,
productivity as well as the propensity to go private (using several variables shown to predict who will
go private), the improved productivity found in these earlier papers and shown in our before-after
analysis is not due to the change in the agency problems and capital market myopia of no longer
being a public firm. We therefore conclude that agency problems and capital market myopia in public
markets lead to no greater impact on productivity than that experienced by similar private firms.
References
[1] Ackerberg, D. , K. Caves and G. Frazer, 2006, “Identification of Production Functions”, Working
Paper, UCLA.
[2] Asker, J, J Farre-Mensa, and A. P. Ljungqvist, 2012, “Comparing the Investment Behavior of
Public and Private Firms”, ECGI Working Paper Series in Finance, No. 282/2010.
[3] Bebchuk, L.A., and L.A. Stole, 1993, “Do Short-Term Objectives Lead to Under- or Over- Invest-
ment in Long-Term Projects?”, Journal of Finance, 48, 719-729.
[4] Becker, G. and W. Gray, 2009, NBER Productivity database.
[5] Bharath, S.T., A. Dittmar, 2010, “Why do firms use private equity to opt out of public markets?”,
Review of Financial Studies, 23(5), 1771-1818.
[6] Bhojraj, S., P. Hribar, M.Picconi, and J. McInnis, 2009, “Making Sense of Cents: An Examination
of Firms That Marginally Miss or Beat Analyst Forecasts”, Journal of Finance, 64(5), 2361-2388.
[7] Bizjak, J., J. Brickley,and J. Coles, 1993, “Stock-Based Incentive Compensation, Asymmetric
Information, and Investment Behavior”, Journal of Accounting and Economics, 16, 349-372.
31
[8] Blundell, R., and S. Bond, 1998, “Initial conditions and moment restrictions in dynamic panel
data models.”, Journal of Econometrics, 87: 115-43.
[9] Blundell, R., and S. Bond, 2000, “GMM estimation with persistent panel data: an application to
production functions”, Econometric Reviews, 19, pp. 321-340.
[10] Bond, S., and M. Soderbom, 2005, “Adjustment costs and the identification of Cobb Douglas
production functions”, IFS Working Papers W05/04, Institute for Fiscal Studies.
[11] Campbell, T., and A. Marino, 1994, “Myopic investment decisions and competitive labor mar-
kets”, International Economic Review 35(4), 855875.
[12] Davis, S., Haltiwanger, J., Jarmin, R., Lerner, J., and Miranda, J., 2008a, “Private Equity and
Employment”, In Gurung, A. and Lerner, J. (eds.) Globalization of Alternative Investments Working
Papers Volume 1: Global Economic Impact of Private Equity, New York: World Economic Forum
U.S.A, 43-64.
[13] Davis, S., Haltiwanger, J., Jarmin, R., Lerner, J., and Miranda, J., 2008b, “Private Equity, Jobs
and Productivity”, Working paper, University of Chicago.
[14] Fine, J. P., and R. J. Gray., 1999, “A proportional hazards model for the subdistribution of a
competing risk”, Journal of the American Statistical Association 94: 496509.
[15] Y. Gorodnichenko and M. Shapiro, 2011, “Using the Survey of Plant Capacity to Measure Capital
Utilization,” with Matthew D. Shapiro, US Census Bureau Center for Economic Studies Discussion
Paper 11-18.
[16] Graham, John R., Campbell R. Harvey, and Shiva Rajgopal, 2005, “The economic implications
of corporate financial reporting”, Journal of Accounting and Economics, 40, pp. 3-73.
[17] Greenstone, Michael., Rick Hornbeck, and Enrico Moretti, Identifying Agglomeration Spillovers:
Evidence from Winners and Losers of Large Plant Openings, Journal of Political Economy, 2010,
118 (3): 536-598.
[18] Griliches, Zvi and Jacques Mairesse, 1995, “Production Functions: The Search for Identification”,
NBER Working Paper No 5067, March 1995.
[19] Guo, S., Hotchkiss, E., Song, W., 2011, “Do buyouts (still) create value?” Journal of Finance
66, 479-517.
[20] Holmstrom, B., and J. Ricart i Costa, 1986, “Managerial incentives and capital management”,
Quarterly Journal of Economics 101(4), 835860.
[21] Jarmin, R. S. and J. Miranda, 2002, “The Longitudinal Business Database”, Technical Report,
US Census Bureau.
[22] Jensen, M.C., 1986, “Agency costs of free-cash-flow, corporate finance, and takeovers”, American
Economic Review 76, 323-329
[23] Jensen, M. C. and Meckling, W. H., “Theory of the Firm: Managerial Behavior, Agency Costs
and Ownership Structure”, Journal of Financial Economics, 3, 305-360
[24] Kaplan, S.N., 1989a, “The Effects of Management Buyouts on Operating Performance and
Value”, Journal of Financial Economics, 24, 217-254.
[25] Kaplan, S.N., 1989b, “Management Buyouts: Evidence on Taxes as a Source of Value”, Journal
of Finance 44(3), 611-632.
32
[26] Kaplan, S.N., 1991, The Staying Power of Leveraged Buyouts, Journal of Financial Economics
29, 287-314.
[27] Kovenock, D. and Phillips, G. M, 1997, “Capital Structure and Product Market Behavior: An
Examination of Plant Exit and Investment Decisions,” Review of Financial Studies, vol. 10(3), pages
767-803.
[28] Levinsohn, J. and A. Petrin, 2003, ”Estimating Production Functions Using Inputs to Control for
Unobservables,” Review of Economic Studies, Vol. 70(2), No. 243, pp. 317-342
[29] Lichtenberg, F. R., 1992, Corporate Takeovers and Productivity, Cambridge: MIT Press.
[30] Lichtenberg, F. R., and D. Siegel, 1991, “The Effects of Leveraged Buyouts on Productivity and
Related Aspects of Firm Behavior”, Journal of Financial Economics, 27:1, pp. 165-94.
[31] Maksimovic, V., and G. Phillips, 2001, “The Market for Corporate Assets: Who Engages in
Mergers and Asset Sales and are there Gains?”, Journal of Finance, 56(6), 2019-2065.
[32] Maksimovic, V., and G. Phillips, 2002, “Do Conglomerate Firms Allocate Resources Inefficiently
Across Industries?”, Journal of Finance, 57(2), 721-767.
[33] Maksimovic, V., and G. Phillips, 2008, “The Industry Life Cycle, Acquisitions and Investment:
Does Firm Organization Matter?”, Journal of Finance, 63(2), 673-708.
[34] Maksimovic, Vojislav, G. Phillips, and N.R. Prabhala, 2011, “Do Conglomerate Firms Allocate
Resources Inefficiently Across Industries?”, Journal of Financial Economics, 2011, 102 317-43.
[35] Marschak, J. and W. Andrews, 1944, “Random Simultaneous Equations and the Theory of
Production, Econometrica 12, pp. 143-205.
[36] McGuckin R.H., and Nguyen S.V., 1995, “On Productivity and Plant Ownership Change - New
Evidence from the Longitudinal Research Database“, Rand Journal of Economics, 26 (2): 257-276
SUM.
[37] Narayanan, M.P, 1985, “Managerial Incentives for Short-term Results“, Journal of Finance 40,
1469-1484.
[38] Olley, S., and A. Pakes, 1996, “The Dynamics of Productivity in the Telecommunications Equip-
ment Industry”, Econometrica, 64(6), 1263-1298.
[39] Pintilie, M., 2006, Competing Risks: A Practical Perspective. Chichester, UK: Wiley.
[40] Porter, M.E., 1992, “Capital Disadvantage: America’s Failing Capital Investment System”, Har-
vard Business Review, September-October.
[41] Rosenbaum, P., and D. Rubin, 1985, “Reducing Bias in Observational Studies Using Subclassi-
fication on the Propensity Score,” Journal of the American Statistical Association, 79, 516-524.
[42] Rumelt, 1987, “Theory, Strategy, and Entrepreneurship,“ in David Teece (ed.) The Competitive
Challenge: Strategies for Industrial Innovation and Renewal. Cambridge, Mass.: Ballinger, 1987,
137-158.
[43] Schoar A, 2002, “Effects of corporate diversification on productivity”, Journal of Finance 57 (6):
2379-2403.
[44] Shleifer, A., and Vishny, R., 1997. “A survey of corporate governance”, Journal of Finance 52:
737-783.
[45] Stein, J. C., 1989, “Efficient capital markets, inefficient firms: A model of myopic corporate
behavior”, Quarterly Journal of Economics 104, 655-669.
33
Data Appendix
The Census of Manufactures (CMF) covers all establishments is a quinquennial census that isundertaken in years 1977, 1982, 1987, 1992, 1997 and 2002. For the other years used in ouranalysis, we use data from the ASM that surveys: (i) All establishments with greater than (orequal to) 250 employees; (ii) All establishments of multi-unit firms; and (iii) a stratified random-ized sample of establishments with less than 250 employees. For certain small establishmentsin the CMF, the employment data is imputed based on reported payroll from administrativerecords data. Following the practice in the literature, such establishments (which are flaggedby an ‘Administrative Records’ dummy variable) are excluded from our analysis. As very few(less than 1%) of the going-private establishments belong to this category, this exclusion hasvery little impact on our sample size.
Key variables used in the analysis are as defined below. Deflators used for obtaining realvalues are taken from the NBER-CES manufacturing industry database (Becker and Gray 2009).
(i). Output measures
(a) Log real sales is defined as value of shipments deflated using 4-digit SIC industry-specific output deflators.
(b) Log real value added is defined as log of (real sales - real materials - real energy costs).
(ii). Input measures
(a) Log employment is the log of the total number of employees reported in the ASM-CMFdatabase.
