1. Contribution incentives and the TCJA

A DB pension plan is a promise of predictable retirement benefits from a plan sponsor (typically an employer) to participants (employees). Plans are funded by employer and employee contributions. In this section we review the main factors underpinning these transfers, and we discuss how the TCJA created tax-based incentives for sponsors to increase contributions.

Since corporate DB plans are subject to funding rules under U.S. law, the size of employer contributions depends on the funding status of the plan.Footnote ⁶ If a plan is overfunded, its sponsor has to contribute the present value of the expected yearly change in accrued benefits (normal or service cost), net of excess assets. Sponsors of overfunded plans have little incentive to contribute more than required, as the fiscal regime penalises them for drawing down plan assets net of liabilities.Footnote ⁷ Sponsors of underfunded plans, by contrast, are required by law to contribute more than the service cost. The Pension Protection Act of 2006 stipulates that plan funding should equal 100 per cent of the plan's liabilities. As a result, minimum required contributions (MRCs) are typically set according to rules which prescribe that sponsors contribute the service cost plus a fraction of the funding shortfall (shortfall amortisation charge). MRC schedules are intended to close funding deficits over a medium-term horizon. Sponsors of underfunded plans might also choose to improve funding status by making voluntary contributions in excess of MRCs. Firms subject to federal corporate income taxation (C-corporations) can deduct pension contributions from tax returns. As a result, there are tax-based incentives for sponsors to contribute more than minimum requirements. Section 404 of the Internal Revenue Code (IRC) specifies that contributions made in a particular year can be deducted from previous-year income under two conditions. First, the contribution has to be made on account of pension benefits accrued in the previous year. Second, the contribution has to be made by the employer's tax return due date, including extensions. Concretely, a firm whose fiscal year ends in December (called a calendar-year firm) has until mid-October of the current year to make contributions that are deductible from the previous-year tax return.Footnote ⁸ In practice, if the firm's ‘plan year’ (the 12-month period relevant for plan reporting) also ends in December, the firm would want to make contributions before mid-September. This is because contributions made after this date would not count towards satisfying minimum funding requirements under Section 430 of the IRC.Footnote ⁹ There are limits to deductibility: contributions are only allowed to be tax-deductible up to the point where a plan is 150 per cent funded.

The TCJA made plan contributions counted towards 2017 sponsor income more valuable than contributions counted towards 2018 income. The Act permanently reduced the statutory federal corporate tax rate from 35 to 21 per cent, beginning in 2018. As a result, sponsor contributions made by calendar-year firms within the grace period between January 2017 and mid-September 2018 could be deducted from 2017 income and thus reduce the corporate tax bill at the old, higher tax rate. By contrast, contributions made after mid-September were deducted at a lower rate. As an example, a late-filing sponsor contributing $1bn to its DB pension plan before mid-September 2018 – rather than after the end of the grace period (e.g., December 2018) – would have saved an extra $140m in 2017 taxes. In this sense, the TCJA included a temporary tax break on pension contributions.

Sponsors have other incentives to shore up underfunded pension plans, with rising benefit insurance premia being an oft-mentioned driver by industry commentary (Figure 1).Footnote ¹⁰ The retirement benefits of private sector workers are guaranteed (up to a limit) by the PBGC, a government agency established in the mid-1970s to protect plan beneficiaries in case of sponsor bankruptcy. In addition to a flat-rate premium which applies to all plans, there is a variable-rate premium which applies only to underfunded plans. Variable rate premia grow with plan deficit, so employers have incentives to make voluntary contributions in order to reduce insurance costs. Sufficiently overfunded firms are exempt from paying premia altogether.

Figure 1.

Aggregate contributions and the PBGC variable premium.

Sponsors may also worry about the impact of unfunded pension liabilities on their cost of capital and valuations, particularly if bankruptcy risk is already a concern (Black, Reference Black1980). Since 2006, financial accounting standards require plan sponsors to ‘flow through’ pension fund deficits into their financial statements, meaning that employers must recognise a plan's funded status on their balance sheets (FAS 158). And credit rating agencies took pension liabilities into account even prior the change in reporting standards, when the funded status of plans was disclosed in financial statement footnotes (Clifton et al., Reference Clifton, Oman, Mulvaney and Philippakos2003; Mathur et al., Reference Mathur, Jonas and LaMonte2006; Campbell et al., Reference Campbell, Dhaliwal and Schwartz2012). As a result, unfunded pension liabilities can have material effects on sponsor cost of capital and equity valuations.Footnote ¹¹

That said, there are opportunity costs to diverting firm resources to pension plans through contributions. In the presence of financing frictions, a reduction in internal financial resources may limit a sponsor's ability to finance investment projects. Indeed, Campbell et al. (Reference Campbell, Dhaliwal and Schwartz2012) show that an increase in mandatory pension contributions – which reduces a firm's ability to rely on internal financing for investment projects – increases the cost of capital for firms facing greater constraints on external financing, a result consistent with earlier evidence of a negative relationship between contributions and firm investment (Rauh, Reference Rauh2006).

Sponsor contributions started rising before the TCJA (Figure 1), an increase which would likely have continued through 2017 even in the absence of tax-based incentives. Industry commentary tends to attribute this rebound to the sharp increase in the PBGC insurance premium.Footnote ¹²

2. Conceptual framework

This section sketches a model of contributions. The framework is designed to formalise the tradeoffs faced by a sponsor in choosing contributions (see Section 1), and to derive testable predictions about the impact of the TCJA on optimal contributions. It embeds the idea that sponsors dislike plan funding deficits because they worsen the balance sheet, thereby increasing the costs of finance. As we focus on the TCJA impact on contributions, we do not model the sponsor's investment decision endogenously.Footnote ¹³ However, we do model the opportunity cost of diverting internal resources away from investment by letting investment returns affect the tradeoff that underpins optimal contributions.

