Declining tax payments among profitable U.S. firms: changes in firm characteristics, tax rates, and propensity to pay taxes Yuzhu Lu Lingnan University [email protected]Liang Shao Hong Kong Baptist University [email protected]Yue Zhang Lingnan University [email protected]Current Version: August 2016
54
Embed
Declining tax payments among profitable U.S. firms ... · The relative magnitude of income taxes paid by U.S. has been declining firms since World War II. The total corporate income
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Declining tax payments among profitable U.S. firms: changes in firm characteristics, tax rates, and propensity to pay taxes
Declining tax payments among profitable U.S. firms: changes in firm characteristics, tax rates, and propensity to pay taxes
Abstract Using 153,949 profitable U.S. public firm-years from 1956-2013, we observe that the tax expense to asset ratio (TAX_AT) decreases by around 7 basis points each year. We decompose this trend and find that 36.09, 52.44, and 11.47% result from changes in firm characteristics, the statutory tax rate, and firms’ propensity to pay taxes, respectively. In particular, we find that before 1988, the declining trend in tax payments is mainly driven by changes in firm characteristics and the statutory tax rate. In contrast, after 1988 the trend is mainly driven by a decreasing propensity to pay taxes that, while prevalent across different groups of firms, is more pronounced among firms with multinational operations, high earnings, strong governance, and high demand for earnings management and cash. Our findings facilitate a more comprehensive understanding of the recent trend in firms’ tax avoidance while generating insights into how to improve tax enforcement. JEL Classification: H26, M41 Key words: corporate tax avoidance, downward trend, firm characteristics, statutory tax rate, propensity to pay tax
2
Declining tax payments among profitable U.S. firms: changes in firm characteristics, tax rates, and propensity to pay taxes
1. Introduction
The relative magnitude of income taxes paid by U.S. firms has been declining since
World War II. The total corporate income taxes in 2013 accounted for 1.6 and 9.6% of the
U.S. GDP and fiscal revenues, respectively—a sharp drop from 5.9 and 31.8% in 1952.1
Recently, the blame has fallen on an increase in tax-sheltering activities attributed to large
and multinational firms that manipulate taxable income numbers and defer tax payments by
delaying profit remittances from overseas. 2 Appeals have also arisen for stricter tax
enforcement or tax-rate reductions.3 The prerequisite for any effective tax reform is a sound
understanding of why firms are paying less. To this end, we examine the factors that drive
this declining trend in a sample of U.S. public firms. We include only profitable firm-years
and exclude firms operating in financial/utility industries. For this exercise, we collect
153,949 firm-years spanning 1956-2013.
We first confirm that the declining trend in corporate tax payments exists for our sample
firms. The effective tax rate (ETR, measured by income taxes scaled by pretax income)
dropped from 47% in 1956 to 28% in 2013, falling below the federal statutory tax rate every
1 Data source: https://www.whitehouse.gov/omb/budget/Historicals. 2 For example, the tax receipts on corporate income of U.S. firms dropped from 50% in 1951 to about 17% in 2011 (Federal Reserve Bank of St. Louis: https://research.stlouisfed.org/fred2/graph/?g=aWA), and the deferred corporate foreign earnings increased from 286 billion in 2001 to 1,266 billion in 2010 (Credit Suisse, Parking Earnings Overseas $1.3 Trillion Parking Lot for the S&P 500? Equity Research, April 2011). 3 See, for example, Tax Reform in the 114th Congress: An Overview of Proposals https://www.fas.org/sgp/crs/misc/R43060.pdf.
3
year, and the income taxes to total asset ratio decreases from 8% in 1956 to less than 3% in
2013.
We next investigate whether the declining trend in tax payments results from changes
in tax-saving related firm characteristics, reductions in the statutory tax rate, or a decreasing
propensity to pay taxes. Although alternative measures generate similar results, we employ
TAX_AT , income taxes scaled by total assets, to quantify the extent to which a firm pays
income taxes. It is appropriate to standardize the firm characteristics in our regression, such
as R&D expense and debt, with total assets to ease heterogeneity in scale. At the meantime,
as shown in our regression specification in Section 2, the sensitivity of TAX_AT to return on
assets (ROA) is equivalent to the effective tax rate (ETR), a popular measure of tax payment
in literature.
We choose 1956-1960 as the base period and estimate a regression model of TAX_AT on
return on assets (ROA) and other firm characteristics that the literature has deemed relevant to
tax saving. With the coefficients (including the intercept) from the base-period regression and
the contemporary statutory tax rate, we compute the ‘expected’ tax payment (Expected_Tax)
for each subsequent three-year time window with contemporary firm characteristics.
Expected_Tax thus captures the effects of changes in the statutory tax rate and firm
characteristics, and the difference between the actual tax payment (TAX_AT) and
Expected_Tax is caused by overall changes in the model’s coefficients, which we refer to as
4
firms’ propensity to pay taxes. A decreasing propensity to pay taxes would suggest that firms
have become less likely to pay taxes regardless of their characteristics and changes in
statutory tax rate. We further decompose the propensity to pay taxes into ‘identified’ and
‘unidentified’ propensity. The identified propensity is measured by changes in the coefficients
of the firm characteristics included in the regression except ROA. The unidentified propensity
is measured by changes in the intercept and the coefficient of ROA in short of the statutory
tax rate4. We call it “unidentified” because the intercept and the coefficient of ROA are not
correlated with the controlled firm characteristics. The unidentified propensity captures either
the opportunities available for firms to reduce their tax burden which are not explained by the
traditional tax avoidance model (Equation 6) or firms’ increasing motivations for tax saving.
With this approach, we observe a declining trend in tax payment among U.S. profitable
corporations in the last half century. Over the full sample period (1961-2013), TAX_AT
decreases by around 7 basis points each year, of which 36.09, 52.44, and 11.47% result from
changes in firm characteristics, reductions in the statutory tax rate, and a lower propensity to
pay taxes, respectively. Among the firm characteristics, ROA is the most significant factor.
The downward trend in ROA explains 47.29% of the annual trend of TAX_AT. We also find
that the identified propensity to pay taxes (i.e., changes in the coefficients of the firm
4 Please refer to section 2 – methodology for more detailed explanation on regression specification.5 Including a constant on the right side of Equation (2), assuming that each firm can save a fixed amount of tax expense in year t, does not change our results, and the corresponding variable of this constant in Equation (3) does not have a significant coefficient.
