Professional Documents
Culture Documents
b
6
BM
t1
b
7
ACC
t1
b
8
RET
t1
e
t
; 2
where FE is analyst forecast error. If rms increase losses in year t to capture tax refunds and
analysts do not anticipate this behavior, then we expect a negative coefcient on D_NOL, where
again the coefcient on D_NOL represents the effect that is incremental to the association between
expected losses and realized forecast errors estimated by the coefcient on D_NEG. Following
previous research, we control for other determinants of earnings forecast bias. The literature argues
that analysts have incentives to issue optimistic forecasts for rms with poor information
environments to obtain access to management (Francis and Philbrick 1993; Lim 2001). We include
rm size (MV) and analyst coverage (COV) to capture a rms information environment (Atiase
1987; Zhang 2006b). We also include the book-to-market ratio (BM) to control for growth
opportunities, as growth rms have stronger incentives to meet or beat earnings targets, suggesting
a negative coefcient on BM. Prior literature documents evidence indicating that analysts are not
fully rational (see Kothari [2001] for a review). To allow for this possibility, we include RET and
FE
t1
to control for analyst underreaction to the information contained in stock returns and forecast
errors in the prior period. We further include accruals to control for the possibility that analysts do
not fully understand the implication of accruals for future earnings (Bradshaw et al. 2001; Teoh and
Wong 2002).
Table 4, Panel B summarizes the results of the forecast error regressions with and without the
lagged dependent variable as a control. In column (1), the coefcient on D_NEG is only marginally
signicant (coeff. 0.015, t 1.86), providing weak evidence that analysts are surprised by
negative earnings news even when they expect the rm to report a loss. Turning to our variable of
interest, the coefcient on D_NOL is 0.048 (t 6.25), which is much larger than the coefcient
on D_NEG, and indicates that analysts fail to anticipate the additional losses reported by rms with
tax incentives to shift income and that analysts forecast errors are predictable ex ante. We add the
lagged forecast error as a control in column (2) to control for unobservable rm effects, and nd
that the coefcient on D_NOL drops to 0.027 but remains highly signicant (t 3.81).
Economically, analysts forecast errors, scaled by stock price, are about 2.7 percentage points larger
for loss rms with carryback incentives than for rms without. This translates into about $64.7
billion of unexpected loss shifting for rms with incentives to use those losses to obtain cash tax
refunds between 1981 and 2010.
17
16
The observed negative forecast errors for D_NOL 0 rms is consistent with the optimistic bias in analysts
earnings forecasts documented in the prior literature (OBrien 1987). Analysts forecast errors are reliably
negative when the forecast horizon is long, and turn to zero or positive immediately before earnings
announcements (Richardson et al. 2004).
17
We estimate the $64.7 billion as the 2.7 percent multiplied by the average sample rms market value of $1.204
billion and then multiply that by 1,990 rm-year observations with tax incentives in Model (2).
Tax-Motivated Loss Shifting 1671
The Accounting Review
September 2013
Equity Market Response to Tax-Motivated Loss Shifting
Finally, we examine the reaction of equity investors to tax-motivated loss shifting. If investors
naively follow analysts earnings forecasts, which tend to miss the tax incentives for accelerating
losses, then the market price will not fully reect the expected tax benets and the market reaction
will be determined by analysts forecast errors. Alternatively, if investor reactions are determined
by the revision in expected future cash ows and discount rates, then the positive cash ow effects
associated with accelerating a tax loss should lead to more positive (or less negative) market
reactions for rms with the tax-based incentives to shift losses, independent of the size of the
forecast error.
We consider both annual returns and earnings announcement returns and report the results in
Table 5. The rst two columns are based on annual returns, and we estimate them with and without
the analyst forecast error as a control. In the full model, we control for analyst forecast errors, rms
with expected losses at the beginning of the period using D_NEG, as well as size, book-to-market,
and prior returns. Column (1) shows that the partial correlation between D_NOL and annual returns
is positive, albeit insignicant, suggesting that the market reaction to negative earnings
announcements by D_NOL rms is similar to those by non-D_NOL rms, even though D_NOL
rms report much larger losses as shown in Tables 2 and 4. Since D_NOL is negatively correlated
with the forecast error, we add analyst forecast errors as a control variable in column (2), and nd a
positive and signicant coefcient on D_NOL (coeff. 0.074, t 2.17), implying that the annual
stock return of a rm with tax-motivated loss-shifting incentives is about 7.4 percent greater than a
loss rm without such incentives, holding analyst forecast errors constant.
