Pension contributions and firm performance: evidence from frozen defined benefit plans.
In recent years, an increasing number of corporations have resorted to altering the structure of employee retirement plans from defined benefit (DB) to defined contribution (DC) in response to escalating pension costs. For example, in 2008, the dramatic decline in the stock market resulted in a sharp drop in DB pension plan assets, whereas the fall in interest rates raised the value of pension liabilities. Consequently, pension plans of S&P 1,500 companies showed an aggregate deficit of $409 billion with overall funding at 75% of pension obligations. Press reports suggest that the shortfall forced companies to divert funds from growing their businesses to pay for pension obligations, to freeze DB plans, and to replace them with DC plans.
A DB plan promises employees a stream of monthly retirement benefits that are determined based on their age, earnings, and years of service. Typically, only the employer makes regular and consistent contributions to the DB plan and bears the full investment risk and beneficiaries' longevity risk. In contrast, a DC plan specifies the contributions that the employee and the employer choose to make and promises no specific retirement benefits, except that the employee is entitled to the investment results derived from those contributions. DB plan contributions are higher, on average, and less predictable than DC plan contributions (Munnell et al., 2006). The former tends to depress corporate liquidity and capital expenditures as a firm with underfunded pension obligations is required, under pension laws, to make annual contributions by applying an arbitrary nonlinear formula based on its pension funding status (Rauh, 2006). In comparison, DC plans are viewed as more flexible because they allow firms to vary contributions according to their cash flow and lower their operating leverage (Petersen, 1992). Retirement analysts observe that though DB plans tend to place too much burden on employers, DC plans swing to the opposite end of the spectrum by dumping much of the burden on the individual.
In this paper, we investigate the response of corporate liquidity, financial leverage, investment, and firm value to potential reductions in legally required pension contributions caused by the strategic decision of sponsoring firms to freeze DB plans and replace them with DC plans. Our research is closely related to some recent papers. First, Shivdasani and Stefanescu (2010) observe that the magnitude of the liabilities arising from DB pension plans is substantial and firms incorporate the magnitude of their pension assets and liabilities into their capital structure decisions. Additionally, examining the impact of negative shocks to internal resources caused by required pension outlays based on sharply nonlinear funding rules under pension laws governing DB plans, Rauh (2006) reports that capital expenditures decline with mandatory contributions (MCs), particularly so for financially constrained firms. Moreover, Franzoni and Marin (2006) find that the emergence of large pension deficits for sponsors is followed by negative stock returns. Finally, Franzoni (2009) concludes that MCs associated with underfunded DB plans lead to negative stock returns and are more pronounced for firms exposed to financial constraints and strong governance structures. In light of this evidence, one would expect that a strategic decision to close existing DB pension plans (eliminating future mandatory pension outlays) and substitute DC plans would relieve the debt pressure and attenuate the negative effects of DB plans on corporate liquidity, investment, and value, particularly for financially constrained firms. However, if the decision to switch from DB to DC plans is prompted primarily by a severe liquidity crunch or financial distress, then the near-term net benefits from switching in terms of investment efficiency and shareholder value would be small. Further, because capital investment requires both immediate outlays, high adjustment costs, and future capital commitment, financially constrained firms would give priority to building up liquidity and recapitalizing in the short-term before embarking on new investment (Pulvino and Tarhan, 2006; Almeida and Campello, 2010; Marchica and Mura, 2010; Dasgupta, Noe, and Wang, 2011; among others). These arguments suggest that we need to perform empirical tests to sort out the liquidity, financing, investment, and net firm value effects of the decision to freeze existing DB pension plans and replace them with DC plans.
Studies by Rauh (2006) and Franzoni (2009) depend on the level of MCs associated with the DB plans to identify exogenous shocks to internal liquidity and the resulting effect on investment and firm value. Re-examining Rauh's (2006) findings that mandatory pension contributions cause a sharp drop in capital expenditures, Bakke and Whited (2012) find little evidence that firms cut back on investment and conclude that Rauh's (2006) results are likely due to the endogeneity of benefit contributions. This research is complimentary to Rauh's (2006) study of exogenous drops in internal liquidity due to mandatory pension contributions. We focus our analysis on the potential financial, investment, and value effects of changes in internal liquidity due to a regime shift in pensions, that is, a deliberate change from required contributions under DB plans to discretionary contributions under DC plans. Regime shifts in pensions provide a unique setting that is particularly advantageous in identifying the financial and real effects of improvements in internal resources because the change in the pension plans has the potential to increase the share of free cash flows allocated to equity investors as compared to employees given the evidence that DC plans are less costly, on average. It is true that this shift in the pension regime is itself endogenous to the sponsoring firm. Yet it mitigates the mandatory nature of DB plan pension outlays that the firm faces when experiencing adverse economic conditions. Moreover, we employ different techniques to explicitly address the potential endogeneity bias caused by the shift to ensure the robustness of our results.