The data only allow us to test the impact of TCJA on sponsor contributions deducted from 2017 and 2018 tax returns. Contributions counted towards 2019 tax returns are affected by the CARES Act, which gave DB sponsors the option to wait until January 2021 to make contributions deductible from 2019 returns. We thus assume there are three periods, t = 0, 1, 2. In periods t = 0, 1 (corresponding to 2017 and 2018), the management of a firm sponsoring a DB pension fund chooses plan contributions, c _t, to maximise the value of the firm. We capture the impact of the CARES Act of 2019 on contributions by assuming that sponsors did exercise the option to wait, and we let c ₂ = 0.Footnote ¹⁴

The pension plan funding status affects insurance costs, contribution requirements, and the costs of external finance. Plan funding depends positively on contributions and negatively on the service cost, s _t. The service cost is determined by previous decisions about wages and by factors outside of management's control (e.g., interest rates), and it is therefore exogenous to current contribution and investment decisions. Letting z _t represent plan surplus – the difference between plan assets and plan liabilities – the law of motion of the funding status is

(1)$$z_{t + 1} = z_t + c_t-s_t + \omega _t, \;$$

where ω _t is a catch-all random variable capturing all uncertainty about pension assets and liabilities (e.g., uncertainty about investment returns). The sponsor does not observe ω _t before choosing contributions. The funding shock ω _t is i.i.d. over time, with bounded support $\omega _t\in [ {\underline{\omega } , \;\bar{\omega }} ]$.

If its pension plan is underfunded, the firm has to pay the variable PBGC insurance premium. The insurance premium, q(z _t), is piece-wise linear:

(2)$$q( {z_t} ) \equiv {\rm max}[ {0, \;-\bar{q}z_t} ] , \;\quad \bar{q}\in ( {0, \;1} ) , \;$$

with derivative ${q}^{\prime}( {z_t} ) = {-}\bar{q}$ if z _t < 0 and 0 otherwise. In addition, a sponsor with an underfunded pension plans must contribute more. Regulatory requirements mandate that contributions have to be at least as high as the service cost:

(3)$$c_t{\kern 1pt} {\rm \geqslant }{\kern 1pt} {\kern 1pt} {\rm max}[ {s_t, \;s_t-z_t} ] \equiv {\rm \Psi }( {z_t} ) , \;$$

so an underfunded sponsor with z _t < 0 has to contribute more that the service cost. The lower bound on contributions, Ψ(z _t), is piece-wise linear with derivative equal to Ψ′(z _t) = −1 if z _t < 0 and 0 otherwise.

If a plan is underfunded, the sponsor suffers a loss. We think of the loss as a reduced form representation of the costs of obtaining external finance. Rather than modelling external finance costs endogenously, we follow Gomes (Reference Gomes2001) and Whited (Reference Whited2006) and assume that when contributions are large relative to the sponsor's internal resources, the firm can only go ahead if it obtains external funds at a premium. External finance costs depend on plan funding and cash flows, x _t. In addition, the plan sponsor faces some uncertainty about the costs of finance. Concretely, we define external finance costs as

(4)$$R( {x_t, \;z_t} ) = r_0-r_{x, t}x_t + r_z\displaystyle{{z_t^2 } \over 2}1_{z_t < 0}\quad {\rm if\;}x_t < 0$$

and 0 otherwise, with r ₀, r _z > 0 represents the minimum external finance costs that the sponsor can obtain external funds at. The linear term in (4) implies that a larger external finance need (a more negative x _t) makes external finance more expensive. The quadratic term denotes the underfunding penalty. Conditional on there being an external finance need, a larger plan funding shortfall (a more negative z _t) increases the cost of external finance. We thus let R _x(x _t, z _t) = −r _x,t if x _t < 0 and 0 otherwise; and R _z(x _t, z _t) = z _t < 0 if x _t, z _t < 0, and 0 otherwise (see the Appendix A).

The external finance cost function parameter r _x,t represents the sensitivity of external finance costs to changes in financing needs. We can thus interpret it as the shadow value of relaxing an external finance constraint, with a larger r _x,t implying that the sponsor is more constrained. In Section 1 we argued that in the presence of financing frictions, a reduction in internal financial resources may limit a sponsor's ability to finance investment projects. In order to allow for this possibility we assume that r _x,t = r(a _t), and we interpret the shock a _t as the productivity of investment/investment returns. A higher a _t realisation is associated with a higher r _x,t realisation. The higher investment returns, the more valuable relaxing the external finance constraint (equivalently, the tighter the constraint). Formally, we assume that r _x,t is i.i.d. and independent of ω _t, for all t, and that a _t is a mean-preserving shock, so the mean of r _x,t is constant.Footnote ¹⁵

The sponsor chooses contributions to maximise the value of the firm, which is equal to the present discounted value of expected cash flows. Cash flows are given by

(5)$$x_t = y_t-( {1-\tau_t} ) ( {c_t + q( {z_t} ) } ) , \;$$

where y _t denotes the component of cash flows that does not depend on contributions. We take it as exogenous but dependent on the shock y _t = f(a _t). The higher a _t, the larger y _t. τ _t is the corporate tax rate. Letting Λ^j denote the (constant) discount factor applied to cash flows received in period t + j, for some Λ ∈ (0, 1), the firm's problem is thus given by

(6)$$\mathop {\max }\limits_{_{\{ {c_0, c_1} \} } } \quad {\rm {\opf E}}_t\left[{\sum\limits_{\,j = 0}^2 {\Lambda^j} ( {x_{t + j}-R( {x_{t + j}, \;z_{t + j}} ) } ) } \right]{\rm , \;}$$

subject to the law of motion of plan funding (1), the expression for the variable insurance premium (2), the regulatory requirement on contributions (3), the definition of external finance costs (4), the expression for cash flows (5), and c ₂ = 0. A typical specification for the discount factor would take β ^ju ^′(C _t+j)/u ^′(C _t) for β ∈ (0, 1), where u ^′(C _t) is the marginal utility of consumption of a representative household at date t + 1. By assuming a constant discount factor we are implicitly assuming linear utility.