In columns (3) and (4), we use abnormal return during the earnings announcement window as
the dependent variable. This allows us to focus on the importance of the earnings release in
conveying information about cash ow effects of the tax loss. Similar to the annual return
regressions, we nd a positive yet insignicant coefcient on D_NOL without controlling for
analyst forecast errors, and this turns positive and signicant once we include analyst forecast errors
(coeff. 0.006, t 3.23). Overall, the results suggest that investors understand either the tax
benets of the loss carryback, the transitory nature of the accelerated losses claimed by carryback
rms, or both.
In summary, we nd that the market reaction to earnings reported by rms with incentives to
accelerate tax losses is less negative than the market reaction to earnings by rms without such
incentives. This occurs despite the fact that rms with tax incentives to shift losses actually report
larger losses using both prior earnings and analyst forecasts as a benchmark, and is consistent with
these losses being transitory, increasing cash ows and providing liquidity.
18
V. ADDITIONAL ANALYSIS AND SENSITIVITY CHECKS
Cross-Sectional Variation in Tax-Motivated Loss Shifting
In Table 4, we provide some evidence of the time-series variations in rms incentives to
engage in tax-motivated loss shiftingrms have particularly strong tax incentives to shift losses
around TRA 86. As Maydew (1997) shows, TRA 86 reduced the corporate tax rate from 46 percent
in 1986 to 34 percent in 1988, providing rms with stronger incentives to accelerate losses between
18
Our market reaction results are in contrast to those in Shane and Stock (2006), who show that investors do not
understand income shifting around one-time tax rate change around 1986. We attribute the different results
between these two research settings to the recurring nature of loss carrybacks and better rm communication
with the market in the 1990s and 2000s. Management guidance was virtually non-existent in the 1980s.
1672 Erickson, Heitzman, and Zhang
The Accounting Review
September 2013
1987 and 1990 in order to generate a refund of taxes paid at the higher rate. In this section, we
explore the cross-sectional variations in rms incentives to carry back tax losses.
While rms have a variety of incentives to carry back losses, we specically test whether the
demand for liquidity drives the rms decision to increase cash ow through a tax refund. This
implicitly assumes that the incremental cost of external funds is higher than the cost of funds
obtained through a tax refund. Empirically, we measure nancial constraints (FC) using a rms
debt minus cash and short-term investment scaled by total assets. We then add FC, measured at
prior year-end, and its interaction term with D_NOL (D_NOL FC) to Equations (1) and (2) as
follows:
E
t
b
0
b
1
D NOL b
2
FC
t1
b
3
D NOL FC
t1
b
4
D NEG b
5
D NEG FC
t1
b
6
E
t1
b
7
E
t2
b
8
LOSS
t1
b
9
LOSS
t2
b
10
LOSS
t1
E
t1
b
11
LOSS
t2
E
t2
b
12
LogMV
t1
b
13
ACC
t1
b
14
RET
t1
e
t
: 3
TABLE 5
Analysis of Stock Returns
Dependent Variable
RET
t
(1)
RET
t
(2)
ARET
t
(3)
ARET
t
(4)
Intercept 0.153 0.277 0.0026 0.0055
(2.07) (3.68) (1.67) (3.54)
FE
t
1.554 0.1109
(10.72) (8.73)
D_NOL 0.003 0.074 0.0014 0.0058
(0.09) (2.17) (0.71) (3.23)
D_NEG 0.018 0.004 0.0086 0.0070
(0.39) (0.09) (6.61) (5.09)
ln(MV
t1
) 0.007 0.023 0.0001 0.0008
(0.99) (3.04) (1.22) (4.10)
BM
t1
0.032 0.063 0.0031 0.0050
(1.60) (3.19) (4.52) (6.67)
RET
t1
0.016 0.064 0.001 0.0029
(0.47) (1.95) (0.24) (6.43)
Adj. R
2
0.043 0.113 0.008 0.045
RET
t
is year ts annual returns starting from eight months prior to scal year-end. ARET
t
is the average three-day earnings
announcement return in year t, measured as raw returns minus value-weighted market returns over the three-day [1, 1]
period, where day 0 is the earnings announcement date. FE
t
is the forecast error for year t, calculated as the difference
between I/B/E/S actual earnings and the median analyst forecast made eight months prior to a rms scal year-end,
scaled by stock price at the forecast date. NOLC is the net operating loss carryback capacity limit (see Appendix B for
variable measurement). D_NOL is a dummy variable with the value of 1 if NOLC is positive for a rm in an expected
loss position, and 0 otherwise, indicating the rms incentives to carry back NOLs and to claim tax refunds. D_NEG is a
dummy variable with the value of 1 if analysts forecasts of year ts earnings are negative, where forecasts were made
eight months prior to scal year-end. MV is a rms market value of equity at the end of year t1. BM is the book-to-
market ratio at the end of year t1. RET
t
is the 12-month buy-and-hold return starting from eight months before a rms
scal year-end. Please see Appendix A for detailed variable denitions. The sample includes 83,875 rm-quarter
observations with non-missing observations of FE and return variables from 1981 to 2010. In all columns, t-statistics in
parentheses are based on the Fama-MacBeth regression approach. FE and BM are winsorized at the 1st and 99th
percentiles.