5
characteristics, except ROA) contributes little to the declining trend of TAX_AT, and most of
the change in the propensity to pay taxes results from the coefficient of ROA drifting farther
beneath the statutory tax rate over time.
Interestingly, we also find that there is a structural change in why corporations pay less
taxes and the turning point coincides with the last overhaul of the U.S. statutory tax rate in
1987. Over 1961-1987, the tax payment trend is mainly explained by changes in firm
characteristics, whereas over 1988-2013 the trend is mainly explained by a lower propensity
to pay taxes. In 1961-1987, changes in firm characteristics and the statutory tax rate explain
66% and 35.51% of the decreasing TAX_AT trend, respectively, whereas in 1988-2013, the
unidentified and identified propensity accounted for 105.63% and 39.92% of the trend,
respectively.
To further examine which group of firms is important in driving the decreasing
propensity to pay taxes in 1988-2013, we repeat the analysis for the sub-samples based on
TAX_ATi,t, ROAi,t, and Fn,i,t are taxi,t, pretaxi,t, and fn,i,t scaled by a firm’s total assets,
respectively, in order to overcome heteroskedasticity, following Frank et al. (2009). In this
annual regression, Interceptt measures firms’ tax-saving efforts (as a portion of total assets)
that are not correlated with pretax income. The vector of βn,t captures the sensitivities of tax
responsibilities to firm characteristics Fn,i,t. If firms are more aggressive in avoiding tax
through managing the variables captured by Fn,i,t, we should observe more negative values for
βn,t.
Given the regression coefficients in Equation (3), the annual average tax responsibilities
as a portion of total assets can be calculated by
ATTAX _ t = Interceptt + (Tt - θt) × ROA t + ∑n βn,t × F̅n,t. (4) 5 Including a constant on the right side of Equation (2), assuming that each firm can save a fixed amount of tax expense in year t, does not change our results, and the corresponding variable of this constant in Equation (3) does not have a significant coefficient.
10
If we pick the first year (year 0) in the sample as the base year for studying the time trend of
where TAX_AT is the total income taxes (COMPUSTAT item TXT) scaled by total assets
(COMPUSTAT item AT).6 All of the independent variables in Equation (6) are scaled by
total assets and are winsorized at the 1% and 99% level to amend the effect of outliers on the
results. ROA is pretax income (COMPUSTAT item PI). LEV is the total debt (COMPUSTAT
item DLTT plus DLC). According to prior research, leverage is negatively related to effective
tax rate because interest on debt is tax deductible (Skickney and McGee, 1982). RND is
6 Untabulated results using cash taxes paid (COMPUSTAT item TXPD) scaled by total assets and using
effective tax rate (COMPUSTAT item TXT divided by item PI) as alternative measures of tax responsibilities
generate similar results.
12
research and development expense (COMPUSTAT item XRD). Expenditure on research and
development is deductible for tax purposes, and thus is predicted to be negatively related to
our dependent variable (Dyreng et al., 2016). We include PPE and INTAN because of the
differential rules for fixed and intangible assets between financial and tax accounting. PPE is
the total property, plants, and equipment (COMPUSTAT item PPENT). Research has also
identified a negative relationship between capital intensity and effective tax rate due to the
government’s favorable tax treatment of investments in depreciable assets (Skickney and
McGee, 1982; Manzon and Plesko, 2002). INTAN is the intangible assets (COMPUSTAT
item INTAN). The relationship between the percentage of intangible assets and effective tax
rate can be positive or negative depending on the amortization periods (Manzon and Plesko,
2002). EQINC is the equity income in earnings (COMPUSTAT item ESUB). This variable is
included to control for differential book and tax treatments of consolidated earnings based on
the equity method and is expected to be negatively related to TAX_AT (Chen et al., 2010).
ΔNOL is the change in a firm’s net operating loss carryforward (COMPUSTAT item TLCF).
We include ΔNOL to control for its association with changes in the valuation allowance
account (Frank et al., 2009) and predict it to be positively related to TAX_AT. CAPX is the
capital expenditure (COMPUSTAT item CAPX). We expect a negative relationship between
TAX_AT and CAPX due to the government’s favorable tax treatment encouraging investments.
SPI is the amount of special items (COMPUSTAT item SPI), which are deductible for tax
13
purposes later than when they are recognized for financial reporting purposes, and hence SPI
is predicted to be negatively related to TAX_AT (Dyreng et al., 2016).
Appendix 1 presents the descriptive statistics for the variables in Equation (6). The mean
(median) of TAX_AT equals 4.3% (3.45%). The mean (median) of ROA is approximately
11.3% (9.5%). The relatively high value of ROA results from our sample only comprising
profitable firms. Consistent with other studies on tax avoidance (e.g., Dyreng et al., 2016),
our sample firms are highly levered (the mean/median LEV is 22.2%/20.5%), and SPI has a
negative mean of -0.1% and a zero median.
3. Explaining the trend in tax payments
3.1 Declining tax payments in the 1956-2013 sample period
Our sample includes all of the U.S. public firms in COMPUSTAT from 1956 to 2013,
excluding firm-years (1) from financial (SIC codes 6000-6999) and utilities (SIC codes
4900-4999) industries, (2) with negative assets or total liabilities exceeding total assets, (3)
with missing information for the variables in Equation (6), and (4) with zero or negative
pretax income. These screenings result in a sample of 153,949 firm-years from 15,091 firms.
We plot the average effective tax rate (ETR) for each year in Figure 1 and the average
ratio of income tax to total assets (TAX_AT) in Figure 2. Both Figures 1 and 2 demonstrate a
declining trend in tax payments in the past 50 years. The mean ETR was about 47% in 1956
but declined to about 28% in 2013. Even after the last overhaul on the statutory tax rate in
14
1987, since which the statutory tax rate has been stable (34-35%), the trend still exists.
TAX_AT declined from 8% in 1956 to under 3% in 2013, suggesting a yearly trend of
-0.086%.