We study a sample of 1,071 firms that sponsor DB pensions, of which 179 firms freeze at least one DB plan, from 2001 to 2008. Our results indicate that though MCs under DB plans continue to drain liquidity and increase financial leverage, the freeze tends to mitigate the negative impact of MCs on corporate liquidity and relieve the pension-induced debt burden. Further, consistent with Rauh (2006), we find evidence in support of the argument that MCs depress investment, but there is little evidence that the shift from DB to DC plans improves corporate investment in a significant way over the following three years. These findings are robust to alternative corrections for endogeneity of the freeze decision. The regime shift from DB to DC plans appears to generate stronger liquidity and leverage benefits to sponsors facing financial constraints. Finally, our analysis of a subsample indicates positive abnormal stock returns during the DB-DC shift announcement window. However, we find little significant change in the effects of MCs on long run stock returns up to three years following the DB-DC shift.
This study makes several important contributions to the literature. First, our research offers new evidence taken from recent data (2001-2008) relative to the findings of Rauh (2006) and Franzoni (2009) (based on the 1990-1998 period) that a negative shock to internal cash flow, caused by the legally required DB pension contributions, tends to hurt corporate investment and firm value, particularly that of financially constrained firms. In our full sample, covering both frozen and nonfrozen DB plans, we document a substantial decline in the marginal negative investment and value effects of MCs despite the sharp market declines in recent years. Our point estimates are on the order of $0.05-$0.21 reduction in capital expenditures per dollar of required contributions, as compared to the $0.60 drop in investment reported by Rauh (2006). Nonetheless, these adverse investment effects are not inconsequential in light of the average positive investment cash flow coefficient of 6% in our sample. Further, we find that a one dollar increase in MCs depresses shareholder value by $0.17-$0.21, as compared with a $0.99 drop in value reported by Franzoni (2009) based on a 1990-1998 sample.
Additionally, prior research has underscored the serious identification problem arising from omitted investment opportunities in the investment-cash flow sensitivity tests. We believe that a study of regime shifts that fundamentally alter MCs, with suitable controls for the endogeneity of the freeze decision itself, improves our understanding of the linkages among corporate pensions, investment, liquidity, capital structure, and firm value. Specifically, consistent with recent evidence documented in the literature (Dasgupta et al., 2011), we find that a positive change in internal cash flow of constrained firms, induced by the shift from DB to DC plans, has a first order impact on liquidity and financing, but the real effect (i.e., investment effect), if any, appears to be secondary. Moreover, previous research on the immediate market reaction to changes in corporate pension plans does not find a significant impact of the change announcement on firm value (McFarland, Pang, and Warshawsky, 2009; Milevsky and Song, 2010). (1) In addition, when examining the effects of MCs on stock performance in a long run event study, Franzoni (2009) documents that required contributions have a negative impact on stock performance. Our investigation provides new evidence regarding the short-term and long-term effects of the DB-DC shift on firm value. Finally, because adverse capital market conditions aggravate the funded status of DB pension plans and can pose a real threat to their viability and sustainability, our comprehensive analysis of the financial and the real (investment) effects of regime switches in pensions is expected to be of value in crafting corporate policy on pension plans and regulatory policy on employee retirement income security and welfare.
The remainder of the paper proceeds as follows. Section I describes the freeze mechanisms and discusses the expected impact of pension freezes on liquidity, capital structure, and investment. We describe the sample and measurements in Section II and present empirical results in Section III. This is followed by an analysis of the firm value effects of the freeze in Section IV. Section V provides our conclusions.
I. Potential Impact of Pension Freezes
This section discusses the main factors that motivate firms to close out DB plans and replace them with DC plans, as well as the potential effects of pension freezes on liquidity, leverage, investment and firm value. The high costs and risks of DB plans are often advanced as reasons for freezing pensions and replacing a DB plan with a DC plan. Munnell et al. (2006) report that the average funding contribution to DB plans is almost three times that of a typical DC plan. In an industry where competitors adopt DC plans, a firm is likely to weaken its own competitive position if it maintains its DB plans. The values of DB plan assets and liabilities are sensitive to changes in interest rates and equity returns (Feldstein and Seligman, 1981; Franzoni and Marin, 2006). A mismatched duration between pension assets and liabilities subjects the pension plan to interest rate risk, and exposes the plan to underfunding status which triggers the required pension contributions. Indeed, the market decline in 2000-2001 led to a sharp increase in pension contributions, from an average annual amount of about $30 billion per year between 1980 and 2000 to $45 billion in 2001 and about $100 billion in 2002 and 2003 (Munnell and Soto, 2007). In 2005, the Government Accountability Office (GAO) estimated that, in aggregate, DB plans were underfunded by a total of $450 billion, with more than half of the largest plans underfunded. Moreover, legacy costs, which originate from the agreements between corporations and their unions to increase pension benefits in place of wage increases, also worsen a DB plan's funding situation. As more people retire, there are fewer active employees on the payroll to contribute to the funding of rising pension costs.