(b) Log real materials is the log of the deflated cost of materials used.(c) Log real energy costs is the log of the deflated cost of fuel, electricity and other energy
sources used.(d) Log real capital is defined as the log the real depreciated capital stock. The real
depreciated capital stock is constructed using the perpetual inventory method. Thedepreciation rates (and deflators) used to construct the plant specific real depreciatedstructures and equipment stocks were taken from Becker and Gray, 2009.
(iii). Blundell-Bond system-GMM TFP measureBasic definitions of the productivity measures are provided in the text. Here we describeelaborate on the specific methodology used to define the Blundell-Bond total factor pro-ductivity measure (more detail is available in Blundell and Bond 2000, and Bond andSoderbom 2005).As in Blundell and Bond (2000), we assume a gross output production function with anAR1 component in the productivity term:
yit = βl.lit + βn.nit + βk.kit + βmmit + βeeit + ηi + νit +mit
νit = ρνit−1 + ϵit |ρ| < 1ϵit,mit ∼ MA(0)
where output and inputs are as defined in section 2.2. The model has a dynamic (commonfactor representation):
yit = πl.lit + π2.lit−1 + π3.nit + π4.nit−1 + π5.kit + π6.kit−1 + π7mit + π8mit−1
+π9eit + π10eit + π11yit−1 + η∗i + ωit
subject to 5 common factor restrictions: π2 = −π1 ∗π11, π4 = −π3 ∗π11, π6 = −π5 ∗π11,π8 = −π7 ∗ π11 and π10 = −π9 ∗ π11, and where η∗i = ηi(1− ρ). The standard Arellano-Bond moment (1991) conditions are
E[xit−j∆ωit] = 0 where xit = (lit, nit, kit,mit, eit, yit)
34
for j ≥ 3 (assuming ωit ∼ MA(1)). This allows the use of suitably lagged levels of thevariables as instruments, after the equation has been first differenced to eliminate fi, theplant specific fixed effects. Blundell and Bond (2000) show that by assuming input andoutput in first differences depend only on the history of the productivity shock till time tbut do not depend on the fixed effect fi, one can obtain additional moment conditionsthat use lagged first differences as valid instruments for the equation in levels which greatlyimprove upon the properties of the estimator. These conditions are:
E[∆xit−j(η∗i + ωit)] = 0 where xit = (lit, nit, kit,mit, enit, yit) (5)
for j=2 (assuming ωit ∼ MA(1)). Both sets of moment conditions can be exploited as alinear GMM estimator in a system containing both first-differenced and levels equations.Combining both sets of moment conditions provides the Blundell and Bond (2000) systemGMM estimator. The underlying production function parameter estimates are recoveredby imposing the common factor restrictions using a minimum distance estimator.In the literature on production function estimation, two innovative two-stage approacheswere suggested by Olley and Pakes (OP) (1996) and Levinsohn and Petrin (LP) (2003),based on using investment and other inputs as proxies to condition out endogenous partof the productivity in the first stage. However, these approaches have been critiqued byAckerberg, Caves and Frazer (2006). Importantly, Bond and Soderbom (2005) show thatthe Blundell and Blond (2000) estimator addresses a critique of the OP and LP approachesput forth by Ackerberg, Caves and Frazer (2006).
35
Table 1 Panel A: Sample characteristics, by years around the “going-private” decisionThe data used to construct this sample is taken from the Census of Manufactures (CMF) and the Annual Survey
of Manufactures (ASM). For each establishment in the going private sample, we include upto eight establishments
(based on data availability) that are closest in size (employment) to the going private establishment from within
the same 3-digit SIC industry, and belonging to the same age quartile as controls.
Years Total Number Number Total Number Numberfrom number of of matched number of going- of controlgoing of going-private control of private firmsprivate establishments establishments establishments firms firms
-6 13,368 2,189 11,179 7,210 490 6,720
-5 13,808 2,198 11,610 7,284 469 6,815
-4 15,907 2,480 13,427 8,368 447 7,921
-3 16,017 2,413 13,604 8,546 449 8,097
-2 17,483 2,652 14,831 9,410 440 8,970
-1 20,114 2,701 17,413 11,255 406 10,849
0 16,095 2,274 13,821 8,567 421 8,146
1 15,515 2,280 13,235 7,984 454 7,530
2 14,350 2,228 12,122 6,821 469 6,352
3 12,961 2,020 10,941 6,048 462 5,586
4 11,684 1,886 9,798 5,339 420 4,919
5 9,724 1,693 8,031 4,495 378 4,117
6 8,883 1,504 7,379 4,358 371 3,987
Total 185,909 28,518 157,391 95,685 5,676 90,009
36
Table 1 Panel B: Sample characteristics: Breakdown by acquirer typeWe classify the sample firms that went private into three categories: buyouts by (i) private operating firms, (ii)
private equity firms, and (iii) management. We source the classifications for these deals using newspaper reports
from Factiva. The residual category is unclassified. The category-types are non-exclusive, so that some deals are
classified under more than one type; hence the net total number is not the sum of the numbers in the sub-categories.
The data used to construct this sample is taken from the Census of Manufactures (CMF) and the Annual Survey
of Manufactures (ASM). For each establishment in the going private sample, we include upto eight establishments
(based on data availability) that are closest in size (employment) to the going private establishment from within
the same 3-digit SIC industry, and belonging to the same age quartile as controls.
Acquirer Total Number Number Total Number Numbertype number of of matched number of going- of control
of going-private control of private firmsestablishments establishments establishments firms firms
Unclassified 2,157 259 1,898 1,186 48 1,138
Management 5,966 819 5,147 3,426 112 3,314
Operating 9,056 1,207 7,849 5,020 188 4,832
Private equity 6,838 958 5,880 3,761 115 3,646
Net total 20,114 2,701 17,413 11,255 406 10,849
37
Table 1 Panel C: Characteristics of going private (target) establishments: Breakdown byacquirer typeThis table presents sample characteristics of the going private establishments. The figures in the last 3rows are demeaned using the 3 digit SIC industry mean for the sample.
Characteristics Going Unclassified Operating Private Management(based on Year -1 to Year -3) Private Sample Equity
N=7766 N=721 N=3524 N=2734 N=2373
Mean Mean Mean Mean Mean
Log (deflated capital) 8.575 8.430 8.699 8.481 8.563
Log (employment) 4.937 4.608 5.010 4.889 5.058
Labor Productivity 3.318 3.304 3.296 3.422 3.293
OLS TFP 3.618 3.571 3.606 3.683 3.621
Blundell-Bond TFP 2.426 2.422 2.405 2.456 2.432
Characteristics N=2701 N=259 N=1207 N=958 N=819
(based on Year -1 ) Mean Mean Mean Mean Mean
Number of Plants Per Firm 27.146 20.259 19.499 41.992 42.958
Number of Industries 7.401 4.259 5.843 10.795 12.247
Firm Sales ($, ’000s) 1576 327 1778 1964 2130
Labor Productivity (Industry demeaned) 0.056 0.045 0.028 0.130 0.059
OLS TFP (Industry demeaned) 0.040 0.055 0.025 0.075 0.033
Blundell-Bond TFP (Industry demeaned) -0.006 -0.016 -0.024 0.036 -0.013
38
Table 1 Panel D: Characteristics of acquirer establishments: Breakdown by acquirer typeThis table presents sample characteristics of the acquirer establishments. Because the acquirer forManagement takeovers is the management of the original firm, we exclude that category from this table.(For this category, the figures in Panel C above can be considered as applicable for acquirers as well.) Thefigures in the last 3 rows are demeaned using the 3 digit SIC industry mean for the sample.
Going Private Sample Unclassified Operating Private Equity
Acquirer Characteristics N=5088 N=212 N=2729 N=1918
(based on Year -1 to Year -3) Mean Mean Mean Mean
Log (deflated capital) 8.383 8.424 8.543 8.143
Log (employment) 4.661 4.941 4.694 4.482
Labor Productivity 3.529 3.023 3.616 3.56
OLS TFP 3.665 3.373 3.696 3.668
Blundell-Bond TFP 2.519 2.408 2.524 2.533
Acquirer Characteristics N=1756 N=73 N=948 N=657
(based on Year -1 ) Mean Mean Mean Mean
Number of Plants Per Firm 44.726 10.699 28.082 74.906
Number of Industries 10.380 4.849 8.028 14.327
Firm Sales ($, ’000s) 2150 271 2210 2414
Labor Productivity (Industry demeaned) 0.195 -0.167 0.2163 0.240
OLS TFP (Industry demeaned) 0.094 -0.070 0.1 0.112
Blundell-Bond TFP (Industry demeaned) 0.059 -0.046 0.057 0.078
39
Table 2 Panel A: Changes in productivity around the “going-private” decision: Before-After(BA) and Difference-in-Differences (DID) specificationsThis table specifications 1-3 presents regression results for each of the productivity measures for the sample of
establishments of firms that went private. The data used to construct this sample is taken from the Census of
Manufactures (CMF) and the Annual Survey of Manufactures (ASM). Dummy variables LR PRE equals 1 for
years -6 to -4 from going private and zero otherwise, SR PRE equals 1 for years -3 to -1 from going private
and zero otherwise, SR POST equals 1 for years 0 to 3 from going private and zero otherwise, and LR POST
equals 1 for years 4 to 6 from going private and zero otherwise. P-values based on standard errors clustered by
establishment are in parentheses. This table specifications 4-6 presents difference-in-differences regression results
for each of the different productivity measures. For each establishment in the going private sample, we include
upto eight establishments (based on data availability) that are closest in size (employment) to the going private
establishment from within the same 3-digit SIC industry, and belonging to the same age quartile as controls.