The first-order condition for contributions illustrates the intertemporal tradeoff faced by the sponsor:

(7)$$\eqalign{( {1-R_x( {x_t, \;z_t} ) } ) ( {1-\tau_t} ) = & \lambda _t + {\rm {\opf E}}_t\{ {\Lambda [ {-( {1-\tau_{t + 1}} ) q^{\prime}( {z_{t + 1}} ) } ] ( {1-R_x( {x_{t + 1}, \;z_{t + 1}} ) } ) } \cr & { -\lambda_{t + 1}\Psi^{\prime}( {z_{t + 1}} ) -\Lambda R_z( {x_{t + 1}, \;z_{t + 1}} ) } \} .} $$

Here, λ _t denotes the Lagrange multiplier on the period-t regulatory requirement on contributions (3). The left-hand side of this equation is the marginal cost of current contributions. Higher contributions today lower current cash flows (but less than one for one, thanks to the contributions tax shield) and possibly raise the cost of external finance, ${-}R_x( {x_t, \;z_t} ) {\kern 1pt} {\rm \geqslant }{\kern 1pt} {\kern 1pt} 0$. The right-hand side represents the marginal benefit of contributing today. Higher current contributions relax the current regulatory constraint ($\lambda _t{\kern 1pt} {\rm \geqslant }{\kern 1pt} {\kern 1pt} 0$). They also increase next-period plan surplus. In turn, this raises next-period cash flows by lowering the PBGC premium ($-{q}^{\prime}( {z_{t + 1}} ) {\kern 1pt} {\rm \geqslant }{\kern 1pt} {\kern 1pt} 0$), and reduces the need to rely on external finance and the corresponding costs. At the same time, a higher future surplus slackens the regulatory constraint by lowering the MRC ($-\lambda _{t + 1}{\rm {\Psi }^{\prime}}{\kern 1pt} {\rm \geqslant }{\kern 1pt} {\kern 1pt} 0$). In addition, a higher future surplus reduces the costs of future external finance, $-\Lambda _{t + 1}R_z( {x_{t + 1}, \;z_{t + 1}} ) {\kern 1pt} {\rm \geqslant }{\kern 1pt} {\kern 1pt} 0$.

Modelling the impact of TCJA. We now derive a prediction for the impact of the TCJA on the time profile of contributions of underfunded sponsors. To that end, we introduce a distinction between the tax rate at which contributions can be deducted, $\tau _t^c$, and the corporate tax rate, τ _t. As a result, the tax rate entering the left-hand side of the first-order condition (7) need not be the same as the tax rate entering the right-hand side. We assume that until period t − 1, both tax rates are equal and constant at the level τ. In period t, it is announced that the corporate tax rate will decline from then on, τ _t+j = τ(1 − Δ)for all $j{\kern 1pt} {\rm \geqslant }{\kern 1pt} {\kern 1pt} 0$, with Δ ∈ (0, 1). By contrast, the tax rate relevant for contributions stays at the old level in period t, before dropping down in all subsequent period, $\tau _t^c = \tau$and $\tau _{t + j}^c = \tau ( {1-{\rm \Delta }} )$ for all $j{\kern 1pt} {\rm \geqslant }{\kern 1pt} 1$.

TCJA causes a steepening in the time profile of underfunded sponsor contributions, with c _t increasing and c _t+1 decreasing. Consider an underfunded sponsor that expects to continue to be underfunded in the immediate aftermath of the TCJA. The firm's pension plan is underfunded in t, t + 1, and t + 2. As a result, the firm contributes more than the minimum requirement, λ _t+j = 0 for j = 0, 1, 2. The firm relies on external finance, x _t+j < 0 for j = 0, 1, 2.Footnote ¹⁶ A higher Δ increases the marginal benefit of contributing today (see (7)). Intuitively, given lower expected corporate taxes tomorrow (t + 1), an extra dollar put to work to reduce plan expenses (by lowering PBGC payments) has a bigger impact on tomorrow's cash flows and finance costs. As a result, current contributions, c _t, increase in Δ.

Future contributions, c _t+1, are different because in addition to lower expected taxes tomorrow, sponsors also face lower taxes today. As Δ rises and the value of the contributions tax shield falls, the marginal cost of c _t+1 rises. Higher Δ raises both the marginal costs and the marginal benefit of future contributions, c _t+1. At the same time, higher c _t contributions have already boosted plan fundingz _t+1, in turn decreasing the expected funding gap, ${\open E}$_t+1[z _t+2]. And the smaller the funding gap, the less responsive the costs of external finance to further funding improvements brought about by additional t + 1 contributions. As a result, the TCJA has an indirect negative effect on the marginal benefit of contributions, which dampens the direct (positive) effect. The marginal cost channel then dominates and c _t+1 falls. We formalise in the claim below.