Tax-Motivated Loss Shifting 1673
The Accounting Review
September 2013
FE
t
b
0
b
1
D NOL b
2
FC
t1
b
3
D NOL FC
t1
b
4
D NEG b
5
D NEG FC
t1
b
6
LogMV
t1
b
7
LogCOV
t
b
8
BM
t1
b
9
ACC
t1
b
10
RET
t1
e
t
: 4
Our main prediction is that the coefcients on the interaction term D_NOL FC should be
signicantly negative in Equations (3) and (4), suggesting that more nancially constrained rms
have stronger loss-shifting incentives. In Table 6, we nd this is indeed the case. In the regression
of future earnings, the coefcient on the interaction term D_NOL FC is 0.105 (t 1.97). In the
regression of analyst forecast errors, the coefcient on the interaction term D_NOL FC is 0.028
(t 2.24). If we further add the prior-period forecast errors as an additional control variable to
Equation (4), then the interaction term D_NOL FC becomes marginally signicant (coeff.
0.018, t 1.65). The results provide some support for the role of liquidity benets because
nancially constrained rms appear more likely to accelerate losses to access cash from tax refunds,
and this behavior results in more negative earnings and analyst forecast errors for those rms.
Analyst Revisions of Future Earnings
Section IV shows that analysts do not fully incorporate tax-related losses in their earnings
forecasts. It is possible that some analysts may learn about such tax-motivated loss shifting over
time, for example when the rm announces earnings or an asset sale, and revise their earnings
forecasts accordingly. If tax-related losses represent one-time attempts to shift income, then a
testable implication is that current-year earnings forecasts should fall, but future earnings forecasts
should be relatively unchanged, suggesting a smaller correlation in analysts forecast revisions
between the current and subsequent years earnings. To test this prediction, we analyze the
following model:
REV
t1
b
0
b
1
D NOL b
2
D NEG b
3
REV
t
b
4
REV
t
D NOL b
5
REV
t
D NEG
b
6
LogMV
t1
b
7
LogCOV
t
b
8
BM
t1
b
9
RET
t1
b
10
ACC
t1
e
t1
;
5
where REV
t
measures analyst forecast revision for year ts earnings from eight months before year
ts scal year-end to one month after, and REV
t1
measures analyst forecast revision for year t1s
earnings during the same period. We expect that on average b
3
will be positive (revisions are
positively correlated), b
5
will be negative (less correlated for negative earnings shocks), and b
4
will
be negative (revisions of t1 forecasts are even less sensitive to current revisions given tax
incentives to report losses).
The results are consistent with our predictions. Untabulated results show a positive coefcient
on REV
t
(b
3
0.558, t 22.50) and a negative coefcient on REV
t
D_NEG (b
5
0.114, t
4.44). More importantly, the coefcient on REV
t
D_NOL is signicantly negative (b
4
0.068,
t 2.05). The results suggest that revisions of future years earnings are less sensitive to revisions
in current-year earnings for rms with tax loss carryback incentives, consistent with the idea that
tax-related losses reect one-time attempts to shift income. In other words, to the extent that
analysts revise their current-year forecasts downward to reect better information about the rms
tax-based incentive to accelerate losses into that year, they are less likely to revise forecasts of
future earnings in the same direction.