Firms generally use one of the three types of freezes of their DB plans: 1) hard freeze, 2) soft freeze, and 3) partial freeze. When a hard freeze is introduced, all employees become fully vested immediately in benefits earned under the plan, but they are not entitled to earn additional benefits. Under a soft freeze, pension credits for future years of work are limited although allowances for compensation increases may be made. In a partial freeze, the firm ceases or limits the accrual of additional benefits for some, usually new hires, but not all employees (Beaudoin, Chandar, and Werner, 2010). The National Compensation Survey conducted by the Department of Labor in 2008 reports that to offset the employees' loss of benefits due to a freeze, DB sponsors offered alternative pension programs such as new DC plans (typically 401 (k) plans), enhanced existing DC plans, but certainly with lower benefit accruals, to over 95% of the frozen DB plans, on average.
We will only focus on hard freezes when studying the impact of pension restructuring as they are likely to result in measurable financial, investment, and value effects. It is important to emphasize that a hard DB plan freeze will eliminate the sponsor's exposure to required contributions for future years of service and alter the structure of its pension obligations in favor of discretionary contributions, but the sponsor will continue to be liable for MCs with respect to benefits that the participants have already earned. Thus, even if a firm had only DB plans before a hard freeze, its postfreeze pension assets, liabilities, and outlays would include the old DB and the new DC plans.
For each outcome variable (i.e., liquidity, financial leverage, investment, and value of equity), we explore three types of effects related to the pension freeze. The first is the marginal (slope) effect of mandatory pension contributions on the outcome variable of interest. The second is the average (intercept change) effect of the pension freeze on the dependent variable. The final effect is the interaction (change in slope) effect of the freeze with required contributions on the outcome variable of interest. The interaction effect reflects whether the freeze effect varies with the level of statutory contributions imposed on the firm. Our expectation is that, in general, the benefits of a hard freeze would be more pronounced for firms facing financial constraints. Because firms facing smaller financial constraints are likely to fund their pension obligations fully, MCs, which represent exogenous shocks in cash flows, are likely to be positively correlated with the degree of financial constraints of pension sponsors. Therefore, we expect the interaction term to capture the freeze effects on the respective outcome variables of financially constrained firms. Furthermore, we expect the aforementioned three effects to only be concentrated in financially constrained firms that face higher external costs of capital. As emphasized by the related literature, required contributions, as well as the hard freeze, are not expected to affect the liquidity, leverage, investment, and firm value of financially unconstrained firms.
A. Liquidity Effects
Because required pension contributions drain internal resources of financially constrained firms, the marginal effect on liquidity would be negative for financially constrained firms as MCs increase. Further, we expect a hard DB freeze to generate opposing types of effects on the average liquidity position of a constrained sponsor. The lower average contributions of DC plans that replace DB plans result in immediate and future net pension savings and improve the average level of liquidity. However, by substituting discretionary contributions under DC plans for the MCs associated with underfunded DB plans, the freeze would relieve the strain on corporate liquidity and reduce the need to hold liquid assets, thus lowering the liquidity level. (2) Firms are arguably more likely to freeze underfunded pension plans. As such, the potential improvement in liquidity can be directed to pension contributions to shore up the funding status of the existing underfunded DB plans.
Additionally, as a freeze is likely triggered by increasing financial constraints, we expect the sponsors to deploy the resulting pension savings to build up liquidity or to other financial and real investment uses. Investigating the behavior of financially constrained firms following a positive cash flow shock, Almeida and Campello (2010) and Dasgupta et al. (2011) suggest that these firms, in anticipation of future costly external financing, are expected to build up liquidity to smooth out the investment process. Further, Dasgupta et al. (2011) argue that constrained firms use the improved cash to pay down debt to build financial slacks for future investment. However, Almeida and Campello (2010) put forward three important reasons as to why constrained firms should use incremental cash flow to build up liquidity instead of reducing external financing. First, because financially constrained firms face high opportunity costs of investment, they may use internal funds for additional capital spending. Additionally, financially constrained firms also worry about future investment. The need to fund future investments under credit constraints increases their demand for liquid assets. Moreover, internal funding and external financing could be complements rather than substitutes implying that improved liquidity facilitates even more external financing. Together, these opposing freeze effects lead to an ambiguous net effect on the average level (i.e., change in intercept) of liquidity of financially constrained firms.