P-values based on standard errors clustered by industry-size-age cells are in parentheses.
Labor OLS Blundell- Labor OLS Blundell-productivity TFP Bond TFP productivity TFP Bond TFP
Before-After DiD
1 2 3 4 5 6
LR PRE 3.237 3.593 2.389 0.066 0.030 0.000(0.000) (0.000) (0.000) (0.000) (0.002) (0.983)
SR PRE 3.342 3.627 2.426 0.080 0.037 0.005(0.000) (0.000) (0.000) (0.000) 0.000 (0.543)
SR POST 3.405 3.659 2.463 0.090 0.027 (0.001)(0.000) (0.000) (0.000) (0.000) (0.005) (0.930)
LR POST 3.414 3.691 2.477 0.085 0.026 0.006(0.000) (0.000) (0.000) (0.000) (0.028) (0.592)
CHANGES RELATIVE TO SR PRESR POST - SR PRE 0.063 0.032 0.037 0.011 (0.011) (0.006)
(0.000) (0.000) (0.000) (0.510) (0.202) (0.479)
LR POST - SR PRE 0.072 0.064 0.051 0.005 (0.011) 0.001(0.000) (0.000) (0.000) (0.817) (0.374) (0.924)
TEST FOR PRE-EXISTING TREND
SR PRE - LR PRE 0.105 0.034 0.037 0.014 0.008 0.005(0.000) (0.000) (0.000) (0.362) (0.342) (0.492)
Fixed effects Plant Plant Plant Industry-size-age-yearNumber of Observations 28,518 28,518 28,518 185,909 185,909 185,909
40
Table 2 Panel B: Changes in productivity around the “going-private” decision: BA and DIDspecifications, excluding sold plants (from both going-private and control samples)The data used to construct this sample excludes all sold plants (i.e., those that underwent a change in ownership
after the going private date) in both the treatment and control samples and is taken from the Census of
Manufactures (CMF) and the Annual Survey of Manufactures (ASM). This table specifications 1-3 presents
regression results for each of the productivity measures for the sample of establishments of firms that went private.
Dummy variables LR PRE equals 1 for years -6 to -4 from going private and zero otherwise, SR PRE equals 1
for years -3 to -1 from going private and zero otherwise, SR POST equals 1 for years 0 to 3 from going private
and zero otherwise, and LR POST equals 1 for years 4 to 6 from going private and zero otherwise. P-values
based on standard errors clustered by establishment are in parentheses. This table specifications 4-6 presents
difference-in-differences regression results for each of the different productivity measures. For each establishment
in the going private sample, we include upto eight establishments (based on data availability) that are closest in
size (employment) to the going private establishment from within the same 3-digit SIC industry, and belonging
to the same age quartile as controls. P-values based on standard errors clustered by industry-size-age cells are in
parentheses.
Labor OLS Blundell- Labor OLS Blundell-productivity TFP Bond TFP productivity TFP Bond TFP
Before-After DiD
1 2 3 4 5 6
LR PRE 3.250 3.619 2.401 0.070 0.033 0.000(0.013) (0.006) (0.006) (0.023) (0.013) (0.011)
SR PRE 3.372 3.656 2.444 0.088 0.041 0.003(0.010) 0.005 (0.005) (0.021) (0.012) (0.011)
SR POST 3.439 3.689 2.484 0.097 0.030 -0.005(0.008) 0.005 (0.004) (0.023) (0.013) (0.011)
LR POST 3.463 3.718 2.507 0.101 0.033 0.010(0.008) (0.005) (0.004) (0.030) (0.016) (0.016)
CHANGES RELATIVE TO SR PRESR POST - SR PRE 0.189 0.070 0.083 0.009 (0.010) (0.008)
(0.019) (0.010) (0.009) (0.679) (0.367) (0.450)
LR POST - SR PRE 0.213 0.099 0.106 0.013 (0.008) 0.007(0.025) (0.013) (0.012) (0.690) (0.655) (0.663)
TEST FOR PRE-EXISTING TRENDSR PRE - LR PRE 0.122 0.037 0.043 0.018 0.008 0.003
(0.016) (0.007) (0.007) (0.384) (0.468) (0.746)
Fixed effects Plant Plant Plant Industry-size-age-year
Number of Observations 20,788 20,788 20,788 139,826 139,826 139,826
41
Table 3 Panel A: Changes in productivity measures around the “going-private” decision: DIDspecifications using alternative propensity score and Past TFP matched control groupsThis table presents 3 regression results for each of three productivity measures for the sample of establishments
of firms that went private. In Columns 1 to 2, for each establishment in the going private sample, we construct
the closest propensity to go private score matched firm(s) but did not go private, as control firm(s). We use the
firm specific control variables at the time of the IPO to estimate the propensity to go private as in Bharath and
Dittmar (2010). Since these control variables are not available for all firms (whose establishments we consider
in the regression), the number of establishments of firms that went private drops to 22,488 (from 28,518 in
table 2) in these propensity-matched estimations. In specifications 4-6 we match establishments based on past
TFP (using the Blundell-Bond measure) within the same industry and age quartile. The data used to construct
these samples is taken from the Census of Manufactures (CMF) and the Annual Survey of Manufactures (ASM).
Dummy variables LR PRE equals 1 for years -6 to -4 from going private and zero otherwise, SR PRE equals 1 for
years -3 to -1 from going private and zero otherwise, SR POST equals 1 for years 0 to 3 from going private and
zero otherwise, and LR POST equals 1 for years 4 to 6 from going private and zero otherwise. P-values based
on standard errors clustered by industry-propensity cell year (specifications 1-3) and industry-past TFP cell year
(specifications 4-6) are in parentheses.
Labor OLS Blundell- Labor OLS Blundell-productivity TFP Bond TFP productivity TFP Bond TFP
1 2 3 4 5 6
LR PRE -0.052 -0.018 -0.002 0.0531 0.0429 (0.005)
(0.238) (0.489) (0.942) (0.015) (0.008) (0.007)
SR PRE -0.004 0.004 0.006 0.085 0.0763 0.002
(0.916) (0.874) (0.769) (0.012) (0.006) (0.004)
SR POST -0.030 -0.026 -0.015 0.0998 0.058 (0.002)
(0.445) (0.273) (0.484) (0.015) (0.008) (0.007)
LR POST 0.036 0.033 0.037 0.098 0.049 (0.006)
(0.509) (0.292) (0.189) (0.021) (0.011) (0.011)
CHANGES RELATIVE TO SR PRE
SR POST - SR PRE -0.026 -0.030 -0.021 0.015 (0.019) (0.0038)
(0.497) (0.187) (0.310) (0.319) (0.012) (0.584)
LR POST - SR PRE 0.040 0.029 0.031 0.0122 (0.0271) (0.0036)
(0.488) (0.366) (0.292) (0.568) (0.017) (0.490)
TEST FOR PRE-EXISTING TREND
SR PRE - LR PRE 0.048 0.022 0.007 (0.032) (0.033) (0.007)
(0.246) (0.339) (0.720) (0.025) (0.000) (0.245)
Fixed effects Industry-propensity cell-year Industry-past TFP-age-year
Number of Observations 55,358 55,358 55,358 179,043 179,043 179,043
42
Table 3 Panel B: Changes in productivity measures around the “going-private” decision:DID specifications using alternative propensity score and Past TFP matched control groups,excluding sold plantsThe data used to construct this sample excludes all sold plants (i.e., those that underwent a change in ownership
after the going private date) in both the treatment and control samples. For each establishment in the going
private sample, we construct the closest propensity to go private score matched firm(s) but did not go private, as
control firm(s). We use the firm specific control variables at the time of the IPO to estimate the propensity to
go private as in Bharath and Dittmar (2010). Since these control variables are not available for all firms (whose
establishments we consider in the regression) in our sample, the number of establishments of firms that went
private drops in these estimations. In specifications 4-6 we match establishments based on past TFP (using the
Blundell-Bond measure) within the same industry and age quartile. The data used to construct these samples is
taken from the Census of Manufactures (CMF) and the Annual Survey of Manufactures (ASM). Dummy variables
LR PRE equals 1 for years -6 to -4 from going private and zero otherwise, SR PRE equals 1 for years -3 to -1
from going private and zero otherwise, SR POST equals 1 for years 0 to 3 from going private and zero otherwise,
and LR POST equals 1 for years 4 to 6 from going private and zero otherwise. P-values based on standard errors
clustered by industry-propensity cell year (specifications 1-3) and industry-past TFP cell year (specifications 4-6)
are in parentheses.