3.2 Triennial regressions
To study the changes in firms’ propensity to pay taxes, we estimate annual regressions
based on Equation (6) and track the time-series trends of the regression coefficients. To
reduce the effect of firms’ inter-temporal tax management and facilitate concise presentations
in tables and figures, we estimate triennial regressions for each three-year window between
1961 and 2013, which generates qualitatively similar results as the annual regressions. Panels
A and B in Table 1 present the coefficients for the 1961-1987 and 1988-2013 periods,
respectively. The signs of the coefficients are largely consistent with our predictions in
Section 2. Several patterns in the coefficient time series stand out. First, the coefficient on
ROA shows a mostly monotonous down-trend over the entire sample period, from 0.499 in
1961-1963 to 0.285 in 2012-2013. We also find that this coefficient has been drifting further
underneath the contemporary statutory tax rate, which should have been the coefficient on
ROA if the other control variables captured all of the untaxed income. This finding suggests
that the portion of pretax income that corporations claim to be untaxable has been growing,
and hence the sensitivity of tax responsibilities to pretax income declines. Second, the
coefficients of LEV and RND become more negative over time, suggesting more tax-saving
15
propensities from holding debt and R&D investment. This should mostly result from growing
tax deductions for interest expenses and innovative investments. Third, the coefficients of
PPE and EQINC increase over time, suggesting that corporations have been less able to save
taxes from holding fixed assets and from the exemption of earnings arising from holding
other companies’ equity. Fourth, the intercept does not have a statistically significant trend in
the sample, and hence the tax savings demand curve has not shifted up or down in the last 50
years. Fifth, R2 dropped from 91.7% in 1961-1963 to 58.9% in 2012-2013. The
cross-sectional difference in tax savings has, to a larger extent, been driven by factors other
than the controls in Equation (6) in recent years. In our empirical setting, the average effect of
these uncontrolled factors is captured by the extent to which the coefficient of ROA deviates
from the statutory tax rate.
The results in Table 1 suggest that the descending trend in tax payments may result from
changes in firms’ propensity to pay taxes (i.e., the decreasing coefficients of ROA, LEV, and
RND over time). Admittedly, changing firm characteristics, in addition to a declining
statutory tax rate, provide another non-exclusive explanation. Table 2 reports the means of the
firm characteristics in Equation (6) for every three-year window. We also estimate the
coefficient, namely Trend, of regressing the means of each variable on the first year of each
three-year window and report Trend in the bottom row of Table 2. We find that firms have
been less profitable (i.e., decreasing ROA over time) and investing more in R&D that is
16
eligible for tax credits, and Trend of these two variables are statistically significant. These
changes can possibly drive lower tax payments. In summary, the findings in Tables 1 and 2
necessitate further decomposition analysis of the tax payment trend.
3.3 Decomposing the declining trend in tax payment
We now investigate the sources leading to the declining trend in tax payments. We first
estimate a regression model based on Equation (6) for the time window 1956-1960 as the
base period (year 0 in Equation 5).7 Following Equation (5), we decompose the average
TAX_AT into Expected_Tax and Propensity. Expected_Tax is the total expected tax payment.
It captures the trend of expected tax payment due to the changes in both firm characteristics
and the statutory tax rate. Propensity is firms’ propensity to pay taxes. It is the difference
between TAX_AT and Expected_Tax and captures firms’ propensity to pay taxes due to the
changes in the coefficients of firm characteristics and other unidentified factors associated
with tax payment.
We further decompose Expected_Tax into .Expected_Tax_Firm and Expected_Tax_Rate.
Expected_Tax equals the sum of Expected_Tax_Firm and Expected_Tax_Rate.
Expected_Tax_Firm for each three-year window t (e.g., 1961-1963 and so on) is computed by
substituting the means of firm-level variables from three-year window t into the base-period
regression (1956-1960):
7In untabulated tests, we find that alternative base periods such as 1956-1969 generate similar results.
17
0.00229 + 0.52 × ROA t + 0.00204 × LEV t – 0.00114 × RND t – 0.0145 × PPE t + 0.0146 ×
INTAN t – 0.543 × EQINC t – 0.0377 × NOL∆ t + 0.000378 ×CAPX t – 0.0147× SPI t,
where the coefficients are from the base-period regression. Expected_Tax_Rate captures the
effect of the changing statutory tax rate on average tax payment and is calculated as {∆Tt ×
ROA t}, where ΔTt is the difference in the average statutory tax rate between three-year time
window t (Tt) and the base period (T0).
Figure 3 plots the actual average tax payment (TAX_AT) and expected tax payment
controlling for contemporary firm characteristics (Expected_Tax_Firm) and total expected tax
payment controlling for both contemporary firm characteristics and the statutory tax rate
(Expected_Tax) across the time windows. In the time window of 1988-1991, both
Expected_Tax and TAX_AT exhibit a sudden drop, in line with the tax rate overhaul in 1987.
More importantly, we observe distinct patterns in Figure 3 before and after 1988. First, the
declining trend in TAX_AT exists throughout the sample period but slows down somewhat
after 1988. Second, Expected_Tax largely overlaps with TAX_AT before 1988, implying that
changes in firm characteristics and the statutory tax rate explain most of the declining trend in
tax payments, and firms’ propensity to pay taxes is stable between 1961 and 1987. Third,
after 1988, the declining trends in Expected_Tax_Firm and Expected_Tax disappear and there
is a growing divergence between TAX_AT and Expected_Tax, suggesting that changes in firm
characteristics and the statutory tax rate cannot explain the trend of TAX_AT in the post-1988
18
period, and Propensity (the difference between TAX_AT and Expected_Tax) become the main
factor.
To accentuate the dynamics of the propensity to pay taxes over time, in Figure 4 we
plot Propensity; that is, the deviation of TAX_AT from Expected_Tax. Compared with the
1961-1987 period, Propensity shows a dramatic declining trend during the 1988-2013 period
from 0.00136 in 1988-1990 to -0.00799 in 2012-2013. This suggests that after 1988, firms
may have become more aggressive in tax saving with strategies not captured by the
controlled firm-level variables.
Figures 3 and 4 suggest that the declining trend in tax payments is driven by different
factors before and after 1988. Changes in firm characteristics and the statutory tax rate are the
major reasons before 1988, whereas a decreasing propensity to pay taxes takes their place
after 1988. It is necessary to conduct sub-period analyses separately for 1961-1987 and
1988-2013.