Other Robustness Checks
Our main analysis is based on an indicator variable version of NOLC, the capacity for tax loss
carrybacks. In untabulated analysis, we incorporate the magnitude of tax loss carrybacks by
replacing D_NOL with NOLC and re-estimate the regressions. The results continue to hold. For
example, the coefcient on NOLC in Equation (2) is 0.795 (t 3.54), while the coefcient on
1674 Erickson, Heitzman, and Zhang
The Accounting Review
September 2013
TABLE 6
Analysis of Financial Constraints
Panel A: Regression of Earnings (E
t
)
Intercept 0.118
(9.46)
D_NOL 0.060
(2.48)
D_NEG 0.133
(4.47)
FC
t1
0.055
(5.52)
D_NOL FC
t1
0.105
(1.97)
D_NEG FC
t1
0.241
(3.65)
E
t1
0.509
(7.79)
E
t2
0.038
(0.45)
LOSS
t1
0.029
(2.49)
LOSS
t2
0.029
(3.23)
E
t1
LOSS
t1
0.174
(2.43)
E
t2
LOSS
t2
0.133
(1.20)
ln(MV
t1
) 0.016
(10.77)
ACC
t1
0.108
(4.58)
RET
t1
0.079
(5.94)
Adj. R
2
0.259
Panel B: Regression of Analyst Forecast Errors (FE
t
)
Intercept 0.082
(14.21)
D_NOL 0.030
(4.86)
D_NEG 0.014
(2.02)
FC
t1
0.021
(6.40)
D_NOL FC
t1
0.028
(2.24)
D_NEG FC
t1
0.062
(2.85)
(continued on next page)
Tax-Motivated Loss Shifting 1675
The Accounting Review
September 2013
NOLC in Equation (3) is 0.151 (t 2.29). One caveat of the NOLC regressions is that actual
losses may not be linearly correlated with loss carryback capacity.
We use annual observations in the main analysis on the premise that tax reporting is done
annually. As a robustness check, we rerun our analysis based on quarterly observations by
assigning D_NOL in year t1 to four quarterly earnings surprises in year t. A caveat of the quarterly
approach is that four quarterly observations within a rm-year are not independent. As an
alternative, we execute our analyses using only the fourth-quarter observations. In both
specications, the tenor of the results is unchanged (e.g., signicantly negative coefcients on
D_NOL in the regressions of earnings and analyst forecast errors, and positive coefcients on
D_NOL in the return regressions). We also examine how the coefcient pattern varies across
quarters and nd that our results are driven by the third and fourth scal quarters, consistent with
the idea that rms are more likely to shift losses in later quarters as they receive more precise
information about expected tax and accounting prots, loss-shifting opportunities, and liquidity
needs.
VI. CONCLUSION
We identify rms with an incentive to exercise an option to obtain a cash refund of prior tax
payments that would otherwise expire at the end of the year. Over the 19812010 period with
relatively stable statutory tax rates, rms appear to accelerate reported losses to generate cash ows
from tax refunds, and do so to a greater extent when they are nancially constrained. This evidence
suggests that such tax-based incentives have pervasive effects on reporting decisions across time
even when statutory tax rates are stable and thus have broad implications for capital markets
research.
TABLE 6 (continued)
ln(MV
t1
) 0.010
(11.18)
ln(COV
t
) 0.003
(2.70)
BM
t1
0.018
(6.29)
RET
t1
0.029
(7.27)
ACC
t1
0.013
(2.42)
Adj. R
2
0.128
Panels A and B report regression results for earnings (E
t
) and analyst forecast errors (FE
t
), respectively. E is annual
Compustat earnings before extraordinary items scaled by stock price at a rms scal year-end. FE is the difference
between I/B/E/S actual earnings and the median analyst forecast made eight months prior to a rms scal year-end,
scaled by stock price at the forecast date. D_NOL is a dummy variable with the value of 1 if a rm has carryback
incentives, and 0 otherwise. D_NEG is a dummy variable with the value of 1 if analysts forecasts of year ts earnings are
negative, where forecasts were made in the fourth month after scal year-end. FC is nancial constraint measured as a
rms debt minus cash and short-term investment scaled by total assets. LOSS
t1
(LOSS
t2
) is a dummy variable with the
value of 1 if E
t1
(E
t2
) is negative, and 0 otherwise. MV is a rms market value of equity at the end of year t1. ACC is
total accruals scaled by average assets. RET is the 12-month buy-and-hold return starting from eight months before a
rms scal year-end. COV is the number of analysts covering the rm at the forecast date. Please see Appendix A for
detailed variable denitions. The sample includes 85,577 rm-year observations with non-missing earnings variables (E
t
,
E
t1
, and E
t2
) in Panel A and 83,875 rm-year observations with non-missing FE in Panel B from 1981 to 2010.