The overall impact of the TCJA on sponsor contributions depends on the time profile of the sponsor's financial constraint. If a sponsor is more financially constrained in period t + 1 than it expects to be in period t + 2, then t + 1 contributions decrease by more than t contributions increase (reversal). To see why this is the case, consider the first-order conditions for t + 1 contributions:

(8)$$( {1-\tau ( 1-\Delta ) } ) ( {1 + r_{x, t + 1}} ) = \Lambda ( {1 + {\rm {\opf E}}\% _{t + 1}[ {r_{x, t + 2}} ] } ) ( {1-\tau ( {1-\Delta } ) } ) \bar{q}-\Lambda {\rm {\opf E}}_{t + 1}[ {z_{t + 2}} ] .$$

The marginal cost of contributions (the left-hand side of (8)) increases in Δ. As the corporate tax rate decreases, the negative impact of higher contributions on current after-tax cash flows x _t rises. And the more constrained the sponsor, the larger the negative impact on current free cash flows, x _t − R(x _t, z _t). The marginal benefit of contributions (the right-hand side of (8)) has two terms. The first one increases in Δ, because a lower corporate tax rate raises the positive impact of lower PBGC payments on after-tax cash flows. Like the marginal cost, this term is also larger for a more constrained sponsor, because an increase in after-tax cash flows x _t+1 has a larger impact on future free cash flows, x _t+1 − R(x _t+1, z _t+1). The second term, −ΛR _z = Λ${\open E}$_t+1[z _t+2], depends on the corporate tax rate only through plan funding, z _t+2. By (1), z _t+2 is linear in the sum of past contributions, c _t + c _t+1. Using (1) and (8) we thus obtain:

$$\displaystyle{{d( c_t + c_{t + 1}) } \over {d\Delta }} = \displaystyle{{-( {1 + r_{x, t + 1}} ) \tau + \Lambda ( {1 + {\rm {\opf E}}\% _{t + 1}[ {r_{x, t + 2}} ] } ) \tau \bar{q}} \over \Lambda }, \;$$

which is negative (reversal) if the sensitivity of the marginal cost of contributions with respect to Δ, (1 + r _x,t+1)τ, is larger than the sensitivity of the marginal benefit, $\Lambda ( {1 + {\rm {\opf E}}_{t + 1}[ {r_{x, t + 2}} ] } ) \tau \bar{q}$.Footnote ¹⁷ As both the marginal cost and marginal benefit of contributions are more sensitive to the corporate tax rate when the sponsor is more constrained, but the former accrues earlier than the latter, reversal happens when r _x,t+1 is sufficiently large relative to ${\open E}$_t+1[r _x,t+2].

We next turn to taking the predictions formalised in claims 1 and 2 to the data.

3. Data and construction of variables

Our plan-sponsor level data come from Schedules SB and H of the electronic IRS 5500 filings from the Department of Labor. All employers sponsoring funds with more than 100 employees must file Schedules SB and H of the IRS 5500 Form on an annual basis.Footnote ^{18,Footnote 19} We match the plans with Compustat employers to obtain sponsor-level information.

An alternative source of pension data for Compustat firms is annual 10-K forms filed with the SEC. Unlike IRS data, SEC filings data do not include MRCs, making it harder to disentangle the voluntary component of contributions from the mandatory. In addition, pension variables obtained from SEC filings (contributions, plan assets, and liabilities) do not distinguish between domestic plans and plans pertaining to foreign subsidiaries. By contrast, the TCJA tax break only applies to contributions made to domestic plans. Similarly, the PBGC premium only applies to funding shortfalls of domestic plans.

3.2 Outcome variables

We study the impact of the TCJA tax break on voluntary sponsor contributions, total sponsor contributions, and plan funding.

3.2.1 Contributions

As discussed in Section 1, tax-based incentives affect only the voluntary component of sponsor contributions. We compute voluntary contributions by a particular sponsor to a particular plan by subtracting mandatory contributions from total contributions.Footnote ²³ We define the mandatory component of pension contributions as the sum of MRCs (both legacy and current) and of special contributions made to avoid restrictions on the timing of benefits payment for underfunded plans.Footnote ²⁴ As a firm may sponsor multiple plans, we aggregate over all the plans sponsored by the same firm to obtain sponsor-level contributions (Voluntary Contributions and Total Contributions, respectively).Footnote ²⁵

As larger firms naturally tend to contribute more (e.g., because they have higher service costs), we scale both our contribution variables – voluntary and total – by sponsor size, captured by sponsor assets at the beginning of the current year (Assets (book)). Normalising by firm assets is standard in papers studying either pension contributions from the sponsor's perspective (e.g., Rauh, Reference Rauh2006) or the impact of tax-based incentives on other firm choices, such as capital expenditures (e.g., Zwick and Mahon, Reference Zwick and Mahon2017; Xu and Zwick, Reference Xu and Zwick2018).Footnote ²⁶

Voluntary contributions represent 0.02 per cent of sponsor assets at the mean and 0.002 per cent of assets at the median (Table 1). Total pension contributions are 0.31 per cent of assets at the mean and 0.05 per cent at the median. Both total and voluntary contributions grew in 2016 and in 2017. They declined sharply in 2018 (Figure 2, left panel).

Table 1.

Summary statistics

Notes: Panel (a) presents plan-level and sponsor-level summary statistics for our sample. There are 4,105 plan-year observations and 2,506 firm-year observations during the period 2014–18 (some firms sponsor multiple plans). All plans in the sample are middle- and large-scale plans covering more than 100 employees. Plan-level data are from IRS 5500 filings. Sponsor-level data are from Compustat. Voluntary and Total contributions, PBGC premium, CAPEX, Payout, and Cash flows are scaled by beginning-of-year sponsor balance sheet assets. Voluntary contributions, Total contributions, and PBGC premium are winsorised at the top 1% level. Funding ratio, Cash flows, Tobin's Q, Altman's Z, CAPEX, Payout, Return on investment, Discount rate, and DB pension plans significance are winsorised at the top and bottom 1% level. Panel (b) shows time variation in DB pension plans significance, Tax exposure, and PBGC premium. Columns (1) and (2) show the number of retirement plans and sponsoring firms in each sample year. Columns (3) and (4) break the sample down by tax-exposure. Column (5) shows the Pension Benefit Guarantee Corporation variable premium rates. Rates are quoted per $1,000 of unfunded vested benefits for single-employer plans.