In Panel A of Table 3 we present, for every three-year window, the actual tax payment
(TAX_AT), expected tax payment (Expected_Tax), and firms’ propensity to pay taxes
(Propensity), together with their sub-components, i.e., expected tax payment controlling for
firm characteristics (Expected_Tax_Firm), expected tax payment controlling for statutory tax
rate (Expected_Tax_Rate), identified propensity to pay tax (Propensity_Identified) and
unidentified propensity to pay tax (Propensity_Unidentified). In addition, we report the year
19
trends (Trend) of TAX_AT and each component for the full sample period and sub-periods of
1961-1987 and 1988-2013.
Over the full sample period (1961-2013), changes in firm characteristics and the
statutory tax rate together explain 88.53% of the trend in TAX_AT. Expected_Tax_Firm,
which captures the effects of changing firm characteristics, has a year trend of -0.00027,
meaning on average Expected_Tax_Firm decreases 0.00027 per year; such trend is equivalent
to 36.09% of for the trend of TAX_AT. On the other hand, Expected_Tax_Rate, which
captures the effect of a changing statutory tax rate, has a year trend of -0.039% (equivalent to
52.44% of that for TAX_AT). The remaining 11.47% of the trend of TAX_AT is explained by
firms’ decreasing propensity to pay taxes (Propensity).
The statistics for the sub-periods in Panel A of Table 3 are consistent with the
observations in Figures 3 and 4. In the 1961-1987 sub-period, Expected_Tax shows a year
trend of -0.00069, accounting for 101.51% of the trend of TAX_AT, whereas Propensity
shows a slightly upward trend, the magnitude of which equals 1.51% of that of TAX_AT. The
trend of TAX_AT is hence mostly explained by changing firm characteristics and the statutory
tax rate over the 1961-1987 period, which accounts for 66 and 35.51% of the total trend,
respectively. In the 1988-2013 sub-period, the trend of TAX_AT (-0.0003) is about 44% of
that in 1961-1987 (-0.00068). Interestingly, both Expected_Tax_Firm and Expected_Tax_Rate
20
show an upward trend, but it is completely submerged by a sharply declining trend in
Propensity, leading to a declining overall trend in TAX_AT. The two sub-components of
Propensity—Propensity_Identified and Propensity_Unidentified—explain 39.92 and
105.63% of the TAX_AT trend, respectively.
In summary, the declining trend in firms’ actual tax payments is driven by distinct
factors before and after 1988 for our sample. Changes in firm characteristics and reductions
in the statutory tax rate explain the trend over the 1961-1987 period, and a decreasing
propensity to pay taxes plays that role over the 1988-2013 period. Accordingly, we examine
which firm characteristics and propensity types are important for the trends over the
1961-1987 and 1988-2013 periods, respectively.
In Panel B of Table 3, we report a breakdown in the trend of Expected_Tax attributable
to the changes in firm characteristics and the statutory tax rate for the full sample period and
two sub-periods. Among the firm characteristics, declining profitability is the major factor
driving the trend of TAX_AT across the sub-periods. For the entire sample period, changes in
ROA generate a decreasing trend of -0.00035 in tax payments, accounting for 47.29% of the
total decreasing trend in TAX_AT. Consistent with the results in Panel A, the effect of
declining profitability is much smaller in 1988-2013 than in 1961-1987; specifically, the
contributions of declining profitability are 67.40 and 17.59% of the trend in TAX_AT in
21
1961-1987 and 1988-2013, respectively. None of the other firm characteristics contribute
more than 10% of the total trend of TAX_AT in any period.
In Panel C of Table 3, we report a breakdown in the trend of Propensity. Our focus is the
1988-2013 sub-period because the effect of Propensity on the trend of TAX_AT is
concentrated in that period. Propensity_Unidentified dominates Propensity_Identified in
driving the declining trend in actual tax payments, with the effect of a larger –θt in Equation
(5), which measures the deviation of ROA’s coefficient in a time window from the statutory
tax rate. The changes in –θt generate a trend of -0.00028 in tax payments and explain 92.97%
of the trend of TAX_AT in 1988-2013. In other words, there might be an increasing number of
unidentified strategies used by firms for tax saving beyond the traditional approaches
captured by the firm characteristics (except for ROA). This may, in turn, drives the coefficient
of ROA away from the statutory tax rate. With regard to Propensity_Identified, three
firm-level factors stand out. The changes in the coefficients of LEV, RND, and INTAN explain
10.48, 18.73, and 19.90% of the trend of TAX_AT in 1988-2013, suggesting that either firms
have been managing debt, R&D investment, and intangible assets to be more eligible for tax
savings, or government has been providing more tax deductions or credits for firms holding
innovative investments.
3.4 Sub-sample tests
The findings reported in Section 3.3 show that a decreasing propensity to pay taxes,
22
especially the unidentified portion, plays a major role for the downward trend in tax
payments since 1988. This trend of unidentified propensity can either be due to the
availability of more unobserved opportunities for firms to reduce their tax burden, or more
likely firms’ decreasing willingness to pay taxes (i.e., increasing motivations for tax saving).
We are particularly interested in the trend of Propensity_Unidentified, because it is very
relevant to whether tax enforcement should be tightened. In this section, we provide some
insights into why firms have become more aggressive about tax saving by examining whether
the downward trend is pervasive across all firms, or limited to (or more evident in) particular
sub-sets of firms. To this end, we estimate annual regressions between 1991 and 2013 based
on Equation (6)8 and reproduce the analysis on Propensity_Unidentified in Panel A of Table
3 for different sub-samples, divided according to factors that are related to tax-saving
motivations in 1991-2013, using 1988-1990 as the base period. 9 We report the
Propensity_Unidentified for different groups of firms in Table 4 and plot the
Propensity_Unidentified for each sub-samples in Figure 5. In addition, we apply chaw test to
check whether there is significant difference in the trend of Propensity_Unidentified between
different groups and report the p-value in the bottom row of Table 4.
3.4.1 Multinational vs. domestic firms
8 Standard errors in annual regressions are clustered at the firm level to alleviate concerns over residual serial correlation. 9 For the sub-sample based on corporate governance, we use 1990 as the base year, constrained by data availability for the G-index.
23
There have been growing concerns about the tax avoidance activities of multinational
firms (Dyreng et al., 2016). Empirical evidence has shown that U.S. firms have become more
active in shifting income out of the United States (Klassen and Laplante, 2012), and that
multinational firms have lower global effective tax rates than domestic firms (Rego, 2003).