Furthermore, prior studies by Rauh (2006) and Franzoni (2009) report that measures of financial constraints are significantly positively correlated with the level of MCs. We expect that lower contributions and the increased flexibility of DC plans following the freeze would be especially valuable to firms facing financial constraints when compared to unconstrained firms. This argument suggests that for financially constrained firms, the interaction effect between the pension freeze and MCs would be positive.
B. Leverage Effects
Examining the interplay between pension and financial leverage, Shivdasani and Stefanescu (2010) observe that firms consider the magnitudes of pension assets and liabilities when they make capital structure decisions. DB plan liabilities are fixed long-term obligations and the contributions are tax deductible. (3) If firms choose to fund DB plans with borrowed funds, their combined fixed obligations (pension plus debt) would be larger, and they would reap more tax shields. Empirically, they estimate that the tax benefits of consolidated leverage (pension and financial debt) are 47% higher than the tax benefits of financial debt alone. This tax shield-based argument suggests a positive correlation between financial leverage and required contributions (i.e., a positive marginal effect) of financially constrained firms.
In addition, the drop in MCs triggered by the freeze should lower the average level of debt over the postfreeze years. However, the freeze shrinks fixed obligations represented by the pension liability enabling firms to raise debt to recoup the lost tax shields. Also, pension liability is senior even to the most senior lenders (Stewart, 2003; Shivdasani and Stefanescu, 2010). Therefore, the DB plan freeze would dampen the growth in the size and volatility of pension liabilities reducing the default risk faced by subordinate debt holders leading to a possible decrease in interest rates. Moreover, the hard freeze shifts pension risk with respect to future years of service to the employees, which is likely to lower firms' borrowing costs and increase their incentive to borrow. (4) These arguments suggest that the pension freeze would likely increase the average level of debt employed by financially constrained firms to capture the lost pension tax shields and lower borrowing costs (i.e., a positive intercept effect).
The positive intercept effect discussed above reflects the increase in leverage after the freeze of a firm with average financial constraints. However, firms with more than average financial constraints face increased costs of borrowing due to information asymmetry and moral hazards. As the hard freeze is likely triggered by the poor financial health of sponsors, firms facing more financial constraints are less likely to be able to borrow to fund MCs over the subsequent years despite the reduced pension burden, diminished pension risk, and lower interest rates. In contrast, their peers, with lower than average financial constraints, would have greater incentives to borrow after the freeze. Recall that the degree of financial constraints is positively correlated with MCs. Therefore, we would expect the leverage of financially constrained firms to be less responsive to MCs following the hard freeze. In other words, the interaction effect of the freeze and the required contributions on financial leverage of financially constrained firms would be negative.
C. Investment Effects
Financial theory suggests that financial constraints due to capital market frictions increase the sensitivity of a firm's investment to cash flows after controlling for investment opportunities (Fazzari, Hubbard, and Petersen, 1988). However, the empirical investment-cash flow sensitivity model pioneered by Fazzari et al. (1988) tends to bias the coefficient estimate of cash flows due to potential measurement error problems when using the average Tobin's Q as a proxy for investment opportunities (Erikson and Whited, 2000; Gomes, 2001; Alti, 2003). To circumvent this measurement error problem, recent research attempts to identify exogenous changes in internal cash flows and examine their effect on investment (Blanchard, Lopez-de-Silanes, and Shleifer, 1994; Fee, Hadlock, and Pierce, 2009, among others). Rauh (2006) argues that nonlinear funding rules under US pension laws for DB plans lead to exogenous shocks to firms' internal cash flows. Employing the regression discontinuity approach to model the sharply nonlinear relationship between the level of mandatory pension contributions and the funding status, he reports that MCs exacerbate financial constraints and cause a sharp decline in capital expenditures of DB sponsoring firms from 1990 to 1998, particularly for financially constrained firms.
Financial constraints not only hamper investment, but also cause other corporate investment distortions, such as inducing firms to select investment projects with shorter payback periods, shifting investment to less risky projects, or projects utilizing more pledgeable assets instead of selecting projects based on the net present value (NPV) criterion (Almeida, Campello, and Weisbach, 2011). Alderson and Betker (2009) examine the impact of required contributions on the internal capital of multi-segment financially constrained firms during the "perfect" pension storm of 2001-2002 and observe that these firms direct more investment toward segments that generate greater cash flows. Although such investment reallocation could be helpful in generating the additional cash flows that constrained firms badly need, it represents a form of investment distortion. Alternatively, firms may lay off employees and cut investment to cope with pension shortfalls and required contributions, or (hard) freeze pensions to sustain their investment programs. When firms freeze DB plans and replace them with DC plans, not only is the risk of future pension deficits and statutory contributions reduced, but even pension contributions to the new DC plans are expected to decrease, thus reducing their exposure to liquidity risk. Although the presence of DB plans amplifies an exogenous negative shock to cash flows through MCs, the subsequent (hard) freeze and ultimate closure of those plans generates a positive shift in internal resources by extinguishing (or attenuating) the exposure to expected required contributions related to future years of service. Firms could use such pension cost savings to improve their investment programs instead, especially when they face financial constraints, to maintain steady growth.