Figure 2.

Aggregate contributions and funding.

3.2.2 Funding ratios

To assess whether the TCJA tax break had an impact on plan funding, we consider the change in funding ratios between 2016 and 2017. We chose this period because contributions made up until the end of the contribution ‘grace period’ (in theory, mid-October 2018; in practice, mid-September 2018), are counted towards 2017 contributions for financial reporting purposes and thus flow into 2017 assets and funding. We compute the funding ratio for a particular plan-sponsoring firm in any given year (Funding Ratio) in a few steps. First, for each plan sponsored by a particular firm, we sum of reported plan assets (Assets) and sponsor contributions (Total Contributions) net of credit balances (Credit Balances).Footnote ²⁷ Reported plan assets are measured at year-end market value, and they do not include contributions. Second, we aggregate the resulting plan-level asset measure over all the plans sponsored by the firm, and we thus obtain the funding ratio numerator. To get the denominator, we sum plan-level liabilities (Liabilities) over all the plans sported by the firm. Liabilities are the present discounted value of future pension benefits accumulated to year-end. MAP-21 allows sponsors to discount plan liabilities using an average of market rates on corporate bonds over the past 25 years.Footnote ²⁸ Plans in our sample are 107.6 per cent funded at the mean and 104.4 per cent funded at the median, with a standard deviation of 16 per cent.

3.3 Explanatory variables

3.3.1 Tax-based incentives

For the TCJA tax break to affect voluntary contributions, two conditions need to be satisfied. First, the firm has to have a positive corporate income tax bill before deducting contributions (tax-paying sponsor).Footnote ²⁹ Second, the funding ratio has to be below the 150 per cent bound above which contributions stop being deductible (funding ratio below 150%), for at least one of the sponsored plans.

We say that a sponsor is exposed to tax-based incentives – including the TCJA tax break – if it meets both these conditions. We define the tax exposure of sponsor s at time t (Tax Exposure) as a dummy variable which is equal to 1 if Gross Tax >0 and if Funding Ratio <150% for at least one plan i of sponsor s. Here, Gross Tax denotes the Federal corporate tax bill of sponsor s before deducting pension contributions. Since we do not observe Gross Tax, we obtain it by adding back the contribution deduction to the corporate tax bill from Compustat. Concretely, Gross Tax = Net Tax + τ ×sum of Total Contributions over sponsored plans, where Net Tax is the Federal corporate income tax expense from Compustat and τ is the statutory corporate income tax rate.

By using Tax Exposure as a proxy for sponsor exposure to the TCJA tax break, we assume that the gross corporate tax bill (Gross Tax) is exogenous to the sponsor's contribution decision. This assumption is justified by the timing of tax-based incentives for retirement plan contributions, which suggests that a sponsor is likely to take the pre-contribution tax bill as given when choosing how much to (voluntarily) transfer to its pension plans. To account for any possible endogeneity and as a robustness check, we will rerun the analysis by relaxing the definition of tax exposure of sponsor s at time t (Tax Exposure) as a dummy variable which is equal to 1 only if Funding Ratio <150% for at least one plan i of sponsor s.

Tax-based incentives for sponsors to contribute could be captured by other proxies. These include estimates of corporate marginal tax rates (Graham, Reference Graham1996a, Reference Graham1996b) and measures of tax exposure based on sponsor tax credits such as net operating loss carryforwards and investment tax credits.Footnote ³⁰ According to the latter set of proxies, a firm is not exposed to tax-based incentives if its accumulated tax credits are large enough to cause it not to report any taxable income. Our tax-based incentives measure is positively correlated with marginal corporate tax rates, and negatively correlated a set of dummies capturing lack of exposure due to tax credits (see Table 2).

Table 2.

Tax exposure and other proxies for tax-based incentives

Notes: This table presents pooled probit regression estimates of the impact of different proxies for tax-based incentives to make pension contributions on the probability that a sponsor is tax exposed. The dependent variable is Tax exposure, a dummy variable = 1 if (i) a firm has a positive ex-contributions tax bill and (ii) the firm sponsors at least one plan with funding ratio below 150%. Marginal tax rate before interest is a simulated corporate marginal tax rate based on income before interest expense has been deducted. Marginal tax rate after interest is a simulated corporate marginal tax rate based on income after interest expense has been deducted. See https://faculty.fuqua.duke.edu/~jgraham/read.html for more details. Net operating loss (NOL) and Investment tax credit are scaled by beginning-of-year sponsor balance sheet assets. NOL dummy 1 is a dummy variable which is = 1 if a sponsor has a positive carryforward balance and it pays no current U.S. income tax. NOL dummy 2 is a dummy variable which is = 1 if a sponsor has a positive carryforward balance and it reports no pre-tax income. NOL dummy 3 is a dummy variable which is = 1 if a sponsor does not report any pre-tax income. Z-statistics obtained using robust standard errors are in parentheses.