Earnings variables FE, BM, and ACC are winsorized at the 1st and 99th percentiles, and t-statistics in parentheses are
based on the Fama-MacBeth regression approach.
1676 Erickson, Heitzman, and Zhang
The Accounting Review
September 2013
Our evidence also suggests that analysts do not incorporate these tax-motivated loss-shifting
incentives into their earnings forecasts because we nd that analyst forecast errors are signicantly
more negative for rms with incentives to accelerate tax losses. This evidence contrasts with the
view that analyst forecast errors proxy for earnings surprises, and reinforces the evidence on
analysts disregard for the information in the rms tax disclosures. Despite the relatively larger
unexpected losses for rms with tax incentives to accelerate losses, the stock market reacts in a way
that suggests investors value tax-motivated loss shifting. If we control for the size of the reported
loss, then the stock returns of rms with tax-motivated loss-shifting incentives are signicantly less
negative than similar loss rms without such incentives.
Our evidence is relevant to understanding the growing importance of tax losses on rm value.
Recent innovations in takeover defenses are designed specically to protect the value of tax losses,
and within the last decade, Congress has argued that the temporary liberalization of tax loss
carryback rules injects needed liquidity into the business sector. Taken together, our evidence
suggests that the incentive to carry back tax losses plays a material and persistent role in corporate
reporting decisions and capital markets activities.
REFERENCES
Amir, E., and T. Sougiannis. 1999. Analysts interpretation and investors valuation of tax carryforwards.
Contemporary Accounting Research 16: 133.
Atiase, R. 1987. Market implications of predisclosure information: Size and exchange effects. Journal of
Accounting Research 25: 168176.
Blouin, J., J. Core, and W. Guay. 2010. Have the tax benets of debt been overestimated? Journal of
Financial Economics 98 (2): 195213.
Bradshaw, M., S. Richardson, and R. Sloan. 2001. Do analysts and auditors use information in accruals?
Journal of Accounting Research 39: 4574.
Brown, L., R. Hagerman, P. Grifn, and M. Zmijewski. 1986. Security analyst superiority relative to
univariate time-series models in forecasting quarterly earnings. Journal of Accounting and
Economics 9: 6187.
Chen, K., and M. Schoderbek. 2000. The 1993 tax rate increase and deferred tax adjustments: A test of
functional xation. Journal of Accounting Research 39: 2344.
Dhaliwal, D., C. Gleason, and L. Mills. 2004. Last-chance earnings management: Using the tax expense to
achieve analysts forecasts. Contemporary Accounting Research 21: 431459.
Doyle, J., R. Lundholm, and M. Soliman. 2006. The extreme future stock returns following extreme
earnings surprises. Journal of Accounting Research 44: 849887.
Engel, E., M. Erickson, and E. Maydew. 1999. Debt-equity hybrid securities. Journal of Accounting
Research 37: 249274.
Erickson, M., and S. Heitzman. 2010. NOL poison pills: Selectica v. Versata et al. Tax Notes (June 23).
Francis, J., and D. Philbrick. 1993. Analysts decisions as products of a multi-task environment. Journal of
Accounting Research 31: 216230.
Graham, J. 1996. Proxies for the corporate marginal tax rate. Journal of Financial Economics 42: 187221.
Graham, J., and C. Smith. 1999. Tax incentives to hedge. Journal of Finance 54: 22412262.
Graham, J., and H. Kim. 2009. The effects of the length of the tax-loss carryback period on tax receipts and
corporate marginal tax rates. National Tax Journal 52: 413427.
Guenther, D. 1994. Earnings management in response to corporate tax rate changes: Evidence from the
1986 Tax Reform Act. The Accounting Review 69: 230243.
Hanlon, H., and S. Heitzman. 2010. A review of tax research. Journal of Accounting and Economics 50:
127178.