However, it is plausible that firms may decide to switch pension plans when their financial constraints are so acute that they can no longer support the required contributions under the DB plans. Because the hard freeze is replaced by a DC plan, albeit less costly, that serves to reduce only future pension obligations, it may take years before the resulting cost savings can ameliorate the liquidity position of constrained firms in a significant way. Specifically, the hard DB pension freeze does not relieve the firms from required contributions to bring their existing underfunded DB plans, if any, to the full funding status implying a continuing liquidity drain even after the freeze. Because capital investment requires both immediate outlays and future capital commitment, as well as high adjustment costs, a financially constrained firm would give priority to building up liquidity and recapitalizing in the short run before embarking on any new investment (Pulvino and Tarhan, 2006; Almeida and Campello, 2010; Marchica and Mura, 2010; Dasgupta et al., 2011). Furthermore, cash starved firms choose to hard freeze DB plans in hard times when investment opportunities are low, so an uptick in investment may not be observed until several years after the freeze. Overall, these considerations suggest that capital expenditures may continue to remain depressed even after a hard pension freeze. The above discussion suggests that though the marginal effect of MCs on capital expenditures of financially constrained firms would be negative, the average (i.e., change in intercept) effect of the freeze and its interaction effect with required contributions on capital expenditures of financially constrained firms is more difficult to predict.
D. Value Effects
Franzoni (2009) argues that MCs should have no impact on firm value in a frictionless world as such payments are nothing but the conversion of cash into assets in the pension plan, which is effectively a fully owned subsidiary of the firm. However, he finds that investors react negatively to required pension contributions under DB plans, more significantly so if the sponsors are financially constrained. Mitchell and Mulherin (1989) report a positive stock price reaction around the date DB plan terminations are filed. In our context, we would expect a negative marginal effect of required contributions on long-term abnormal stock returns of financially constrained firms before the freeze based on the findings of Franzoni (2009). However, the pension freeze is associated with several counteracting value effects. A shift from a DB to a DC plan can help firms save pension costs and allow the sponsor the flexibility to scale DC contributions up or down as dictated by exogenous shocks to its financial condition. In the presence of market frictions and the resulting costly external finance, the improved liquidity and flexibility resulting from a shift in pension regimes is likely to allow financially constrained firms to undertake positive NPV projects, which they would have otherwise forgone, thus enhancing firm value.
Nevertheless, the bargaining power of labor unions and labor market competition to recruit and retain talent could put some limits on that flexibility. Freezing a costly and risky DB plan and replacing it with a DC plan is analogous to a wealth transfer from employees to the firm and a risk shift in the opposite direction. Employees have their human capital tied to the firm, which is difficult to diversify. Less favorable pension policies toward employees could pose a real challenge to firms in maintaining productivity and attracting and retaining talent, resulting in unfavorable long-term financial consequences. Moreover, a pension switch may simply signal to the market the severity of the firm's financial distress, leading to an expected negative market reaction. Finally, as discussed above, it may take quite a while to realize any gains in real investment from a hard freeze, thus making it difficult to detect the potential change in firm value even over periods up to three years after the freeze. Therefore, the net average short-term and long-term effects (i.e., change in intercept), as well as the interaction effect of the shift from a DB to a DC plan and MCs on firm value of financially constrained firms, are an empirical question.
II. Data and Summary Statistics
Our main source of information about firms that sponsor DB plans and those that freeze DB plans and shift to DC plans is Form 5500 that firms file with the Internal Revenue Service (IRS) annually. These data are compiled from 1990 to 2007 by Boston College's Center for Retirement Research. DB freezes at the plan level were first reported in Form 5500 in 2002. Therefore, our analysis focuses on the period starting one year earlier and ending one year later from 2001 to 2008. Because a firm may have more than one type of pension plan and freeze those plans at different times, when constructing our sample of DB freeze firms, we identify the timing of the first plan freeze that a firm makes during our sample period. Current accounting and actuarial practices allow for smoothing of reported pension assets and liabilities, so pension accounting data may not reflect the true pension obligations of firms. Consequently, we base our analyses on mandatory pension contributions and actual pension contributions, both of which are aggregated, using the Employer Identification Number (EIN), to firm-level from individual DB plans as reported in Form 5500. Then, we match the firm-level pension data with Compustat data based on EIN. For those firms that cannot be matched with Compustat data, we manually match by firm name using the pension plan sponsor's name in Form 5500 and the company name in Compustat. We acknowledge that our matching process is not perfect as a firm's subsidiary may have its own EIN number and report its pension plans on a separate Form 5500. If a subsidiary's name (i.e., sponsor's name) is not identical to its parent's name, it is possible that the subsidiary's pension data is not aggregated to its parent firm's data resulting in an underestimate of pension assets, liabilities, and MCs for some firms in our sample.