Nonetheless, Dyreng et al. (2016) find that the decline in the effective tax rate of
multinational firms is similar to that of domestic firms. We focus on the unidentified
propensity to pay taxes and analyze whether the decline is mainly driven by multinational
firms. We divide our sample into multinational and domestic firms for each year after 1988
and reproduce the analysis in Panel A of Table 3, only reporting Propensity_Unidentified for
the two sub-samples in columns 1 and 2 of Table 4. We classify a firm as multinational in
year t if either its pretax foreign income (COMPUSTAT item PIFO) or the absolute value of
its foreign tax expense (COMPUSTAT item TXFO) is greater than zero. Table 4 shows that
both domestic and multinational firms have experienced a significantly declining trend in
their unidentified propensity to pay taxes, but the magnitude of the declining trend of
multinational firms is significantly greater (Trend = -0.0004) than that of domestic firms
(Trend = -0.0003), with a p-value of chaw test equal to 0.042.
3.4.2 Top profitable firms
As Manzon and Plesko (2002) argue, more profitable firms can make more efficient use
of tax deductions, credits, and exemptions than less profitable firms. Moreover, earnings
24
become concentrated in a small number of U.S. corporations (DeAngelo, DeAngelo, and
Skinner, 2004). In our sample, the 30 firms with the highest pretax income account for
around 50% of the total pretax income. Thus, it is meaningful to investigate whether the
decrease in Propensity_Unidentified is concentrated in firms with large earnings numbers.
We present Propensity_Unidentified by year in columns 3 and 4 of Table 4 for the 30 firms
with the highest earnings and the rest of the sample, respectively. The decreasing trend in
Propensity_Unidentified is significantly more pronounced among the top 30 profitable firms
(-0.0011) than in the other firms (-0.0003), with a p-value of chaw test equal to 0.038.
3.4.3 Corporate governance
Recent studies examining the link between corporate governance and tax avoidance
provide evidence that the variation in corporate governance explains the cross-sectional
differences in the level of tax avoidance. Desai et al. (2007) argue that self-interested
managers can structure firms in complex ways to facilitate transactions that reduce corporate
taxes and divert corporate resources for their private interests. Wilson (2009) states that tax
sheltering can be a tool for wealth creation for shareholders in well-governed firms but not in
poorly governed firms. Armstrong et al. (2015) observe a positive relation between corporate
governance and tax avoidance in the lower tail of the tax avoidance distribution, and a
negative relation in the upper tail. We investigate whether corporate governance influences
the time trend of tax saving. We use the G-index constructed by Gompers, Ishii, and Metrick
25
(2003) as our proxy for corporate governance. The G-index is a measure of managerial
entrenchment. Firms with a high G-index value are considered as having more entrenched
management and thus weaker corporate governance. We divide our sample from 1990 to
2010 into quartiles for each industry-year (based on two-digit SIC codes) according to the
strength of corporate governance. Columns 5 and 6 of Table 4 report annual
Propensity_Unidentified for quartiles with the strongest and weakest corporate governance.
Both quartiles experienced a significant decrease in Propensity_Unidentified between 1988
and 2010, but the downward trend was more pronounced in strongly governed firms than in
poorly governed firms (-0.0005 vs. -0.0004), with a p-value of chaw test equal to 0.042.
3.4.4 Earnings per share
The literature also shows that managers engage in tax planning specifically with the
objective of improving accounting outcomes (Desai and Dharmapala, 2009; Robinson, Sikes,
and Weaver, 2010; Shackelford, Slemrod, and Sallee, 2011; Graham, Hanlon, Shevlin, and
Shroff, 2014). To test whether the downward trend in an unidentified propensity to pay taxes
is driven by financial reporting incentives, we divide our sample according to whether a
firm’s reported EPS increases in the current year from the previous year.10 Graham et al.
(2014) find that a reported EPS increase is a very important signal of tax planning. Columns
7 and 8 of Table 4 present the evolution of Propensity_Unidentified among firms with
10 We use reported EPS (COMPUSTAT item EPSPX) divided by Share Adjustment Factor (COMPUSTAT item AJEX) in year t minus (EPSPX/AJEX) in year t-1 as the measure for an EPS increase/decrease.
26
increased and decreased EPS. The magnitude of the decreasing trend in
Propensity_Unidentified for firms with increased EPS (Trend = -0.0005) is significantly
greater than that for firms with decreased EPS (Trend = -0.0001), with a p-value of chaw test
equal to 0.004.
3.4.5 Cash flow volatility
According to precaution theory for holding cash, firms with greater cash flow volatility
are motivated to hold more cash. Bates et al. (2009) find that the increase in cash ratio in
1980-2006 is concentrated in firms that experience the greatest increase in cash flow
volatility. It is possible that firms with high cash flow volatility have stronger motivations to
save taxes as a way of accumulating cash. We test whether this consideration sheds light on
firms’ decreasing unidentified propensity to pay taxes by calculating Propensity_Unidentified
for firms with high and low cash flow volatility. We follow Bates et al. (2009) to measure
cash flow volatility. For each firm-year, we compute the standard deviation of cash flow to
assets ratio11 for the previous 10 years and require that each firm have at least three
observations. We divide our sample into quartiles for each industry-year by cash flow
volatility. Columns 9 and 10 of Table 4 present Propensity_Unidentified for the top and
bottom quartiles. The time trend of Propensity_Unidentified is significantly larger in the top
(-0.0008) than in the bottom (-0.0005) quartile, with a p-value of chaw test equals to 0.000. 11 We use net cash flow from operating activities (COMPUSTAT item OANCF) divided by total assets as cash flow to assets. If OANCF is missing, we replace it with operating income before depreciation (OIBDP) - interest expense (XINT) - tax expense (TXT) - dividends (DVC).
27
3.4.6 Changes in cash holdings
Similar to Sub-section 3.4.5, we estimate Equation (6) annually for quartiles divided
according to the change in cash holdings in each industry-year. Columns 11 and 12 of Table
4 report the annual Propensity_Unidentified of the top and bottom quartiles. The firms with
the largest increases in cash holdings exhibit a greater decline in Propensity_Unidentified
(Trend = -0.0004), which is two times as big as the firms with the highest reductions in cash
holdings (Trend = -0.0002), with a p-value of chaw test equal to 0.008.