Hanlon, M. 2003. What can we infer about a rms taxable income from its nancial statements? National
Tax Journal 56: 831863.
Tax-Motivated Loss Shifting 1677
The Accounting Review
September 2013
Hayn, C. 1995. The information content of losses. Journal of Accounting and Economics 20: 125153.
Kothari, S. P. 2001. Capital markets research in accounting. Journal of Accounting and Economics 31: 105
231.
Leone, M. 2010. Your loss is your gain. Available at: http://www.cfo.com/article.cfm/14476845/c_
14477052
Lim, T. 2001. Rationality and analysts forecast bias. The Journal of Finance 56: 369385.
Matsunaga, S., T. Shevlin, and D. Shores. 1992. Disqualifying dispositions of incentive stock options: Tax
benets versus nancial report costs. Journal of Accounting Research 30 (Supplement): 3768.
Maydew, E. 1997. Tax-induced earnings management by rms with net operating losses. Journal of
Accounting Research 35: 8396.
Mills, L., K. Newberry, and G. Novack. 2003. How well do Compustat NOL data identify rms with U.S.
tax return loss carryovers? The Journal of the American Taxation Association 25: 117.
OBrien, P. 1987. Analysts forecasts as earnings expectations. Journal of Accounting and Economics 10
(1): 5383.
Outslay, E., and G. McGill. 2002. Did Enron pay taxes? Using accounting information to decipher tax
status. Tax Notes (August 19).
Plumlee, M. 2003. The effect of information complexity on analysts use of that information. The
Accounting Review 78: 275296.
Richardson, S., S. Teoh, and P. Wysocki. 2004. The walkdown to beatable analyst forecasts: The roles of
equity issuance and insider trading incentives. Contemporary Accounting Research 21: 885924.
Scholes, M., M. Wolfson, M. Erickson, E. Maydew, and T. Shevlin. 2008. Taxes and Business Strategy. 4th
edition. Upper Saddle River, NJ: Prentice Hall.
Scholes, M., P. Wilson, and M. Wolfson. 1992. Firms responses to anticipated reductions in tax rates: The
Tax Reform Act of 1986. Journal of Accounting Research 30 (Supplement): 161185.
Shackelford, D., J. Slemrod, and J. Sallee. 2010. Financial Reporting, Tax, and Real Decisions: Toward a
Unifying Framework. Working paper, University of Michigan.
Shane, P., and T. Stock. 2006. Security analyst and stock market efciency in anticipating tax-motivated
income shifting. The Accounting Review 81: 227250.
Shevlin, T. 1990. Estimating corporate marginal tax rates with asymmetric treatment of gains and losses.
Journal of the American Taxation Association 11: 5167.
Sloan, R. 1996. Do stock prices reect information in accruals and cash ows about future earnings? The
Accounting Review 71: 289315.
Teoh, S. H., and T. J. Wong. 2002. Why new issues and high-accrual rms underperform: The role of
analysts credulity. Review of Financial Studies 15: 869900.
Zhang, F. X. 2006a. Information uncertainty and stock returns. Journal of Finance 61: 105137.
Zhang, F. X. 2006b. Information uncertainty and analyst forecast behavior. Contemporary Accounting
Research 23: 565590.
1678 Erickson, Heitzman, and Zhang
The Accounting Review
September 2013
APPENDIX A
Variable Denitions
E
t
E is annual Compustat earnings before extraordinary items per share scaled by stock price
at a rms scal year-end;
FE
t
analyst forecast error, dened as (E F)/P
t1
, where E is reported annual earnings per
share from I/B/E/S, F is the median analysts earnings-per-share forecast made eight
months prior to scal year-end, and P
t1
is stock price at the forecast date;
NOLC
t1
net operating loss carryback capacity (see the estimation process in Appendix B for
details), indicating a rms tax paid in the earliest year of the NOL carryback window
that has not been refunded yet;
D_NEG a dummy variable with the value of 1 if analysts forecasts of year ts earnings are
negative, where forecasts were made eight months prior to scal year-end;
D_NOL a dummy variable with the value of 1 if the rm has an expected loss (D_NEG 1) and
the net operating loss carryback capacity is positive (NOLC . 0), and 0 otherwise;
MV
t1
the market value of equity at the end of year t1;
COV
t
the number of analysts following the company at the forecast date (eight months prior to
scal year-end);
BM
t1
the book-to-market ratio, measured as the book value of equity divided by its market
value, at the end of year t1;
RET
t
the 12-month buy-and-hold return starting from eight months before a rms scal year-
end;
ARET
t
average earnings announcement return of four quarterly earnings announcements for year t.