As discussed above, firms can adopt a hard freeze, a soft freeze, or a partial freeze. The definition of a freeze reported in Form 5500 instructions suggests that it is a hard freeze representing a regime shift in pension obligations that marks the end of the DB era with immediate and significant reductions in pension liabilities. In our sample of frozen DB plans, we find that firms freeze 85% of the number of DB plans, on average, when they begin to restructure DB programs and contributions to the frozen DB plans account for 81% of total DB contributions of the firm in the year preceding the freeze. The ratio of active workers covered by frozen DB plans over the total number of active workers covered is 90%, on average, when a firm first freezes its DB plans. In practice, sponsors frequently supplement frozen DB plans with an increase in contributions to DC plans to (partially) offset the employees' loss of benefits due to the freeze. Thus, a DB plan freeze could reasonably be interpreted as a shift from a DB plan to a DC plan. For this reason, we use the terms DB freeze and DB-DC regime shift interchangeably in this paper.
After carefully matching the pension data, we gather a preliminary sample of 1,610 unique firms that sponsor at least one DB plan from 2001 to 2008, of which 388 firms hard freeze at least one of their DB plans for the first time from 2002 to 2008. Panel A of Table I provides the (annual) temporal distribution of the first hard freeze events and the top 15 industries with DB freezes from 2002 to 2007. The number of firms that first freeze their DB plans increased sharply in 2003, probably due to the market recession in 2000-2001, and remained stable after. Although DB freezes were distributed over almost all industries, they were particularly concentrated in manufacturing, mining, and financial institutions. From this initial set, we keep only those firm-year observations that have nonmissing data for assets, investment, and mandatory pension contributions. This filter yields our final unbalanced panel of 1,071 DB sponsoring firms (with 5,531 firm-year observations) including 179 firms (with 414 postfreeze firm-year observations) that hard freeze at least one DB plan. We label this sample as the DB sample throughout our study, although some firms covered by this sample sponsor both DB and DC plans. Following the hard freeze, we track each firm for roughly three years, on average. It is worth highlighting that the temporal and industrial distribution pattern of this 179 freeze subsample is similar to those of the 388 freeze sample documented earlier.
Because funding status is a key variable in our analysis, we present funding ratios of all DB plans (Panel B) and those of frozen DB plans in the freeze year (Panel C) from 2000 to 2007 in Table I. Funding ratio is measured as the ratio of plan assets to current pension liabilities, and a ratio less than one indicates the underfunded status of the plan. In Panel A, it is noteworthy that the mean (median) funding ratios of DB plans follow a decreasing trend, starting with the overfunded (fully funded) status in 2000 (median = 1.00) and then becoming underfunded for the rest of the period (median = 0.91 in 2007). The mean and median funding ratios of the subsample of frozen plans are below the respective statistics of the full sample of DB plans, with the median ratio dropping from 0.91 in 2002 to 0.88 in 2007. The declining time trend in the funding status indicates that underfunded plans are more likely to be frozen.
Summary statistics for the sample of DB sponsoring firms are reported in Table II. Constructed variables are winsorized at 1% and 99% to remove the effect of outliers that could bias our analysis. Investment is the ratio of capital expenditures to the beginning-of-period book value of assets. Following Rauh (2006), cashflows and nonpension cashflows are calculated as follows:
Cash flows = Net Income + Depreciation and amortization + Pension expenses - Total pension contributions, (1)
Non-pension cash flows = Net Income + Depreciation and amortization + Pension expenses. (2)
Both measures purge the effects of pension smoothing by adding back pension expenses, and nonpension cash flows approximate the operating cash flows (unadjusted for total pension contributions). Tobin's Q is measured as the ratio of (market value of equity + book value of assets--book value of equity--deferred taxes) to the book value of assets. Funding status is the difference between DB plan assets and plan liabilities aggregated to the firm level, then scaled by the beginning-of-period book value of assets.
If a DB pension plan is underfunded in a given year, MCs of the plan are calculated as max (MFC, DRC), where MFC is the minimum funding contribution and DRC is the deficit reduction contribution. Following Munnell and Soto (2004) and Rauh (2006), MFC is approximated as the normal cost (equal to the present value of pension benefits accrued during the year) plus 10% of the previous underfunding status. In our sample period, 2001-2008, DRC is estimated as max (0.30, [0.30 0.40 (funding status - 0.60)]) of the underfunding amount (see Rauh, 2006 for a detailed description of MC calculation). MCs are scaled by the beginning-of-period book value of assets to arrive at the respective ratio.