There are disadvantages to using corporate marginal tax rate estimates or exposure measures based on accumulated tax credits in order to capture the impact of the TCJA tax break on sponsor contributions. First, marginal tax rates may not be the relevant tax rates for sponsor contribution decisions. There is evidence that firms may prefer to use simple heuristics such as statutory and effective tax rates to evaluate incremental decisions, rather than harder-to-estimate marginal tax rates (Graham et al., Reference Graham, Hanlon, Shevlin and Shroff2017). This suggests that our tax-exposure measure, which is based on the statutory tax rate, is a more suitable proxy than the marginal tax rate for capturing the impact of tax-based incentives on sponsor contributions. Second, the tax credit dummies might incorrectly classify some sponsors as not exposed to the TCJA tax break. This is because the exposure measures based on accumulated tax credits reported in Compustat include tax credits accrued to foreign subsidiaries, as well as domestic subsidiaries which are unconsolidated for tax purposes (Corporate Taxes and Defined Benefit Pension Plans, 1988). By contrast, pension contributions are deducted from corporate income net of income from such subsidiaries, so sponsors may be subject to tax-based incentives even if the no-exposure dummies are equal to 1.

4. The effect of TCJA on pension plans

4.1 Identification strategy

Our empirical strategy exploits cross-sectional variation in sponsor exposure to tax-based incentives to assess the impact of the TCJA tax break. We use non-tax-exposed sponsors as a control group to assess the counterfactual level of voluntary and total pension contributions in the absence of the tax break for the tax-exposed firm.Footnote ³² This allows us to estimate of the marginal impact of the TCJA tax break on contributions and funding. The identification strategy depends on the assumption that tax-exposed (treatment) and non-tax-exposed control firms do not differ across dimensions other than tax-based incentives that may affect voluntary contributions during the sample period.

Exposure to tax-based incentives is not random in our sample. Table 3 reports the correlation of our measure of tax-exposure with other variables that are likely to affect pension contribution patterns: plan funding (Funding Ratio), profitability metrics (Return on Investment and Discount Rates), PBGC premia, proxies for sponsor bankruptcy risk (Altman's Z), and the opportunity cost of internal resources (Cash Flows, CAPEX, Payout, Tobin's Q). Tax-exposed firms have more underfunded pension plans, higher PBGC variable premium, higher payout, and higher pre-contributions cash flows, which all push for higher contributions. At the same time, tax-exposed sponsors have higher pension liability discount rates, which would tend to reduce sponsor incentives to contribute. The tax-exposed also display lower CAPEX, which could be associated with relatively lower contributions if resulting from more binding constraints on external finance. On balance, it is not obvious that the significant correlates in Table 3 will bias our estimates in a specific direction. To account for all possible biases, we include the observable correlates as controls in our empirical specifications.

Table 3.

Tax exposure and plan- and sponsor-level characteristics

Notes: This table presents pooled probit regression estimates of the impact of different plan-level (rows (1)–(4)) and sponsor-level characteristics (rows (5)–(9)) on the probability that a sponsor is tax exposed. The dependent variable is Tax exposure, a dummy variable equal to 1 if (i) a firm has a positive ex-contributions tax bill and (ii) the firm sponsors at least one plan with funding ratio below 150%. Z-statistics obtained using robust standard errors are in parentheses.

4.2 The TCJA tax break and contributions

The conceptual framework outlined in Section 2 indicates that TCJA should have has a positive impact on 2017 contributions. By contrast, it should have had a negative impact on 2018 contributions (claim 1). Therefore we should expect 2017 contributions from tax exposed firms to be higher than those of non-tax-exposed firms. We should expect the opposite for 2018 contributions. The model also suggests that whether the drop in 2018 contributions is large enough to offset the 2017 increase depends on the time profile of sponsor financial constraints (claim 2). Empirically, we can gauge whether this was the case or not by comparing the 2017 (differential) response of tax-exposed firms to tax-based incentives to their 2018 response.

A graphical analysis suggests that the TCJA tax break had a positive impact on 2017 contributions and a negative impact on 2018 contributions, as expected. We split the sample into two groups according to tax-based incentives, proxied by our tax exposure measure. The first group includes firms that have pre-pension contribution tax-based incentives (i.e., Tax Exposure =1) and the second group includes firms that have no pre-pension contribution tax-based incentives (i.e., Tax Exposure =0). The left panel of Figure 3 plots average yearly voluntary pension contributions from 2014 through 2018, for both tax-exposed and non-tax-exposed firms. The difference between voluntary contributions from tax-exposed and non-tax-exposed sponsors was relatively stable prior to the TCJA tax break (2014–2016). In 2017, contributions from tax-exposed sponsors increased by 0.008 per cent of sponsor assets. By contrast, contributions from non-tax-exposed sponsors increased by 0.0024 per cent. In 2018, pension contributions from tax-exposed firms decreased significantly relative to those of non-tax-exposed firms. Given the more permanent nature of changes in other time-varying contribution incentives (such as increases in the PBGC variable premium), it is difficult to argue that this increase/decrease pattern can be accounted for by something other than TCJA and the ensuing temporary tax break.

Figure 3.

Voluntary contributions and the TCJA.

Regression analysis confirms the findings of the graphical analysis on impact. We estimate the following cross-sectional specification:

(9)$$\displaystyle{{{\rm Voluntary\;Contribution}{\rm s}_{s, t}} \over {{\rm Assets\;( book}) _{s, t-1}}} = \alpha _t + \beta _t{\rm Tax\;Exposur}{\rm e}_{s, t} + \delta _tZ_{s, t} + \varepsilon _{s, t}, \;\quad {\rm for\;}t = 2014, \;\ldots , \;2018.$$

Here, Tax Exposure is our measure of tax-based incentives and Z is a vector of controls which includes the observable correlates of tax-based incentives. The β _t coefficients are the coefficients of interest, as they capture the impact of tax deductibility of contributions on contributions in each of our sample years. We plot them on the right panel of Figure 3. A priori, we would expect the impact of tax-based incentives in 2017 to be above pre-TCJA average, $\beta _{2017} > \sum\nolimits_{t = 2014}^{2016} {{\hat{\beta }}_t/3}$, and the impact of tax-based incentives in 2018 to be below, $\beta _{2018} < \sum\nolimits_{t = 2014}^{2016} {{\hat{\beta }}_t/3}$.