Quarterly earnings announcement return is dened as raw returns minus value-weighted
market returns over the three-day [1, 1] period, where day 0 is the earnings
announcement date;
ACC
t1
total accruals in year t1, measured as (DCA DCash) (DCL DSTD DTP)
DEPEXP scaled by average total assets, where DCA change in current assets, DCash
change in cash and cash equivalents, DCL change in current liabilities, DSTD
change in debt in current liabilities, and DEPEXP depreciation and amortization
expense;
FC
t1
nancial constraint, measured as total debt minus cash and short-term investment scaled by
total assets, at the end of year t1;
LEV
t1
the leverage ratio, measured as total debt divided by total assets, at the end of year t1;
and
EBITDA
t1
the ratio of earnings before interest, taxes, depreciation, and amortization scaled by average
total assets, measured in year t1.
Tax-Motivated Loss Shifting 1679
The Accounting Review
September 2013
APPENDIX B
Variable Measurement
Timeline of the Variables (using December Fiscal Year-End as an Example)
Estimating Net Operating Loss Carryback Capacity (NOLC)
NOLC is meant to capture tax paid in the earliest year of the NOL carryback window that has
not been refunded by the beginning of year t. We estimate net operating loss carryback capacity
(NOLC) depending on the time period of the loss. Before 1997, rms could carry back NOLs up to
three years and carry forward NOLs up to 15 years. Since 1997, rms can carry back NOLs up to
two years and carry forward NOLs up to 20 years.
The Post-1997 Period (Fiscal Year-End of August 1998 and Later)
Firms can carry back NOLs up to two years:
Our main test is whether rms have incentives to report larger losses in year t based on whether
the rm can carry back losses against income in the earliest carryback year (t2). We dene taxable
income (TI) as current tax expense divided by the top statutory tax rate. We calculate NOLC as
follows:
NOLC MAX 0; TI
t2
MIN0; TI
t1
MAX0; TI
t3
f g: A1
In words, A1 denes the capacity to carry back losses (NOLC) in year t as the amount of taxable
prots in t2 that has not been used to claim a refund from tax losses in year t1. If the rm did not
generate a taxable prot in t2 (TI
t2
0), then there is no tax available to refund, and the outside
maximization function ensures that NOLC will simply equal 0. But if the rm has taxable prots in
t2 (TI
t2
. 0), then we must rst determine how much of those taxable prots in t2 were offset
through tax losses in year t1. This adjustment is given by MIN[0,TI
t1
MAX(0,TI
t3
)] in
Equation (A1). If the rm has a taxable prot in year t1, then this adjustment equals 0 and the full
amount of t2 taxes paid is available for refund. If the rm had taxable losses in year t1 (TI
t1
,
0), then we must ask whether the rm had taxable income in t3. A loss in year t1 will rst be
carried back against prots in year t3, and the prots available to be offset by year t1 losses are
presented by MAX(0,TI
t3
). If the year t1 loss exceeds the year t3 income (if any), then the
1680 Erickson, Heitzman, and Zhang
The Accounting Review
September 2013
excess loss is taken against t2 income, and this is the adjustment given by MIN[0,TI
t1
MAX(0,TI
t3
)].
To illustrate the potential values of NOLC as a function of prior taxable prot and loss patterns,
consider the cases in the following table:
Case
Taxable
Prots/Losses
in Earliest
Carryback Year
TI
t2
Taxable
Prot/Loss
in Year t1
TI
t1
Taxable Prots
in Year t3
Available to Offset
Year t1 Losses
MAX(0,TI
t3
)
Amount of Year
t1 Loss Not Covered
by Prots in t3
MIN[0,TI
t1
MAX(0,TI
t3
)]
Max. Amount
of Loss in Year
t that can be
Carried Back
to Obtain Refund
of t2 Taxes
NOLC
(1) (2) (3) (4) (5)
1. 10 0 20 0 10
2. 5 5 10 0 5
3. 10 10 10 0 10
4. 10 20 10 10 0
5. 10 15 10 5 5
6. 10 0 0 0 0
Cases 1 through 5 represent rms with taxes paid on prots in the earliest carryback year. Cases 1
and 2 represent rms with no tax loss in t1 (and hence no offset of t2 prots). Case 3 provides an
example of a rm whose tax loss in year t1 was offset entirely against prots in year t3,
preserving the entire capacity of t2 prots to offset losses in t. Case 4 represents the case where a
loss in t consumed both t3 and t2 prots, while in Case 5, only a portion of the prots in t2 are
used up to offset losses in t1. The remainder in Case 5 is the amount available for use against
losses in t. The nal case simply illustrates that rms that paid no taxes in t2 (or reported a loss)
will have no capacity for carryback of t losses regardless of what happens in year t1 or t3.