Labor OLS Blundell- Labor OLS Blundell-
productivity TFP Bond TFP productivity TFP Bond TFP
1 2 3 4 5 6
LR PRE -0.034 -0.031 -0.002 0.0486 0.0479 -0.010
(0.072) (0.045) (0.039) (0.020) (0.010) (0.008)
SR PRE -0.007 0.010 0.003 0.105 0.0905 0.001
(0.059) (0.036) (0.031) (0.015) (0.008) (0.005)
SR POST -0.025 -0.019 -0.020 0.1080 0.066 -0.005
(0.068) (0.042) (0.038) (0.021) (0.011) (0.009)
LR POST 0.023 0.034 0.027 0.111 0.060 -0.005
(0.094) (0.060) (0.055) (0.028) (0.015) (0.015)
CHANGES RELATIVE TO SR PRE
SR POST - SR PRE -0.019 -0.029 -0.023 0.003 -0.025 -0.007
(0.776) (0.457) (0.513) (0.876) (0.015) (0.447)
LR POST - SR PRE 0.030 0.024 0.023 0.006 -0.030 -0.006
(0.758) (0.703) (0.677) (0.821) (0.058) (0.674)
TEST FOR PRE-EXISTING TREND
SR PRE - LR PRE 0.027 0.041 0.005 0.056 0.043 0.011
(0.695) (0.282) (0.883) (0.003) (0.000) (0.139)
Fixed effects Industry-propensity cell-year Industry-past TFP-age-year
Number of Observations 36,659 36,659 36,659 136,218 136,218 146,220
43
Table 4: Changes in capital and employment around the “going-private” decision : Before-Afterand Difference-in-Differences Specifications
This table presents before-after and difference-in-differences regression results for capital and employment. In
DID1, for each establishment in the going private sample, we include upto eight establishments (based on data
availability) that are closest in size (employment) to the going private establishment from within the same 3-digit
SIC industry, and belonging to the same age quartile as controls. In DID2, for each establishment in the going
private sample, we include upto eight establishments (based on data availability) that are closest in past TFP
to the going private establishment from within the same 3-digit SIC industry, and belonging to the same age
quartile as controls. In DID3, we include upto eight establishments within the same 3-digit SIC industry closest in
propensity to go private (but whose owner firms did not go private), as control establishments. The data used to
construct this sample is taken from the Census of Manufactures (CMF) and the Annual Survey of Manufactures
(ASM). Dummy variables LR PRE equals 1 for years -6 to -4 from going private and zero otherwise, SR PRE
equals 1 for years -3 to -1 from going private and zero otherwise, SR POST equals 1 for years 0 to 3 from
going private and zero otherwise, and LR POST equals 1 for years 4 to 6 from going private and zero otherwise.
P-values based on standard errors clustered by plant establishments (before-after), industry-size-age cells (DID1),
industry-past TFP-age cells (DID2) and industry-propensity cells (DID3) are in parentheses.
Log deflated capital Log employment
Before-After DID1 DID2 DID3 Before-After DID1 DID2 DID3
1a 1b 1c 1d 2a 2b 2c 2d
LR PRE 8.642 0.181 0.369 -0.223 5.033 0.043 0.204 -0.134
(0.000) (0.000) (0.034) (0.012) (0.000) (0.000) (0.025) (0.033)
SR PRE 8.659 0.195 0.657 -0.124 4.994 0.021 0.423 -0.054
(0.000) (0.000) (0.033) (0.117) (0.000) (0.000) (0.024) (0.345)
SR POST 8.684 0.092 0.460 -0.176 4.936 0.001 0.299 -0.076
(0.000) (0.000) (0.035) (0.038) (0.000) (0.929) (0.026) (0.229)
LR POST 8.722 0.044 0.293 -0.278 4.905 -0.011 0.166 -0.123
(0.000) (0.078) (0.042) (0.005) (0.000) (0.459) (0.031) (0.101)
CHANGES RELATIVE TO SR PRESR POST - SR PRE 0.025 -0.103 -0.197 -0.052 -0.058 -0.021 -0.124 -0.022
(0.001) (0.000) (0.000) (0.405) (0.000) (0.032) (0.000) (0.650)
LR POST - SR PRE 0.064 -0.151 -0.364 -0.154 -0.089 -0.033 -0.257 -0.069
(0.000) (0.000) (0.000) (0.090) (0.000) (0.038) (0.000) (0.321)
TEST FOR PRE-EXISTING TRENDSR PRE - LR PRE 0.017 0.014 0.288 0.099 -0.039 -0.021 0.219 0.08
(0.009) (0.369) (0.000) (0.147) (0.007) (0.012) (0.000) (0.099)
Fixed effects Plant Cell-year Cell-year Cell-year Plant Cell-year Cell-year Cell-year
Number of Observations 28,518 185,909 179,043 55,358 28,518 185,909 179,043 55,358
44
Table
5Panel
A:SummaryStatisticsonExitRates,
byIndustry
oftheGoingPriva
teSample
The3-
(6-)
year
Exitrate
isthemeanof
thecorrespon
dingExitdummy,whichequalson
eiftheplantexited
(was
sold
orclosed)in
thenext
three(six)yearsandzero
otherwise.
The3-
(6-)
year
Soldrate
isthemeanof
thecorrespon
dingSolddummywhichequalson
eiftheplant
was
sold
inthenextthree(six)yearsandzero
otherwise.
The3-
(6-)
year
Closedrate
isthemeanof
thecorrespon
dingCloseddummywhich
equalson
eiftheplantwas
shutdow
nin
thenextthree(six)yearsandzero
otherwise.
(Thesevariablesareundefined
(missing)
foraplantfor
timeperiodsafteritisshutdow
n).
Thevariablesaredefined
usingtheLon
gitudinal
BusinessDatabase;
dataispresentedforsub-industries
within
manufacturingon
ly.
SIC
Industry
No.
ofNo.
of3year
3year
3year
No.
of6year
6year
6year
firm
sPlants
ExitRate
SoldRate
ClosedRate
Plants
ExitRate
SoldRate
ClosedRate
20Food&
KindredProductsMfrs
3577
539
.9%
20.4%
19.5%
683
54.2%
26.2%
28.0%
22Textile
Mill
ProductsMfrs
1733
828
.7%
12.4%
16.3%
293
50.5%
23.2%
27.3%
23Apparel
&18
289
27.3%
4.5%
22.8%
274
50.0%
7.7%
42.3%
Other
Finished
Products-Mfrs
24,25
Lumber
&WoodProds
2127
755
.2%
31.8%
23.5%
245
76.3%
41.2%
35.1%
26Pap
er&
1218
158
.6%
40.9%
17.7%
178
69.1%
45.5%
23.6%
Allied
ProductsMfrs
27PrintingPublishing&
Allied
Industries
3053
236
.1%
15.6%
20.5%
467
60.6%
26.8%
33.8%
28,29
Chem
icals&
Petroleum
Refining
2457
135
.0%
13.7%
21.4%
549
50.3%
20.0%
30.2%
30Rubber
&MiscellaneousPlasticsMfrs
3250
937
.9%
21.2%
16.7%
479
53.0%
26.7%
26.3%
31Leather
&13
355
34.4%
12.4%
22.0%
313
53.0%
21.4%
31.6%
Leather
ProductsMfrs
33PrimaryMetal
Industries
Mfrs
1721
737
.3%
16.6%
20.7%
202
50.5%
21.8%
28.7%
34FabricatedMetal
ProductsMfrs
3854
344
.0%
20.4%
23.6%
507
57.8%
26.8%
31.0%
35Industrial
&48
442
44.1%
17.9%
26.2%
413
62.0%
26.2%
35.8%
Com
mercial
MachineryMfrs
36Electronic
&47
350
46.0%
18.0%
28.0%
333
64.3%
27.3%
36.9%
Other
ElectricalEquip
Mfr
37Transportation
EquipmentMfrs
929
435
.7%
15.6%
20.1%
282
47.9%
21.3%
26.6%
38Measuring&
3320
044
.0%
19.5%
24.5%
179
62.0%
29.1%
33.0%
AnalyzingInstruments-M
frs
39Miscellaneous
1610
132
.7%
9.9%
22.8%
8757
.5%
14.9%
42.5%
ManufacturingIndsMfrs
45
Table
5Panel
B:SummaryStatisticsonExitRatesbyperiodandbyIndustry
Typ
eThevariablesaredefined
usingtheLon
gitudinal
BusinessDatabase.
See
notes
toPanelAfordefinitionof
Exit,SoldandClosedRates.‘M
PP’
refers
tofigu
resfrom
Table
3of
Maksimovic,PhillipsandPrabhala,
2011
.Industry
types
aredefined
asper
Maksimovic
andPhillips(2008).
Sam
ple
No.
ofNo.
of3year
3year
3year
No.
of6year
6year
6year
firm
sPlants
ExitRate
SoldRate
ClosedRate
Plants
ExitRate
SoldRate
ClosedRate
GoingPrivate
Firms
1258
8133
336
.5%
8.6%
27.9%
7085
457
.9%
14.2%
43.7%
Con
trol
Firms
8665
512
8340
34.2%
7.2%
27.0%
1126
1454
.1%
10.9%
43.2%
MPP-MergerFirms(Target)
1289
345
.6%
27.0%
18.6%
MPP-MergerFirms(C
ontrolsforTarget)
12.3%
9.0%
3.3%
GoingPrivate
Firms-Not
Classified
113
6375
36.5%
8.6%
27.9%
6214
54.3%
12.8%
41.5%
Con
trol
Firms-Not
Classified
6663
1005
235
.4%
7.7%
27.7%
9824
50.8%
11.2%
39.7%
GoingPrivate
Firms-Managem
ent
353
1962
130
.0%
8.9%
21.1%
1705
652
.1%
15.5%
36.6%
Con
trol
Firms-Managem
ent
2322
132
635
32.4%
7.6%
24.9%
2869
352
.3%
11.4%
41.0%
GoingPrivate
Firms-Operating
721
4524
336
.4%
7.9%
28.6%
3849
059
.3%
13.8%
45.5%
Con
trol
Firms-Operating
4777
071
563
35.1%
7.2%
28.0%
6226
455
.5%
10.8%
44.6%
GoingPrivate
Firms-Private
Equity
278
2568
241
.1%
12.8%
28.3%
2217
960
.4%
17.3%
43.1%
Con
trol
Firms-Private
Equity
2732
438
957
32.9%
7.0%
25.9%
3287
254
.0%
10.9%
43.1%
Bytimeperiod:1980s
GoingPrivate
Firms
592
4443
037
.2%
11.7%
25.5%
4443
053
.5%
16.8%
36.7%
Con
trol
Firms
4816
973
154
33.8%
7.0%
26.8%
7315
449
.5%
9.7%
39.8%
MPP-MergerFirms(Target)
6710
49.7%
30.3%
19.4%
MPP-MergerFirms(C
ontrolsforTarget)
14.5%
10.8%
3.7%
Bytimeperiod:1990s
GoingPrivate
Firms
715
3957
435
.5%
4.9%
30.6%
2909
563
.1%
9.3%
53.8%
Con
trol
Firms
4184
759
690
34.8%
7.6%
27.2%
4396
461
.2%
12.8%
48.4%
MPP-MergerFirms(Target)
6183
40.9%
23.2%
17.6%
MPP-MergerFirms(C
ontrolsforTarget)
10.2%
7.3%
2.9%
ByIndustry
Typ
e:Growth
GoingPrivate
Firms
2362
338
.4%
18.9%
19.4%
593
53.0%
28.7%
24.3%
Con
trol
Firms
558
1024
30.6%
12.5%
18.1%
966
44.8%
19.3%
25.6%
ByIndustry
Typ
e:Consolid
ating
GoingPrivate
Firms
260
3154
39.4%
19.6%
19.8%
2831
55.6%
26.5%
29.1%
Con
trol
Firms
558
5419
27.6%
12.4%
15.2%
4814
44.8%
19.6%
25.3%
ByIndustry
Typ
e:Tec
hnologicalChange
GoingPrivate
Firms
6712
5437
.6%
13.2%
24.4%
1173
56.3%
20.0%
36.2%
Con
trol
Firms
1086
2121
31.4%
10.2%
21.2%
1970
48.2%
14.3%
33.9%
ByIndustry
Typ
e:Dec
lining
GoingPrivate
Firms
6091
241
.1%
18.2%
22.9%
856
61.7%
26.2%
35.5%
Con
trol
Firms
1133
1590
34.7%
11.4%
23.3%
1497
51.0%
15.6%
35.4%
46
Table 6: Establishment exit after the “going-private” decision: Hazard and propensity analysis
This table presents exponential and competing risks model (Panel A) and linear propensity model (Panel B) results.