Tax-based incentives had a larger impact on contributions in 2017 than in the 3 years pre-TCJA. The 2017 estimate of the tax exposure coefficient, $\hat{\beta }_{2017}$, is positive and significant (Table 4, columns (1) and (2)). This result is robust to including controls for the observable correlates of our tax exposure measure (column (2)), assuaging concerns about identification. According to our preferred specification (with controls, column (2)), voluntary contributions from tax-exposed sponsors were 0.037 percentage points larger than their counterpart from non-tax-exposed firms. By contrast, the average impact of tax-based incentives on voluntary contributions prior to the TCJA, $\sum\nolimits_{t = 2014}^{2016} {{\hat{\beta }}_t/3}$, was around 0.022 percentage points, making the 2017 impact about one-third of a standard deviation higher than the pre-TCJA average.

Table 4.

Pension contributions, tax-based incentives, and the TCJA

Notes: This table presents regression estimates of the response of voluntary pension contributions to fiscal incentives in each year between 2014 and 2018. The dependent variable is voluntary pension contributions deducted from 2014 tax returns in columns (1) and (2); from 2015 returns in columns (3) and (4); from 2016 returns in columns (5) and (6); from 2017 returns in columns (7) and (8); and from 2018 returns in columns (9) and (10). Tax exposure is a dummy variable equal to 1 if (i) a firm has a positive ex-contributions tax bill and (ii) the firm sponsors at least one plan with funding ratio below 150%. The TCJA reduced the federal corporate tax rate from 35% to 21% beginning in 2018. As a result, contributions counted towards the 2017 corporate tax return could be deducted at 35%, while contributions counted towards 2018 returns at 21%. Columns (2), (4), (5), (6), and (8) include the following plan-level controls: Funding ratio, PBGC premium, Return on investment, Discount rate. They also include the following sponsor-level controls: Altman's Z-score, Cash flows, CAPEX, Tobin's Q, Payout, and DB plans share. t-statistics obtained using robust standard errors are in parentheses.

The impact of tax-based incentives on 2018 contributions was below pre-TCJA average, with a large enough deviation to fully offset the above-average 2017 effect (reversal). The 2018 estimates of the tax exposure coefficient, $\hat{\beta }_{2018}$, are at the minimum level over our five-year sample period (Table 4, columns (1) and (2)), implying that tax-based incentives to contribute were at their weakest right after the end of the tax break. In our preferred specification (with controls, column (2)), the impact of tax-based incentives in 2018 amounted to 0.008 percentage points. At around one-third of a standard deviation lower than pre-TCJA average, this decline fully offset the 2017 increase. We interpret this as evidence that tax-exposed firms shifted planned future contributions from 2018 to 2017. Through the lens of our model, the fact that 2018 tax-based incentives completely reversed the effects of 2017 incentives on contributions is consistent with sponsor financial constraints being relatively more binding than future expected constraints (claim 2).

These results are robust to changing the definition of tax exposure to a dummy variable equal to 1 if only the firm sponsors at least one plan with funding ratio below 150 per cent (Tables 5 and 6). These results are also robust to including sector fixed effects. Estimates with sector fixed effects are qualitatively similar to estimates without (Table 7). The second column of Table 7 changes the dependent variable to total pension contributions. We continue to find a positive impact of the TCJA tax break in 2017, followed by a reversal in 2018.

Table 5.

Tax exposure and plan- and sponsor-level characteristics: robustness

Notes: This table presents pooled probit regression estimates of the impact of different plan-level (rows (1)–(4)) and sponsor-level characteristics (rows (5)–(9)) on the probability that a sponsor is tax exposed. The dependent variable is Tax exposure, a dummy variable equal to 1 if the firm sponsors at least one plan with funding ratio below 150%. Z-statistics obtained using robust standard errors are in parentheses.

Table 6.

Contributions, tax-based incentives, and the TCJA: robustness

Notes: This table presents regression estimates of the response of voluntary pension contributions to fiscal incentives in each year between 2014 and 2018. The dependent variable is voluntary pension contributions deducted from 2014 tax returns in columns (1) and (2); from 2015 returns in columns (3) and (4); from 2016 returns in columns (5) and (6); from 2017 returns in columns (7) and (8); and from 2018 returns in columns (9) and (10). Tax exposure is a dummy variable equal to 1 if the firm sponsors at least one plan with funding ratio below 150%. The TCJA reduced the federal corporate tax rate from 35% to 21% beginning in 2018. As a result, contributions counted towards the 2017 corporate tax return could be deducted at 35%, while contributions counted towards 2018 returns at 21%. Columns (2), (4), (5), (6), and (8) include the following plan-level controls: Funding ratio, PBGC premium, Return on investment, Discount rate. They also include the following sponsor-level controls: Altman's Z-score, Cash flows, CAPEX, Tobin's Q, Payout, and DB plans share. t-statistics obtained using robust standard errors are in parentheses.

Table 7.