3.4.7 Future financing needs
Following the previous two sub-sections, tax planning can also provide financing for
future investment needs. We divide our sample into four quartile sub-samples according to
the level of future financing needs. We measure future financing needs by the ratio of capital
investment12 over total assets in year t+1. Columns 13 and 14 of Table 4 show that the
Propensity_Unidentified of the firms with the greatest future financing needs have a
significantly greater downward trend (Trend = -0.0005) than the firms in the lowest quartile
(Trend = -0.0003), with a p-value of chaw test equal to 0.009.
3.4.8 A summary of the sub-sample analyses
The sub-samples are constructed based on factors related to tax-saving motivations, and
we find that the declining trend in Propensity_Unidentified is ubiquitous across all of the 12 The capital investment level is calculated as the sum of capital expenditure (COMPUSTAT item CAPX) and acquisition (COMPUSTAT item AQC).
28
sub-samples. We also find that the trend is more pronounced in some sub-samples, implying
that the importance of these factors in tax-saving decisions has been growing. First, the
propensity to pay taxes exhibited by multinational corporations and corporations with
large-number earnings declined faster in 1988-2013, suggesting stricter tax enforcement
among these tax payers. Second, strong corporate governance seems to increasingly enable
managers to save taxes for shareholders. Third, corporations have been, to a larger extent,
relying on tax savings to increase earnings numbers. Finally, tax savings have become an
increasingly important source of financing for buffering cash flow shocks and investment
needs.
4. Robustness tests
An alternative explanation for the declining trend of the unidentified propensity is the
changes of firms’ industry attribute. To rule out this concern, we divide our sample into
subsamples according to Fama French 12 industry classification and then estimate Equation
(6) for each industry over the period of 1988 – 2013. Table 5 reports the
Propensity_Unidentified for each Fama French 12 Industry13 and shows that the declining
trend in Propensity_Unidentified is ubiquitous across different industries, suggesting that the
declining trend is not driven by changes in the composition of firms from different industries
and is due to firms’ declining propensity to pay taxes unexplained by book-tax related firm
13 We exclude Fama French 8 – Utilities and 11 – Financial industries in order to align with our sample selection.
29
characteristics in the literature.
5. Conclusion
We examine U.S. firms’ tax payments in 1956-2013 and find that the ratio of income tax
over total assets (TAX_AT) continually decreases by around 7 basis points annually. We
decompose this trend into three components: changes in firm characteristics, the statutory tax
rate, and firms’ propensity to pay taxes. We find that in 1961-2013, 36.09, 52.44, and 11.47%
of the declining trend in TAX_AT result from changes in firm characteristics, the statutory tax
rate, and the propensity to pay taxes, respectively. More importantly, we observe a structural
change in firms’ tax paying trend. Before 1988, the trend was mainly driven by changes in
firm characteristics and the statutory tax rate, whereas since 1988, a decreasing propensity to
pay taxes played a major role. Not only does this study confirm the observation made by
policy makers and scholars that corporations have been more aggressive in their tax savings,
but it also reveals that the recent decline in tax payment is not driven by changing firm
characteristics, but rather by a growing propensity to avoid taxes, which suggests stricter tax
enforcement.
Further, we find that the decreasing trend in corporate tax payments is prevalent among
firms regardless of their multinationality, earnings scale, corporate governance strength,
earnings management, and cash demand. However, the declining trend in tax payment is
more pronounced among corporations with multinational operations, high earnings, strong
30
governance, and a high demand for earnings management and cash. These findings provide
possible explanations for the declining trend in tax payments and insights for policy makers.
31
References
Armstrong, C. S., Blouin, J. L., Jagolinzer, A. D., Larcker, D. F., 2015. Corporate governance, incentives, and tax avoidance. Journal of Accounting and Economics 60 (1), 1–17.
Auerbach, A. J., Poterba, J. M., 1987. Why have corporate tax revenues declined? Unpublished working paper. Massachusetts Institute of Technology.
Bates, T. W., Kahle, K. M., Stulz, R. M., 2009. Why do U.S. firms hold so much more cash than they used to? The Journal of Finance 64 (5), 1985–2021.
Chen, S., Chen, X., Cheng, Q., Shevlin, T., 2010. Are family firms more tax aggressive than non-family firms? Journal of Financial Economics 95 (1), 41–61.
DeAngelo, H., DeAngelo, L., Skinner, D. J., 2004. Are dividends disappearing? Dividend concentration and the consolidation of earnings. Journal of Financial Economics 72, 425–456.
Desai, M., 2003. The divergence between book and tax income. In: Poterba, J.M. (Ed.), Tax Policy and the Economy 17. MIT Press, Cambridge, MA, pp. 169–206.
Desai, M., Dharmapala, D. 2009. Corporate tax avoidance and firm value. Review of Economics and Statistics 91, 537–546.
Desai, M., Dyck, I., Zingales, L., 2007. Theft and taxes. Journal of Financial Economics 84, 591–623.
Dyreng, S. D., Hanlon, M., Maydew, E. L., 2008. Long-run corporate tax avoidance. The Accounting Review 83 (1), 61–82.
Dyreng, S. D., Hanlon, M., Maydew, E. L., Thornock, J. R., 2016. Changes in corporate effective tax rates over the past twenty-five years. Unpublished working paper. Duke University.
Fama, E., French, K., 2001. Disappearing dividends: Changing firm characteristics or lower propensity to pay? Journal of Financial Economics 60 (1), 3–43.
Frank, M. M., Lynch, L. J., Rego, S. O., 2009. Tax reporting aggressiveness and its relation to aggressive financial reporting. The Accounting Review 84 (2), 467–496.
Gompers, P. A., Ishii, J. L., Metrick, A., 2003. Corporate governance and equity prices. Quarterly Journal of Economics 118 (1), 107–155.
Graham, J., Hanlon, M., Shevlin, T., Shroff, N., 2014. Incentives for tax planning and avoidance: Evidence from the field. The Accounting Review 89 (3), 991–1023.
32
Gupta, S., Newberry, K., 1997. Determinants of the variability in corporate effective tax rates: Evidence from longitudinal data. Journal of Accounting and Public Policy 16 (1), 1–34.