In Column 1 of Panel B, the 3- (6-) year Exit dummy is a variable that equals one if the plant exited (was sold
or closed) in the next three (six) years and zero otherwise. In Column 2 of Panel B, the 3- (6-) year Sold dummy
is a variable that equals one if the plant was sold in the next three (six) years and zero otherwise. In Column 2
of Panel B, the 3- (6-) year Closed dummy is a variable that equals one if the plant was shut down in the next
three (six) years and zero otherwise. (These variables are undefined (missing) for a plant for time periods after it
is shut down). For both panels, for each establishment in the going private sample, we include two establishments
that are closest in size (employment) to the going private establishment from within the same 3-digit SIC industry
and age quartile as controls. In both Panel A and Panel B, the analysis is done using time-invariant explanatory
variables, so the data has one observation for each of the gone private and control establishments. In Panel C and
D, exclude industry fixed effects in order to explore the role of industry-level variables. Capacity utilization measure
is taken from Gorodnichenko and Shapiro (2011). Industry concentration is measured as in Kovenock and Phillips
(1997) as the market share of the top four plants in the industry. Change in output demand is defined as aggregate
change in industry output. Industry classifications used in Panel D are based on definitions in Maksimovic and
Phillips (2008). The data used to construct this sample is taken from the Longitudinal Business Database (LBD).
P-values based on standard errors clustering by control group cells are in parentheses.
Panel A: Hazard and Exponential Exponential Exponential Competing RisksCompeting Risks Models model model model model
Exit Sold Closed Sold Closed
1 2 3 4a 4b
Log employment -0.110 0.216 -0.217 0.191 0.176
(0.000) (0.000) (0.000) (0.000) (0.000)
Age -0.023 0.007 -0.027 -0.001 -0.002
(0.000) (0.000) (0.000) 0.317 0.046
Gone-private dummy 0.142 0.285 0.059 0.261 0.267
(0.000) (0.000) (0.000) (0.000) (0.000)
Gone-private dummy hazard ratio 1.153 1.330 1.061 1.298 1.306
Control group Industry-size-age matched Industry-size-age matched
Fixed Effects Industry, Year Industry, Year Industry, Year Industry, Year Industry, Year
Observations 203,311 183,329 203,311 203,541 203,311
Panel B: OLS with Cell - Year 3-year Outcome dummy 6-year Outcome dummy
Fixed Effects Exit Sold Closed Exit Sold Closed
1 2 3 1 2 3
Gone-private dummy 0.031 0.015 0.015 0.051 0.043 0.008
(0.000) (0.000) (0.000) (0.000) (0.000) (0.008)
Fixed effects Industry-size-age matched cell - Year Industry-size-age matched cell - Year
Number of Observations 209,673 209,673 209,673 183,468 183,468 183,468
47
Table 6 (..ctd): Establishment exit after the “going-private” decision: Hazard and propensityanalysis
Panel C 3-year outcome dummy 6-year outcome dummy
Exit Sold Closed Exit Sold Closed
1 2 3 1 2 3
Gone-private dummy 0.108 0.050 0.058 0.093 0.047 0.046
(0.000) (0.000) (0.000) (0.000) (0.007) (0.004)
Labor Productivity -0.036 0.001 -0.037 -0.035 0.009 -0.044
(0.000) (0.895) (0.000) (0.002) (0.344) (0.000)
Firm Employment -3.24E-07 -1.77E-07 -1.47E-07 -3.17E-07 -2.26E-07 -9.06E-08
(0.002) (0.006) (0.090) (0.009) (0.006) (0.382)
Number of Plants Owned by Firm 0.00004 0.00003 0.00002 0.00003 0.00003 0.00000
(0.017) (0.051) (0.200) (0.109) (0.075) (0.885)
Capacity Utilization -0.00019 0.00013 -0.00032 -0.00085 -0.00014 -0.00071
(0.656) (0.710) (0.369) (0.069) (0.739) (0.115)
Change in Output Demand -0.262 0.072 -0.333 -0.023 0.114 -0.137
(0.030) (0.502) (0.001) (0.871) (0.366) (0.287)
Lagged Industry Concentration 0.093 0.182 -0.0891 -0.0675 0.157 -0.2246
(0.489) (0.146) (0.456) (0.631) (0.294) (0.105)
Fixed Effects Year Year Year Year Year Year
Observations 3,134 3,134 3,134 2,802 2,802 2,802
Panel D 3-year outcome dummy 6-year outcome dummy
Exit Sold Closed Exit Sold Closed
1 2 3 1 2 3
Gone-private dummy 0.121 0.068 0.053 0.110 0.071 0.039
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Labor Productivity -0.031 0.004 -0.035 -0.034 0.012 -0.046
(0.000) (0.415) (0.000) (0.000) (0.031) (0.000)
Firm Employment -4.48E-07 -2.88E-07 -1.60E-07 -4.59E-07 -3.21E-07 -1.38E-07
(0.000) (0.000) (0.008) (0.000) (0.000) (0.065)
Number of Plants Owned by Firm 0.00006 0.00004 0.00001 0.00006 0.00005 0.00001
(0.000) (0.000) (0.230) (0.000) (0.000) (0.370)
Consolidating Industry 0.006 0.014 -0.007 0.019 0.004 0.015
(0.691) (0.309) (0.567) (0.276) (0.824) (0.358)
Technological Change Industry -0.007 0.005 -0.011 0.006 -0.004 0.011
(0.731) (0.773) (0.492) (0.768) (0.820) (0.594)
Declining Industry 0.024 -0.001 0.025 0.059 -0.010 0.070
(0.256) (0.970) (0.177) (0.011) (0.628) (0.002)
Fixed Effects Year Year Year Year Year Year
Observations 9,380 9,380 9,380 8,363 8,363 8,363
48
Table
7:Exithazard
after
the“going-priva
te”dec
ision:Tests
fordifferen
tialtargeting
Thistable
presents
expon
ential
model
resultsof
theduration
toexit,sale
andclosure
measuresof
thesample
ofestablishments
offirm
sthat
wentprivate.
Incolumns1,
2and3,
foreach
establishmentin
thego
ingprivatesample,weincludeupto
eigh
testablishments
(based
ondata
availability)that
areclosestin
size
(employm
ent)
tothego
ingprivateestablishmentfrom
within
thesame3-digitSIC
industry,andbelon
ging
tothesameagequartile
ascontrols.
Thedatausedto
construct
thissample
istakenfrom
theCensusof
Manufactures(C
MF),
theAnnual
Survey
ofManufactures(A
SM)andthelongitudinal
businessdatabase(LBD).
P-values
based
onstandarderrors
clustered
bycontrol
grou
p
cells
arein
parentheses.