Contributions, tax-based incentives, and the TCJA: robustness

Notes: This table presents regression estimates of the response of pension contributions to fiscal incentives in each year between 2014 and 2018. The dependent variable is voluntary pension contributions deducted from 2014 returns in column (1); from 2015 returns in column (3); from 2016 returns in column (5); from 2017 returns in column (7); and from 2018 returns in column (9). The dependent variable is total pension contributions deducted from 2014 returns in column (2); from 2015 returns in column (4); from 2016 returns in column (6); from 2017 returns in column (8); and from 2018 returns in column (10). Tax exposure is a dummy variable equal to 1 if (i) a firm has a positive ex-contributions tax bill and (ii) the firm sponsors at least one plan with funding ratio below 150%. The TCJA reduced the federal corporate tax rate from 35% to 21% beginning in 2018. As a result, contributions counted towards the 2017 corporate tax return could be deducted at 35%, while contributions counted towards 2018 returns at 21%. All columns include the following plan-level controls: Funding ratio, PBGC premium, Investment return, Discount rate. They also include the following sponsor-level controls: Altman's Z-score, Cash flows, CAPEX, Tobin's Q, and DB plans share. t-statistics obtained using robust standard errors are in parentheses.

In dollar values, our estimates imply a $2.8bn to $5.0bn increase in voluntary contributions to medium- and large-scale plans associated with the tax break, depending on whether or not the specification includes controls (the impact is larger with controls). Our estimates report the TCJA impact in percentage points, so we multiply by tax exposed sponsor assets to obtain a dollar figure. Accordingly, the tax break impact on voluntary contributions in dollars in sample is given by $\left({{\hat{\beta }}_{2017}-\sum\nolimits_{t = 2014}^{2016} {{\hat{\beta }}_t/3} } \right)A_{2016}/100$, where $A_{2016} = \sum\nolimits_{s = 1}^{381} {A_{s, 2016}}$ represents the total assets of the 381 tax-exposed sponsors in our sample at the end of 2016 (beginning of 2017). We obtain a $1.3bn increase in voluntary contributions for the specification without controls and a $2.3bn increase for the specification with controls. By assumption, the TCJA had no impact on contributions from the non-tax-exposed. Assuming that our sample is representative of the broader population of firms submitting Schedule SB of the IRS 5500 filings – some of which are not listed, and therefore do not appear in Compustat – we extrapolate to estimate the TCJA impact on the voluntary contributions of all sponsors of middle- and large-scale plans. To that end, we multiply the in-sample estimates by the ratio of total voluntary contributions by Schedule SB filers to total voluntary contributions by sponsors in our sample, which is equal to $6.7bn/$3.1bn. To compute the tax break impact on total contributions for firms in sample, we repeat the same steps using the estimates in Table 7 instead. This returns a $15.3bn increase in voluntary contribution for the specification without controls and a $37bn increase for the specification with controls. Total contributions by firms in our sample amount to $50bn, while total contributions by Schedule SB filers are equal to $107.7bn. This implies a $33bn to $79.7bn increase in total contributions to medium- and large-scale plans associated with the tax break.

4.3 The TCJA tax break and funding ratios

In this section, we study other effects of the TCJA tax break on firms and their DB retirement plans. We examine whether or not the tax break had an impact on funding ratios. We find that our results on contributions carry over to plan funding ratios.

Our estimates suggest that the TCJA tax break had a short-lived impact on plan funding. While the TCJA increased 2017 funding ratios, by 2018 they were already back where they would have been in the absence of the intervention. We estimate the following specification:

(10)$${\rm \Delta Funding\;Rati}{\rm o}_{s, t, t-1} = \alpha _t + \beta _t{\rm Tax\;Exposur}{\rm e}_{s, t} + \delta _tZ_{s, t} + \varepsilon _{s, t}, \;\quad {\rm for\;}t = 2017, \;\,2018.$$

Here, Funding Ratio is defined as in Section 3.2.2 and Z _t is a vector of controls which includes pre-TCJA plan funding status (Funding Ratio in 2016), the actual investment return on plan assets and the change in discount rates between t and t − 1. Results are reported in columns (1)–(4) of Table 8. Tax-exposed firms experienced an increase of 2.5–3.4 percentage points in the funding status of their corporate pension plans between 2016 and 2017 (relative to non-tax-exposed firms), depending on whether or not the specification includes controls. Firms that were tax exposed in both 2017 and 2018 saw a relative decrease of 2 percentage points in the funding status of their corporate pension plans between 2017 and 2018. Columns (5) and (6) report the results of estimating a variant of (10) which considers the change in funding ratios between end-2016 and end-2018, again focusing on firms that were tax exposed in both 2017 and 2018. The coefficient of 2017 Tax Exposure is not significant, confirming that the temporary increase tax incentives for contributions associated with the TCJA had no long-lasting impact on funding ratios.

Table 8.

Funding ratios and the TCJA

Notes: This table presents regression estimates of the response of pension contributions to fiscal incentives in each year between 2014 and 2018. The dependent variable is voluntary pension contributions deducted from 2014 returns in column (1); from 2015 returns in column (3); from 2016 returns in column (5); from 2017 returns in column (7); and from 2018 returns in column (9). The dependent variable is total pension contributions deducted from 2014 returns in column (2); from 2015 returns in column (4); from 2016 returns in column (6); from 2017 returns in column (8); and from 2018 returns in column (10). Tax Exposure is a dummy variable equal to 1 if (i) a firm has a positive ex-contributions tax bill and (ii) the firm sponsors at least one plan with funding ratio below 150%. The TCJA reduced the federal corporate tax rate from 35% to 21% beginning in 2018. As a result, contributions counted towards the 2017 corporate tax return could be deducted at 35%, while contributions counted towards 2018 returns at 21%. All columns include the following plan-level controls: Funding ratio, PBGC premium, Investment return, Discount rate. They also include the following sponsor-level controls: Altman's Z-score, Cash flows, CAPEX, Tobin's Q, and DB plans share. t-statistics obtained using robust standard errors are in parentheses.