Klassen, K. J., Laplante, S. K., 2012. Are U.S. multinational corporations becoming more aggressive income shifters? Journal of Accounting Research 50 (5), 1245–1285.
Manzon, G., Plesko, G., 2002. The relation between financial and tax reporting measures of income. Tax Law Review 55, 175–214.
Mills, L., 1998. Book-tax differences and Internal Revenue Service adjustments. Journal of Accounting Research 36 (2), 343–356.
Rego, S., 2003. Tax-avoidance activities of U.S. multinational corporations. Contemporary Accounting Research 20 (4), 805–833.
Robinson, J., Sikes, S., Weaver, C., 2010. Performance measurement of corporate tax departments. The Accounting Review 85, 1035–1064.
Shackelford, D., Slemrod, J., Sallee, J., 2011. A unifying model of how the tax system and accounting rules affect corporate behavior. International Tax and Public Finance 18, 461–494.
Stickney, C. P., McGee, V. E., 1982. Effective corporate tax rates the effect of size, capital intensity, leverage, and other Factors. Journal of Accounting and Public Policy 1 (2), 125–152.
Wilson, R., 2009. An examination of corporate tax shelter participants. The Accounting Review 84, 969–999.
33
Figure 1 – Annual Effective Tax Rate
This figure plots the average annual effective tax rate (ETR) of profitable U.S. public firms from 1956 to 2013. ETR is measured as total income taxes (COMPUSTAT item TXT) divided by pre-tax income (COMPUSTAT item PI).
34
Figure 2 – Annual Actual Tax Payment Relative to Total Assets
This figure plots the average annual TAX_AT of profitable U.S. public firms from 1956 to 2013. TAX_AT is measured as total income taxes (COMPUSTAT item TXT) divided by total assets (COMPUSTAT item AT).
35
Figure 3 – Actual Tax Payment, Expected Tax Payment due to Changing Firm Characteristics, and Total Expected Tax Payment
This figure plots the mean of actual tax payment (TAX_AT), expected tax payment due to changing firm characteristics (Expected_Tax_Firm), and total expected tax payment (Expected_Tax) of every three-year window of sample period 1961 to 2013. Refer to Table 3 Panel A for variable definitions.
36
Figure 4 – Firms’ Propensity to Pay Taxes
This figure plots for firms’ propensity to pay taxes (Propensity). Refer to Table 3 Panel A for variable definitions
37
Figure 5 – Unidentified Propensity to Pay Tax for Different Sub-samples This figure plots firms’ unidentified propensity (Propensity_Unidentied), i.e., firms’ propensity to pay taxes unrelated to identified tax saving related firm characteristics, for various sub-samples. Refer to Table 3 Panel A for variable definition. We plot Propensity_Unidentied separately for multinational vs. domestic firms (Panel A), the top 30 vs. non-top 30 profitable firms (Panel B), top vs. bottom corporate governance quartiles (Panel C), firms with increased vs. decreased EPS (Panel D), top vs. bottom cash flow volatility quartiles (Panel E), top vs. bottom changes in cash holding quartiles (Panel F), and top vs. bottom future financing needs quartiles (Panel G) of 1991-2013 period. We classify a firm as multinational in year t if either its pretax foreign income (COMPUSTAT item PIFO) or the absolute value of its foreign tax expense (COMPUSTAT item TXFO) is greater than zero. Pretax income is the proxy for firms’ profitability. Gompers, Ishii, and Metrick (2003)’s G-index is the proxy for corporate governance. EPS increase/decrease is measured as reported EPS (COMPUSTAT item EPSPX) divided by Share Adjustment Factor (COMPUSTAT item AJEX) in year t minus (EPSPX/AJEX) in year t-1. Cash flow volatility is measured as the standard deviation of cash flow to assets ratio (COMPUSTAT item OANCF divided by item AT) of the past 10 years. Change in cash is measured as cash (COMPUSTAT item CHE) in year t minus that in year t-1 over total assets in year t. Future financing needs is measured as the ratio of capital investment over total assets(COMPUSTAT item CAPX plus item AQC divided by item AT) in year t+1. Quartiles are based on the corresponding variable for each two-digit SIC industry-year. Panel A Panel B Panel C Panel D
Panel E Panel F Panel G
38
Appendix 1 Descriptive Statistics
This table reports the descriptive statistics on the profitable U.S. public firms from 1956 to 2013. TAX_AT is actual tax payment, measured as total income taxes (COMPUSTAT item TXT) divided by total assets (COMPUSTAT item AT). ROA is return on assets, calculated as pre-tax income (COMPUSTAT item PI) divided by total assets. LEV is leverage, constructed as the ratio of total debt (COMPUSTAT item DLTT plus DLC) to total assets. RND is the ratio of research and development expense (COMPUSTAT item XRD) to total assets. PPE is constructed as the ratio of total property, plants, and equipment (COMPUSTAT item PPENT) to total assets. INTAN is calculated as the ratio of intangible assets (COMPUSTAT item INTAN) to total assets. EQINC is equity income in earnings, computed as the ratio of equity income in earnings (COMPUSTAT item ESUB) to total assets. ΔNOL is the change in a firm’s net operating loss carryforward (COMPUSTAT item TLCF) in year t-1 divided by total asset. CAPX is the capital expenditures, measured as capital assets (COMPUSTAT item CAPX) divided by total assets. SPI is the special items, measured as special items (COMPUSTAT item SPI) divided by total assets.
Table 2: Firm Characteristics Trends This table reports the mean of each variable of all sample firms in every three-year window from 1961 to 2013. We calculate the average annual change in each variable (Trend) by estimating a regression model with variables in each column (e.g., TAX_AT, ROA, and so on) as the dependent variable and Year as the independent variable. Trend is the coefficient on Year, where Year is the beginning year of each three-year window. Trend values that are significant at less than the 10% level are reported in bold. The variables are defined in Appendix 1.