Panel
A:Hazard
model
withIndustry,
Exit
Sold
Closed
Exit
Sold
Closed
Exit
Sold
Closed
Yea
rFixed
Effec
ts1
23
12
31
23
ProductivityMeasure
=Lab
orProductivity
OLSFETFP
Blundell-Bon
dTFP
Gon
e-Private
Dummy
0.97
81.10
90.57
40.57
51.01
3-0.229
0.87
10.950
0.394
(0.000
)(0.000
)(0.041
)(0.005
)(0.000
)(0.412
)(0.000
)(0.000)
(0.127)
ProductivityMeasure
-0.024
0.13
5-0.163
-0.048
0.16
3-0.224
0.00
80.165
-0.110
(0.408
)(0.001
)(0.000
)(0.307
)(0.018
)(0.000
)(0.880
)(0.028)
(0.116)
Gon
e-privatedummyXProductivityMeasure
-0.115
-0.141
-0.065
-0.054
-0.172
0.12
4-0.198
-0.224
-0.077
(0.003)
(0.008)
(0.237
)(0.343
)(0.029)
(0.107
)(0.005)
(0.020)
(0.446)
Observations
9,52
48,61
69,52
49,52
48,61
69,52
49,52
48,616
9,524
Con
trols
Employm
ent,age
Fixed
effects
Industry,Year
Panel
B:OLSwithCell-Yea
rExit
Sold
Closed
Exit
Sold
Closed
Exit
Sold
Closed
Fixed
Effec
ts(3
year
outcomes)
12
31
23
12
3
ProductivityMeasure
=Lab
orProductivity
OLSFETFP
Blundell-Bon
dTFP
Gon
e-Private
Dummy
0.34
00.14
80.19
10.28
70.16
00.12
70.38
10.273
0.108
(0.001
)(0.048
)(0.009
)(0.007
)(0.057
)(0.104
)(0.000
)(0.000)
(0.097)
ProductivityMeasure
-0.024
-0.002
-0.022
-0.030
-0.011
-0.019
-0.015
0.004
-0.018
(0.134
)(0.856
)(0.067
)(0.230
)(0.563
)(0.317
)(0.579
)(0.849)
(0.368)
Gon
e-privatedummyXProductivityMeasure
-0.040
-0.013
-0.027
-0.044
-0.022
-0.022
-0.102
-0.077
-0.025
(0.035)
(0.353
)(0.054)
(0.129
)(0.339
)(0.295
)(0.003)
(0.003)
(0.317)
Observations
9,64
69,64
69,64
69,64
69,64
69,64
69,64
69,646
9,646
Fixed
effects
Industry-size-agematched
cell-year
Fixed
Effects
49
Table
8Panel
A:Changes
inproductivity:
Differen
ce-in-D
ifferen
cessp
ecifica
tionsbreakdownbyacq
uirer
type
This
table
presents
testsof
differencesin
productivitybased
onseparateregression
sforeach
acquirer
typeidentified
inTable
1,Panel
B.
For
each
establishmentin
thego
ingprivatesample,weincludeupto
eigh
testablishments
(based
ondataavailability)that
areclosestin
size
(employm
ent)
tothego
ingprivateestablishmentfrom
within
thesame3-digit
SIC
industry,andbelon
gingto
thesameagequartile
ascontrols.
Column1includes
allestablishments;in
Column2,
thesample
forboththego
ingprivateandmatched
control
establishments
excludes
allsold
plants
(i.e.,
thosethat
underwentachange
inow
nership
afterthego
ingprivatedate).Thedatausedto
construct
this
sample
istakenfrom
theCensusof
Manufactures(C
MF)andtheAnnual
Survey
ofManufactures(A
SM).
DummyvariablesLR
PREequals
1foryears-6
to-4
from
delistingandzero
otherwise,
SR
PREisadummyvariable
that
equals1foryears-3
to-1
from
delistingandzero
otherwise,
SR
POST
equals1foryears0to
3from
delistingandzero
otherwise,
andLR
POST
equals1foryears4to
6from
delistingand
zero
otherwise.
Standarderrors
clustered
bycontrol
grou
pcells
areusedto
assess
theP-values
(inparentheses)forthedifferenttests.
Unclassified
Managem
ent
Operating
Priva
teEquity
12
12
12
12
Laborproductivity
Shortrunchange
(SR
POST
-SR
PRE)
-0.003
0.01
40.00
20.01
60.01
70.01
2-0.002
-0.004
(0.968
)(0.877
)(0.931
)(0.650
)(0.501
)(0.722
)(0.927)
(0.887)
Lon
grunchange
(LR
POST
-SR
PRE)
-0.049
-0.111
-0.019
-0.012
0.02
40.04
60.003
0.021
(0.582
)(0.366
)(0.651
)(0.822
)(0.473
)(0.335
)(0.932)
(0.706)
Pre-existingtrend(SR
PRE-LR
PRE)
0.08
00.09
0-0.005
-0.002
0.01
10.01
00.008
0.020
(0.143
)(0.221
)(0.856
)(0.955
)(0.628
)(0.736
)(0.785)
(0.581)
OLSTFP
Shortrunchange
(SR
POST
-SR
PRE)
-0.061
-0.048
-0.022
-0.018
0.00
20.00
0-0.009
-0.006
(0.091)
(0.377
)(0.092)
(0.316
)(0.859
)(0.998
)(0.468)
(0.718)
Lon
grunchange
(LR
POST
-SR
PRE)
-0.095
-0.132
-0.019
-0.020
0.01
70.03
8-0.018
-0.022
(0.055)
(0.060)
(0.356
)(0.493
)(0.319
)(0.125
)(0.369)
(0.441)
Pre-existingtrend(SR
PRE-LR
PRE)
0.061
0.068
-0.008
-0.010
0.01
00.00
90.001
0.010
(0.053)
(0.106)
(0.578
)(0.592
)(0.357
)(0.557
)(0.962)
(0.616)
BlundellBondTFP
Shortrunchange
(SR
POST
-SR
PRE)
-0.049
-0.050
-0.005
-0.001
-0.003
-0.009
-0.003
0.001
(0.174
)(0.347
)(0.655
)(0.962
)(0.839
)(0.585
)(0.833)
(0.948)
Lon
grunchange
(LR
POST
-SR
PRE)
-0.080
-0.104
0.00
50.01
10.02
20.043
0.000
0.005
(0.101
)(0.133
)(0.810
)(0.691
)(0.176
)(0.070)
(0.985)
(0.867)
Pre-existingtrend(SR
PRE-LR
PRE)
0.061
0.04
5-0.008
-0.017
0.01
00.00
20.001
0.009
(0.053)
(0.209
)(0.578
)(0.330
)(0.357
)(0.921
)(0.962)
(0.595)
Sam
ple
All
Excludes
sold
All
Excludes
sold
All
Excludes
sold
All
Excludes
sold
Number
ofob
servations
18,683
13,947
56,753
42,209
84,020
62,917
62,795
47,249
Fixed
effects
Industry-size-age-year
50
Table
8Panel
B:Changes
inca
pitalandem
ploym
entaroundthe“going-priva
te”dec
ision:Differen
ce-in-D
ifferen
ces
specifica
tionsbreakdownbyacq
uirer
type
This
table
presents
resultsforthetestsof
changesin
capital
andem
ploym
entover
time,
based
onseparateregression
sforeach
acquirer
typeidentified
inTable
1,Panel
B.For
each
establishmentin
thego
ingprivatesample,weincludeupto
eigh
testablishments
(based
ondata
availability)that
areclosestin
size
(employm
ent)
tothego
ingprivateestablishmentfrom
within
thesame3-digitSIC
industry,andbelon
ging
tothesameagequartile
ascontrols.
Thedatausedto
construct
this
sample
istakenfrom
theCensusof
Manufactures(C
MF)andthe
Annual
Survey
ofManufactures(A
SM).DummyvariablesLR
PREequals1foryears-6
to-4
from
goingprivateandzero
otherwise,
SR
PRE
equals1foryears-3
to-1
from
goingprivateandzero
otherwise,
SR
POST
equals1foryears0to
3from
goingprivateandzero
otherwise,
andLR
POST
equals1foryears4to
6from
goingprivateandzero
otherwise.
Standarderrors
clustered
byindustry-size-agegrou
psareused
toassess
theP-values
(inparentheses)forthedifferenttests.
Unclassified
Managem
ent
Operating
Priva
teEquity
12
12
12
12
Logdefl
atedCapital
Shortrunchange
(SR
POST
-SR
PRE)
-0.158
-0.200
-0.122
-0.122
-0.079
-0.068
-0.110
-0.124
(0.010)
(0.015)
(0.000)
(0.001)
(0.001)
(0.057)
(0.000)
(0.000)
Lon
grunchange
(LR
POST
-SR
PRE)
-0.126
-0.220
-0.217
-0.243
-0.116
-0.083
-0.210
-0.262
(0.136)
(0.068)
(0.000)
(0.000)
(0.001)
(0.102)
(0.000)
(0.000)
Pre-existingtrend(SR
PRE-LR
PRE)
0.08
30.07
60.03
50.03
1-0.004
-0.017
0.012
0.023
(0.157
)(0.342
)(0.214
)(0.441
)(0.860
)(0.601
)(0.666)
(0.555)
Logem
ploym
ent
Shortrunchange
(SR
POST
-SR
PRE)
0.01
90.01
9-0.047
-0.050
-0.018
-0.025
-0.019
-0.019
(0.600
)(0.694
)(0.004)
(0.020)
(0.227
)(0.222
)(0.210)
(0.342)
Lon
grunchange
(LR
POST
-SR
PRE)
-0.031
-0.084
-0.071
-0.064
-0.021
-0.032
-0.061
-0.104
(0.563
)(0.208
)(0.011)
(0.065)
(0.379
)(0.318
)(0.017)
(0.001)
Pre-existingtrend(SR
PRE-LR
PRE)
-0.063
-0.080
-0.022
-0.016
-0.013
-0.019
-0.021
-0.029
(0.041)
(0.041)
(0.128
)(0.441
)(0.280
)(0.260
)(0.136)
(0.130)
Sam
ple
All
Excludes
sold
All
Excludes
sold
All
Excludes
sold
All
Excludes
sold
Number
ofob
servations
18,683
13,947
56,753
42,209
84,020
62,917
62,795
47,249
Fixed
effects
Industry-size-age-year
51
Table 8 Panel C: Establishment exit propensity after the “going-private” decision: breakdownby acquirer typeThis table presents hazard models in Panel C-A, and exit dummy linear propensity models in Panel C-B, for each
acquirer type identified in Table 1, Panel B. In Column 1 of Panel B, the 3- (6-) year Exit dummy is a variable that
equals one if the plant exited (was sold or closed) in the next three (six) years and zero otherwise. In Column 2
of Panel B, the 3- (6-) year Sold dummy is a variable that equals one if the plant was sold in the next three (six)
years and zero otherwise. In Column 3 of Panel B, the 3- (6-) year Closed dummy is a variable that equals one if
the plant was shut down in the next three (six) years and zero otherwise. (These variables are undefined (missing)
for a plant for time periods after it is shut down). For both panels, for each establishment in the going private
sample, we include two establishments that are closest in size (employment) to the going private establishment
from within the same 3-digit SIC industry and age quartile as controls. In both Panel A and Panel B, the analysis
is done using time-invariant explanatory variables, so the data has one observation for each of the gone private and
control establishments. The data used to construct this sample is taken from the Longitudinal Business Database
(LBD). P-values based on standard errors clustering by control group cells are in parentheses.