Some firm characteristics stand out from Table II. The average DB sponsoring firm is large ($17.6 billion in asset value), old (32 years), and profitable (0.03). It has significant debts in its capital structure (book leverage 0.27). It is important to note that the distribution of mandatory pension contributions is heavily skewed to the right with roughly the top quartile of firm-year observations having nonzero outlays. We also run some comparative analysis between the DB and pure DC sponsoring firms and note that the representative DB sponsor makes more than twice the pension contributions per year of pure DC sponsors (total pension contribution ratio of 0.009 vs. 0.004) and the difference is significant at 1% level (results not tabulated for brevity). Overall, these aggregate estimates are consistent with the prior evidence that the financial condition of DB plans is deteriorating and DB plans are more expensive when compared to DC plans.
III. Empirical Analysis
We begin with the univariate analyses of the effects of the regime shift in pensions on MCs, DCs, liquidity (as measured by net working capital), leverage, funding ratio, and capital expenditures. This is followed by multivariate tests of the hard freeze on liquidity, capital structure, investment, and firm value.
A. Univariate Analyses of the DB Freeze Effects
To assess the base level effects of the hard freezes, we conduct univariate analyses of the innovations in the key test and outcome variables over time using the self-adjustment, industry-adjustment, and control group-adjustment for the DB freeze subsample. These freezing firms are likely to encounter varying degrees of financial constraints. The event year, the year in which firms shift from DB to DC, is defined as year t. Self-adjusted values are computed as differences in own value from the base year t-1 (the year preceding the hard freeze year) to the freeze event year and each of the three subsequent years. In addition, we analyze changes in industry-adjusted and control group-adjusted values from the base year. The industry-adjusted values are calculated as the difference between the firm's own value and the contemporaneous median of its industry identified by the two-digit standard industrial classification (SIC) code containing only DB sponsoring firms. We require the industry to have at least five observations in any given year to calculate the industry median. If the two-digit SIC code has less than five observations, we use the one-digit SIC code's median. In a similar fashion, we compare the freezing firm's values with the contemporaneous relevant control group's median. The control group consists of all DB sponsoring firms in the same two-digit SIC code with values in the range of 70% to 130% of that of the freezing firm in the base year. If the control group cannot be formed based on two-digit SIC codes, we step back to one-digit SIC codes instead. We conduct t-tests (Wilcoxon signed rank tests) to test the null hypothesis that the means (medians) of the reported changes from the base year are not different from zero.
Because the distribution of MCs is heavily skewed with positive values only in the top quartile and zeros elsewhere, tests using all firm-year observations on MCs would be downward biased (i.e., biased against finding a significant decline in MCs over the postfreeze years). Therefore, we restrict the analysis to only those firms that have positive values of MCs in the base year and report mean and median changes from the base year in Panel A of Table III. The Wilcoxon signed rank test demonstrates a significant drop in the self-adjusted median MCs from the base to each of the postfreeze years. For example, the median drop in MCs (scaled by book assets) from the base to the third year following the freeze is three basis points (0.0003), which is highly significant. Similarly, the industry-adjusted and control-group adjusted mean and median MCs of the freeze sample register consistent and significant declines ranging from one to four basis points over each of the three years following the freeze.
Panel B reports evidence of a significant increase in alternative measures of DCs in year t and t+1, which is consistent with the idea that freezing firms substitute DC for DB plans. (5) An additional univariate analysis of DB contributions (sum of both MC and VC) indicates a decrease over time (results not tabulated). Next, we examine whether firms with pension deficits preceding the freeze utilize the cost savings to shore up underfunded DB plans. This analysis is complementary to the tests on changes in MCs presented in Panel A as improved funding status leads to lower MCs. From Panel C, we find that the funding ratio shows significant improvement in the postfreeze years. For instance, the industry-adjusted median funding ratio improves by 7.44% in the third year following the freeze. In untabulated analyses, we perform similar comparisons for net working capital, book leverage, and investment and find weak evidence of a decrease in liquidity and an increase in leverage and investment following the freeze (results are available upon request). In sum, the univariate tests suggest that pension freezes are followed by material and significant drops in MCs and improvements in the funding ratio of previously underfunded DB plans over the succeeding three years.