Year TAX_AT ROA LEV RND PPE INTAN EQINC ΔNOL CAPX SPI
Table 3: Actual Tax Payment, Expected Tax Payment, and Propensity to Pay Taxes
Panel A: Actual Tax Payment, Expected Tax Payment, and Propensity to Pay Taxes across Three-year Windows
This table reports actual tax payment, expected tax payment, and propensity to pay taxes of all sample firms for every three-year window of 1961-2013 period. TAX_AT is the actual tax payment, measured as total income taxes divided by total assets. The expected tax payment (Expected_Tax) captures the effects of changes in both firm characteristics and the statutory tax rate. It is the sum of two components: Expected_Tax_Firm and Expected_Tax_Rate. Expected_Tax_Firm captures the trend in expected tax payment due to the changing firm characteristics. It is measured as the sum of the coefficients of Equation (6), estimated using the base period (1956-1960) data multiplied by their corresponding contemporary firm characteristics averaged for every three-year window (e.g., 1961-1963, and so on). TAX_ATi,t = Interceptt + (Tt - θt) × ROAi,t + β1,t × LEVi,t + β2,t × RNDi,t + β3,t × PPEi,t + β4,t × INTANi,t + β5,t × EQINCi,t + β6,t × ∆NOLi,t + β7,t × CAPXi,t + β8,t × SPIi,t + ɛi,t (6) Expected_Tax_Rate captures the trend in expected tax payment due to the changing statutory tax rate. It is measured as the difference in contemporary statutory tax rate and the coefficient on ROA estimated from the base period and multiplied by the contemporary ROA averaged for every three-year window. The propensity to pay taxes (Propensity) is the difference between TAX_AT and Expected_Tax. It captures firms’ propensity to pay taxes due to the changes in the coefficients of firm characteristics and other unidentified factors associated with tax payment. It is the sum of two components: Propensity_Identied and Propensity_Unidentified. Propensity_Identied is the identified propensity. It captures the changes in the coefficients of firm characteristics and is measured as the difference between the coefficients of Equation (6) estimated over a contemporary three-year window and those estimated over the base period multiplied by the contemporary firm characteristics. Propensity_Unidentied is the unidentified propensity. It captures firms’ propensity to pay taxes that cannot be explained by the changes in the coefficients. It is measured as the intercept estimated over contemporary three-year window minus the difference in the coefficient on ROA of the contemporary three-year window and that of the base period times the mean of the ROA of the three-year window. We calculate the average annual change of each variable (Trend) by estimating a regression model with variables in each column (e.g., TAX_AT, Expected_Tax_Firm, and so on) as the dependent variable and Year as the independent variable. Trend is the coefficient on Year, where Year is the beginning year of each three-year window. We also calculate the percentage of Trend of each column relative to Trend in actual tax payment (%) for the full sample and sub-period samples 1961-1987 and 1988-2013, respectively. Trend values that are significant at less than the 10% level are reported in bold.
This table reports the estimated expected tax payment by each firm characteristic, i.e., coefficient on each firm characteristic variable of Equation (6) estimated over the base period, multiplied by its contemporary firm characteristic averaged over every three-year window (e.g., 1961-1963, and so on). Intercept0 and βn,0 are estimated from the base period (1956-1960). TAX_ATi,t = Interceptt + (Tt - θt) × ROAi,t + β1,t × LEVi,t + β2,t × RNDi,t + β3,t × PPEi,t + β4,t × INTANi,t + β5,t × EQINCi,t + β6,t × ∆NOLi,t + β7,t × CAPXi,t + β8,t × SPIi,t + ɛi,t (6) T0 is the statutory tax averaged over the base period. ΔTt is the difference between the contemporary statutory tax averaged over every subsequent three-year window and T0. We then estimate a regression model with variables in each column as the dependent variable and Year as the independent variable. Trend is the coefficient on Year, where Year is the beginning year of each three-year window. We also calculate the percentage of Trend of each column relative to Trend in actual tax payment (%) for the full sample and sub-period samples 1961-1987 and 1988-2013, respectively. Trend values that are significant at less than the 10% level are reported in bold.
Panel C: Breaking Down the Propensity to Pay Taxes This table reports firms’ propensity to pay tax by each firm characteristic, i.e., the difference between the coefficient on each firm characteristic variable of Equation (6) estimated over the contemporary three-year window (e.g., 1961-1963, and so on), then that estimated over the base period (1956-1960) multiplied by its corresponding contemporary firm characteristic. TAX_ATi,t = Interceptt + (Tt - θt) × ROAi,t + β1,t × LEVi,t + β2,t × RNDi,t + β3,t × PPEi,t + β4,t × INTANi,t + β5,t × EQINCi,t + β6,t × ∆NOLi,t + β7,t × CAPXi,t + β8,t × SPIi,t + ɛi,t (6) The difference between the coefficient estimated over the base period and that of each three-year window is represented by Δ. θt is the coefficient on ROA for each three-year window. We then estimate a regression model with variables in each column as the dependent variable and Year as the independent variable. Trend is the coefficient on Year, where Year is the beginning year of each three-year window. We also calculate the percentage of Trend of each column relative to Trend in actual tax payment (%) for the full sample and sub-period samples 1961-1987 and 1988-2013, respectively. Trend values that are significant at less than the 10% level are reported in bold.
Table 4: Unidentified Propensity to Pay Tax for Different Groups of Firms This table reports firms’ unidentified propensity to pay taxes (Propensity_Unidentified) for multinational vs. domestic firms (Columns 1 and 2), the top 30 vs. non-top 30 profitable firms (Columns 3 and 4), corporate governance quartiles (Columns 5 and 6), firms with increased vs. decreased EPS (Columns 7 and 8), cash flow volatility quartiles (Columns 9 and 10), changes in cash holding quartiles (Columns 11 and 12), and future financing needs quartiles (Columns 13 and 14) in the 1991-2013 period.14 Trend measures the average annual change in Propensity_Unidentified. It is the coefficient of regressing Propensity_Unidentified on Year. Trend that are significant at less than the 10% level are reported in bold. We apply chaw test to check whether the difference in Trend of Propensity_Unidentified between different groups is significant and report the p-value in the bottom row. Refer to Figure 5 for definitions of sample classification variables and Table 3 Panel A for definition of Propensity_Unidentifieds.
Table 5: Unidentified Propensity to Pay Tax by Fama French 12 Industries This table reports firms’ unidentified propensity to pay taxes (Propensity_Unidentified) by Fama French 12 industries, excludeing Utilities (8) and Financial industries (11). Trend measures the average annual change in Propensity_Unidentified. It is the coefficient of regressing Propensity_Unidentified on Year. Trend that are significant at less than the 10% level are reported in bold. Refer to Table 3 Panel A for definition of Propensity_Unidentifieds. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)