Panel C - A: Exponential Exit Sold Closed
Hazard model 1 2 3
Management acquirers -0.223 0.094 -0.258
(0.000) (0.000) (0.000)
Management acquirers: hazard ratio 0.800 1.099 0.773
Operating acquirers 0.185 0.227 0.122
(0.000) (0.000) (0.000)
Operating acquirers: hazard ratio 1.203 1.255 1.130
Private equity acquirers 0.285 0.501 0.092
(0.000) (0.000) (0.000)
Private Equity acquirers: hazard ratio 1.330 1.650 1.096
Control group Industry-size-age matched
Other controls Employment, Age Employment, Age Employment, Age
Fixed Effects Industry, Year Industry, Year Industry, Year
Observations 203,311 183,329 203,311
Panel C - B: OLS using 3-year outcome dummy 6-year outcome dummy
Cell-Year fixed effects Exit Sold Closed Exit Sold Closed
1 2 3 1 2 3
Management acquirers -0.009 0.024 -0.033 0.030 0.064 -0.033
(0.072) (0.000) (0.000) (0.000) (0.000) (0.000)
Fixed effects Industry-size-age matched cell - Year Industry-size-age matched cell - Year
Number of Observations 52,256 52,256 52,256 45,749 45,749 45,749
Operating acquirers 0.019 0.003 0.016 0.054 0.037 0.017
(0.000) (0.134) (0.000) (0.000) (0.000) (0.000)
Fixed effects Industry-size-age matched cell - Year Industry-size-age matched cell - Year
Number of Observations 116,806 116,806 116,806 100,754 100,754 100,754
Private equity acquirers 0.081 0.053 0.027 0.066 0.072 -0.006
(0.000) (0.000) (0.000) (0.000) (0.000) (0.230)
Fixed effects Industry-size-age matched cell - Year Industry-size-age matched cell - Year
Number of Observations 64,639 64,639 64,639 55,051 55,051 55,051
52
Table
8Panel
D:Exitafter
“going-priva
te”andproductivity:
Tests
fordifferen
tialtargetingbyacq
uirer
type
Thistable
presents
OLSregression
sof
exitdummyvariables.
InColumn1theExitdummyisavariable
that
equalson
eiftheplantexited
(was
sold
orclosed)in
thenextthreeyearsandzero
otherwise.
InColumn2,
Solddummyisavariable
that
equalson
eiftheplantwas
sold
inthenextthree(six)yearsandzero
otherwise.
InColumn3theCloseddummyis
avariable
that
equalson
eiftheplantwas
shutdow
n
inthenextthreeyearsandzero
otherwise.
(Thesevariablesareundefined
(missing)
foraplantfortimeperiodsafteritisshutdow
n).
For
bothpanels,
foreach
establishmentin
thego
ingprivatesample,weincludetwoestablishments
that
areclosestin
size
(employm
ent)
tothe
goingprivateestablishmentfrom
within
thesame3-digitSIC
industry
andagequartile
ascontrols.
Theanalysisisdon
eusingtime-invariant
explanatoryvariables,
sothedatahas
oneob
servationforeach
ofthego
neprivateandcontrol
establishments.Thedatausedto
construct
this
sample
istakenfrom
theCensusof
Manufactures(C
MF),
theAnnual
Survey
ofManufactures(A
SM)andtheLon
gitudinal
Business
Database(LBD).P-values
based
onstandarderrors
clustered
byindustry-size-agegrou
psarein
parentheses.
Exit
Sold
Closed
Exit
Sold
Closed
Exit
Sold
Closed
12
31
23
12
3
ProductivityMeasure
=Lab
orProductivity
OLSFETFP
Blundell-Bon
dTFP
Operatingacquirers
0.53
30.29
50.23
80.38
40.31
0.07
40.285
0.284
0.001
(0.000
)(0.005
)(0.016
)(0.016
)(0.017
)(0.527
)(0.019
)(0.004)
(0.992)
Managem
entacquirers
-0.082
-0.172
0.09
0.05
2-0.044
0.09
60.166
0.05
0.115
(0.698
)(0.264
)(0.581
)(0.802
)(0.765
)(0.563
)(0.362
)(0.696)
(0.434)
Private
equityacquirers
0.46
20.32
30.14
0.50
30.32
10.18
20.49
0.355
0.135
(0.020
)(0.023
)(0.363
)(0.013
)(0.032
)(0.258
)(0.003
)(0.004)
(0.284)
ProductivityMeasure
-0.023
-0.001
-0.022
-0.026
-0.007
-0.019
-0.026
-0.002
-0.024
(0.125
)(0.928
)(0.067
)(0.279
)(0.697
)(0.291
)(0.317
)(0.920)
(0.207)
OperatingXProductivityMeasure
-0.072
-0.039
-0.033
-0.064
-0.061
-0.004
-0.055
-0.078
0.024
(0.003)
(0.040)
(0.067)
(0.137
)(0.081)
(0.897
)(0.242
)(0.040)
(0.517)
Managem
entX
ProductivityMeasure
0.02
10.03
9-0.018
-0.007
0.01
9-0.026
-0.055
-0.009
-0.046
(0.609
)(0.194
)(0.574
)(0.901
)(0.635
)(0.555
)(0.439
)(0.854)
(0.420)
Private
EquityXProductivityMeasure
-0.059
-0.041
-0.018
-0.095
-0.056
-0.038
-0.134
-0.096
-0.038
(0.111
)(0.129
)(0.520
)(0.079)
(0.162
)(0.366
)(0.039)
(0.041)
(0.429)
Observations
9,64
69,64
69,64
69,64
69,64
69,64
69,646
9,646
9,646
Fixed
Effects
Industry-age-sizematched
cell-year
53
Figure 1: Before-after productivity trends in event timeThis figure displays the evolution of productivity measures in event time for the going private sample. Theconfidence intervals are based on standard errors clustered by plant.
3.1
3.2
3.3
3.4
3.5
−6 −4 −2 0 2 4 6
Labor Productivity: Before−After changes3.
553.
63.
653.
73.
75
−6 −4 −2 0 2 4 6
OLS TFP: Before−After changes
2.35
2.4
2.45
2.5
−6 −4 −2 0 2 4 6
Blundell−Bond TFP: Before−After changes
(5% confidence intervals)
Years from going−private
Productivity Trends
54
Figure 2: Difference-in-differences productivity trends in event time
This figure displays the evolution of productivity measures in event time for the going private sample,relative to a industry-age-initial size matched control group.The confidence intervals are based on standarderrors clustered by control group cells.
0.0
5.1
.15
.2
−6 −4 −2 0 2 4 6
Labor Productivity: DID changes
0.0
2.0
4.0
6.0
8
−6 −4 −2 0 2 4 6
OLS TFP: DID changes
−.0
4−
.02
0.0
2.0
4.0
6
−6 −4 −2 0 2 4 6
Blundell−Bond TFP: DID changes
(5% confidence intervals)
Years from going−private
Productivity trends
55
Figure 3: Before-after and difference-in-differences (DID) trends for capital and employmentin event timeThis figure displays the evolution of capital and employment measures in event time for the going privatesample. The before-after figures use only the going private sample, and the DID figures show trendsrelative to an industry-age-initial size matched control group. The confidence intervals are based onstandard errors clustered by plant in the Before-After figures, and by control group cells in the DID figures.
8.6
8.65
8.7
8.75
8.8
−6 −4 −2 0 2 4 6
Log deflated capital: Before−After changes
4.9
4.95
55.
05
−6 −4 −2 0 2 4 6
Log employment: Before−After changes
−.1
0.1
.2.3
−6 −4 −2 0 2 4 6
Log deflated capital: DID changes
−.0
50
.05
.1
−6 −4 −2 0 2 4 6
Log employment: DID changes
(5% confidence intervals)
Years from going−private
Capital and employment trends
56
SUPPLEMENTARY APPENDIX
TABLES
57
Table A.1: Changes in capital and employment after “going-private” : Before-After andDifference-in-Differences specifications, excluding sold plants
This table presents before-after and difference-in-differences regression results for capital and employment. The
data used to construct this sample excludes all sold plants (i.e., those that underwent a change in ownership
after the going private date) in both the treatment and control samples. In DID1, for each establishment in the
going private sample, we include upto eight establishments (based on data availability) that are closest in size
(employment) to the going private establishment from within the same 3-digit SIC industry, and belonging to the
same age quartile as controls. In DID2, for each establishment in the going private sample, we include upto eight
establishments (based on data availability) that are closest in past TFP to the going private establishment from
within the same 3-digit SIC industry, and belonging to the same age quartile as controls. In DID3, we include upto
eight establishments within the same 3-digit SIC industry closest in propensity to go private (but whose owner
firms did not go private), as control establishments. The data used to construct this sample is taken from the
Census of Manufactures (CMF) and the Annual Survey of Manufactures (ASM). Dummy variables LR PRE equals
1 for years -6 to -4 from going private and zero otherwise, SR PRE equals 1 for years -3 to -1 from going private
and zero otherwise, SR POST equals 1 for years 0 to 3 from going private and zero otherwise, and LR POST
equals 1 for years 4 to 6 from going private and zero otherwise. P-values based on standard errors clustered
by plant establishments (before-after), industry-size-age cells (DID1), industry-past TFP-age cells (DID2) and
industry-propensity cells (DID3) are in parentheses.