The Pre-1997 Period
Firms can carry back NOLs up to three years.
Here, we calculate NOLC as follows:
NOLC MAX
0; TI
t3
MIN
h
0; TI
t2
MAX0; TI
t5
MAX0; TI
t4
i
MIN
h
0; TI
t1
MAX
0; TI
t4
MIN
0; TI
t2
MAX0; TI
t5
: A2
Due to the dynamics involved with three carryback years, the formula is more complex. But the
intuition and results are similar as in A1. Numerical examples using this formula are available from
the authors upon request, but can be easily replicated.
Tax-Motivated Loss Shifting 1681
The Accounting Review
September 2013
D_NOL
D_NOL is a dummy variable that takes a value of 1 if NOLC is positive for a rm in an
expected loss position, and 0 otherwise. Firms with expected losses are those with negative
analysts consensus (median) forecasts made eight months prior to scal year-end (D_NEG 1).
Our NOLC (and hence D_NOL) measure is subject to a potential sources of measurement error.
First, taxable income includes both operating income/loss and capital gain/loss, which have
different carryback and carryforward rules. We do not make such a distinction because the
carryback periods are largely similar and because we can only estimate taxable income. Second,
measurement errors arise because the average tax rate is not equal to the top statutory tax rate owing
to progressive tax schedules. Third, current tax expense does not reect the tax benets from the
exercise of non-qualied employee stock options (reected as a reduction in current tax liability)
and is reported net of tax cushions and tax credits (Hanlon 2003). Fourth, we do not incorporate the
magnitude of expected losses related to previous years prots, as analysts earnings forecasts are a
noisy proxy for a rms tax prots/losses. Incorporating the magnitude of expected losses would
make our NOLC formula too complicated. Fifth, rms were allowed to carryback NOLs for ve
years in 2001/2002 and 2008/2009. We do not incorporate the ve-year carry backs in our measure,
as the ve-year carryback period was often not known until a later date. In addition, a ve-year
carryback would make our measure extremely complicated. Finally, we do not incorporate the
information of net operating loss carryforwards on the balance sheet because of jurisdictional
differences and the impact of NOLs acquired in mergers and acquisitions. Moreover, research by
Mills et al. (2003) raises a number of additional concerns using Compustat data to infer net
operating loss positions. We hand-checked a number of cases and nd that many rms were able to
claim tax refunds when reporting non-missing NOL carryforwards in their nancial statements.
While we can certainly rene our NOLC measure along the directions mentioned above,
incorporating such issues would make our measure unduly complicated. We believe that such
measurement errors introduce noise to our measure and, if anything, are likely to bias against
nding any signicant results. With that in mind, we prefer a relatively simple measure as used in
the paper.
Alternative Specications of NOLC and D_NOL
We consider the following alternative specications of NOLC and D_NOL. In the rst
alternative specication, we replace expected losses proxied by analysts forecasts with those based
on a time-series earnings models in which we regress current years earnings (from Compustat) on
last years earnings and earnings in the year before. Then we use coefcient estimates based on
historical data and the most recent set of earnings data to calculate forecasted earnings for the next
year. D_NOL is equal to 1 if NOLC is positive and forecasted earnings are negative. In the second
alternative specication, we drop the requirement of expected losses. Rather, we require taxable
income to be positive in year t2 and to be negative in year t1 in the post-1997 period. In the pre-
1997 period, we require taxable income to be positive in year t3 and to be negative in year t2 and
t1. In both cases, we require that tax paid in the rst year of the NOL carryback window has not
been fully refunded. We nd robust empirical results across these two alternative specications
(results not tabulated).
1682 Erickson, Heitzman, and Zhang
The Accounting Review
September 2013