Log deflated capital Log employment
Before-After DID1 DID2 DID3 Before-After DID1 DID2 DID3
1a 1b 1c 1d 2a 2b 2c 2d
LR PRE 8.599 0.212 0.402 -0.232 4.984 0.051 0.213 -0.176
(0.000) (0.000) (0.000) (0.117) (0.000) (0.000) (0.000) (0.078)
SR PRE 8.6143 0.223 0.721 -0.106 4.9374 0.025 0.451 -0.071
(0.000) (0.000) (0.000) (0.408) (0.000) (0.001) (0.000) (0.441)
SR POST 8.645 0.1150 0.4900 -0.1510 4.881 0.0003 0.3020 -0.0530
(0.000) (0.000) (0.000) (0.301) (0.000) (0.981) (0.000) (0.614)
LR POST 8.6899 0.057 0.303 -0.249 4.853 -0.029 0.144 -0.089
(0.000) (0.112) (0.000) (0.160) (0.000) (0.136) (0.001) (0.489)
CHANGES RELATIVE TO SR PRE
SR POST - SR PRE 0.0307 -0.1080 -0.2310 -0.0450 -0.0564 -0.0245 -0.1490 0.0178
(0.001) (0.000) (0.000) (0.681) (0.000) (0.056) (0.000) (0.831)
LR POST - SR PRE 0.0756 -0.1656 -0.4180 -0.1430 -0.0844 -0.053 -0.3070 -0.0184
(0.000) (0.000) (0.000) (0.386) (0.000) (0.008) (0.000) (0.881)
TEST FOR PRE-EXISTING TREND
SR PRE - LR PRE 0.0153 0.011 0.319 0.126 -0.0466 -0.026 0.238 0.105
(0.157) (0.611) 0.000 (0.269) (0.000) (0.022) (0.000) (0.180)
Fixed effects Plant Cell-year Cell-year Cell-year Plant Cell-year Cell-year Cell-year
Observations 20,788 139,826 136,218 36,659 20,788 139,826 136,218 36,659
58
Table
A.2:Changes
inproductivitymea
sures:
Differen
ce-in-D
ifferen
cesbyacq
uirer
type,
matchingonlagged
TFP
Thistable
presents
testsof
differencesin
productivitybased
onseparateregression
sforeach
acquirer
typeidentified
inTable
1,Panel
B.For
each
establishmentin
thego
ingprivatesample,weincludeupto
eigh
testablishments
(based
ondataavailability)that
areclosestin
lagged
(Blundell-Bon
d)TFP
tothego
ingprivateestablishmentfrom
within
thesame3-digitSIC
industry,andbelon
gingto
thesameagequartile
ascontrols.
Column1includes
allestablishments;in
Column2,
thesample
forboththego
ingprivateandmatched
control
establishments
excludes
allsold
plants
(i.e.,
thosethat
underwentachange
inow
nership
afterthego
ingprivatedate).Thedatausedto
construct
this
sample
istakenfrom
theCensusof
Manufactures(C
MF)andtheAnnual
Survey
ofManufactures(A
SM).
DummyvariablesLR
PREequals
1foryears-6
to-4
from
delistingandzero
otherwise,
SR
PREisadummyvariable
that
equals1foryears-3
to-1
from
delistingandzero
otherwise,
SR
POST
equals1foryears0to
3from
delistingandzero
otherwise,
andLR
POST
equals1foryears4to
6from
delistingand
zero
otherwise.
Standarderrors
clustered
bycontrol
grou
pcells
areusedto
assess
theP-values
(inparentheses)forthedifferenttests.
Unclassified
Managem
ent
Operating
Priva
teEquity
12
12
12
12
Laborproductivity
Shortrunchange
(SR
POST
-SR
PRE)
0.06
80.06
40.02
20.01
80.01
1-0.002
0.009
0.004
(0.198
)(0.353
)(0.372
)(0.581
)(0.637
)(0.951
)(0.700)
(0.894)
Lon
grunchange
(LR
POST
-SR
PRE)
-0.044
-0.018
-0.017
-0.042
0.03
70.04
00.007
0.007
(0.597
)(0.875
)(0.652
)(0.408
)(0.234
)(0.327
)(0.841)
(0.895)
Pre-existingtrend(SR
PRE-LR
PRE)
0.04
20.05
00.047
0.075
0.01
60.03
20.033
0.068
(0.360
)(0.424
)(0.066)
(0.035)
(0.450
)(0.237
)(0.202)
(0.045)
OLSTFP
Shortrunchange
(SR
POST
-SR
PRE)
-0.037
-0.039
-0.017
-0.024
-0.015
-0.020
-0.010
-0.015
(0.206
)(0.341
)(0.180
)(0.173
)(0.184
)(0.200
)(0.378)
(0.380)
Lon
grunchange
(LR
POST
-SR
PRE)
-0.146
-0.167
-0.042
-0.037
0.00
40.00
7-0.030
-0.041
(0.001)
(0.006)
(0.034)
(0.208
)(0.812
)(0.748
)(0.100)
(0.127)
Pre-existingtrend(SR
PRE-LR
PRE)
0.055
0.065
0.027
0.02
70.033
0.044
0.034
0.044
(0.040)
(0.079)
(0.023)
(0.098
)(0.001)
(0.000)
(0.004)
(0.006)
BlundellBondTFP
Shortrunchange
(SR
POST
-SR
PRE)
-0.019
-0.007
0.00
50.00
2-0.007
-0.012
-0.001
-0.005
(0.476
)(0.832
)(0.661
)(0.894
)(0.504
)(0.410
)(0.902)
(0.760)
Lon
grunchange
(LR
POST
-SR
PRE)
-0.132
-0.137
-0.005
-0.014
0.01
20.01
8-0.004
-0.004
(0.011)
(0.067)
(0.778
)(0.655
)(0.429
)(0.406
)(0.831)
(0.864)
Pre-existingtrend(SR
PRE-LR
PRE)
0.055
0.02
80.02
70.00
30.033
0.01
00.034
0.011
(0.040)
(0.338
)(0.023
)(0.804
)(0.001)
(0.380
)(0.004)
(0.406)
Fixed
effects
Industry-PastTFP-age-year
Sam
ple
All
Excludes
sold
All
Excludes
sold
All
Excludes
sold
All
Excludes
sold
Number
ofob
servations
19,926
15,074
52,863
40,234
79,773
60,773
60,194
45,769
59
Table A.3: Changes in establishment profit measures around the “going-private” decision :Before-After and Difference-in-Differences SpecificationsThis table presents regression results for two profit-related measures. The first is gross profits, defined as sales
less sum of (materials cost, energy costs, blue collar wage bill and white collar wage bill). The second is the ratio
of gross profits to sales. Both measures are winsorized by 2% on both tails to minimize effects of outliers. In
DID1, for each establishment in the going private sample, we include upto eight establishments (based on data
availability) that are closest in size (employment) to the going private establishment from within the same 3-digit
SIC industry, and belonging to the same age quartile as controls. In DID2, we include upto eight establishments
within the same 3-digit SIC industry closest in propensity to go private (but whose owner firms did not go private),
as control establishments. The data used to construct this sample is taken from the Census of Manufactures (CMF)
and the Annual Survey of Manufactures (ASM). Dummy variables LR PRE equals 1 for years -6 to -4 from going
private and zero otherwise, SR PRE equals 1 for years -3 to -1 from going private and zero otherwise, SR POST
equals 1 for years 0 to 3 from going private and zero otherwise, and LR POST equals 1 for years 4 to 6 from going
private and zero otherwise. P-values based on standard errors clustered by plant establishments (before-after),
industry-size-age cells (DID1) and industry-propensity cells (DID2) are in parentheses.
Gross Profit Return on sales
Before-After DID1 DID2 Before-After DID1 DID2
1a 1b 1c 1a 1b 1c
LR PRE 9490.0 34.6 -1160 24.630 0.744 0.732
(0.000) (0.909) (0.277) (0.000) (0.056) (0.446)
SR PRE 11274.0 195.5 -873.8 26.001 0.874 1.165
(0.000) (0.510) (0.358) (0.000) (0.021) (0.165)
SR POST 12869.0 301.0 -1332 27.195 1.223 0.547
(0.000) (0.407) (0.285) (0.000) (0.002) (0.581)
LR POST 14270.0 523.7 -627.4 25.999 0.898 1.620
(0.000) (0.324) (0.706) (0.000) (0.090) (0.223)
CHANGES RELATIVE TO SR PRE
SR POST - SR PRE 1595.0 105.5 -458.2 1.194 0.349 -0.618
(0.000) (0.712) (0.651) (0.000) (0.377) (0.538)
LR POST - SR PRE 2996.0 328.2 246.4 -0.002 0.024 0.455
(0.000) (0.490) (0.875) (0.996) (0.967) (0.749)
TEST FOR PRE-EXISTING TREND
SR PRE - LR PRE 1784.0 160.9 286.2 1.371 0.130 0.433
(0.000) (0.537) (0.743) (0.000) (0.734) (0.664)
Fixed effects Plant Cell-year Cell-year Plant Cell-year Cell-year
Number of observations 28,518 185,909 55,358 28,518 185,909 55,358
60