B. Effects of the DB Freeze on Liquidity and Financial Leverage
Although the foregoing univariate analysis provides weak indication of any effect that DB plan freezes have on the liquidity and leverage of firms with underfunded plans, other factors may also affect the change in liquidity and financial leverage and should be controlled for in a multivariate setting. Following Rauh (2006), we use the following baseline specification to test for the effects on liquidity of an exogenous shift in cash flow caused by mandatory DB contributions while controlling for investment opportunities, cash flows, and the lagged value of funding status. (6)
[NWC.sub.it] = [[alpha].sub.i] + [[alpha].sub.t] + [[beta].sub.1][Q.sub.i,t-1] + [[beta].sub.2][NPC.sub.it] + [[beta].sub.3][MC.sub.it] + [[beta].sub.4][FS.sub.it-1] + [[epsilon].sub.it]. (3)
In Equation (3), [NWC.sub.it] is the net working capital (a proxy for liquidity) of firm i in year t, Q is Tobin's Q (a proxy for unobserved investment opportunities), NPC is nonpension cash flows, and FS is funding status (measured as the difference between pension assets and pension liabilities). MC is the key test variable in the equation. We scale all variables except Tobin's Q by the beginning-of-period book value of assets. Even though MCs are a kinked and discontinuous function of the funding status, Rauh (2006) emphasizes the use of FS as a control for the potential endogeneity of MCs with respect to unobserved investment opportunities. The potential measurement errors of Tobin's Q and other unobserved time invariant effects could be a part of the error term ([[epsilon].sub.it]), which may correlate with NPC and MC, thus biasing the coefficient estimates. Therefore, we additionally control for firm ([[alpha].sub.i]) and year ([[alpha].sub.t]) fixed effects (FEs) and use the FEs model to estimate Equation (3). This two-way FEs model allows us to control for unobserved heterogeneity (time invariant firm FEs and time FEs), which affect pension contributions. (7)
We examine the liquidity effects of the DB-DC shift by augmenting the liquidity model in Equation (3) with a DB-DC shift dummy variable, D, and an interaction term D*MC to account for possible changes in both the intercept and the slope of the liquidity model.
[NWC.sub.it] = [alpha] + [[alpha].sub.t] + [[beta].sub.1][MC.sub.it] + [[beta].sub.2][D.sub.it] * [MC.sub.it] + [delta][X.sub.i,t] + [[epsilon].sub.it], (4)
where D takes a value of one in the years following the shift from DB to DC plans, and zero otherwise. X is a vector of other control variables that includes nonpension cash flows (NPC), lagged values of Tobin's Q, and lagged funding status (FS). Franzoni (2009) finds that MCs are significantly and positively correlated with three alternative indices of financial constraints. In our sample, the simple correlation between MCs and the Whited and Wu (2006) index (WW index) of financial constraints is 0.20 and highly significant. That is, firms with high (low) MCs are subject to more (less) financial constraints and higher (lower) costs of external capital. This finding suggests that MCs not only assess the pension contributions-induced shock to internal capital, but also proxy for the degree of firms' exposure to financial constraints. As discussed earlier, we expect [[beta].sub.1] to be negative because MCs constitute negative liquidity shock to firms with financial constraints. Further, if firms facing financial constraints can increase the average level of liquidity (as represented by a positive change in the intercept) by hard freezing DB plans, [[beta].sub.2] should be positive and significant. Alternatively, if the shift from DB to DC plans lowers the demand for net working capital, we would expect this coefficient to be negative. In addition, if the lower and discretionary pension contributions under the DC plans are more beneficial to financially constrained firms when compared with the frozen DB plans, then the interaction effect, [[beta].sub.3], should be positive and significant.
Because firms have greater incentives to restructure their pension policies when they face financial constraints, distress, and/or see attractive investment opportunities, but feel hampered by the cash flow drain caused by their current pension commitments, the DB freeze sample is not a random sample. As the decision to freeze the DB program is endogenous, the regressor, D, will be correlated with the error term in Equation (4) leading to biased ordinary least square (OLS) estimates of [[beta].sub.1] and [[beta].sub.3]. Therefore, we assume that a firm's decision to freeze DB plan(s) is determined by
[D.sup.*.sub.it] = [beta][X.sub.it] + [gamma][Z.sub.it] + [u.sub.it]
[D.sub.it] = 1 if [D.sup.*.sub.it] > 0 (5)
[D.sub.it] = 0 if [D.sup.*.sub.it] [less than or equal to] 0,
where [D.sub.it.sup.*] is a latent variable, [X.sub.it] is a set of variables common to both the investment and probit equations [i.e., Equations (4) and (5)], [Z.sub.it] is a set of explanatory variables that include firm characteristics and/or industry features that can explain the decision to freeze DB plans, but do not belong to the investment equation, and [[mu].sub.it] is an error term. The correlation between [D.sub.it] and the [[epsilon].sub.it] in Equation (4) will be nonzero if some of the exogenous firm characteristics in [Z.sub.it] that influence the freeze decision captured by Equation (5) affect the firm's investment as well, but are not included in the investment equation or the error terms [[epsilon].sub.it] and [u.sub.it] in Equations (4) and (5), respectively, are correlated. In both cases, the OLS estimates of [[beta].sub.2] and [[beta].sub.3] will be biased. We use three different techniques to correct for the potential endogeneity bias due to the correlation between [D.sub.it] and the [[epsilon].sub.it] and obtain unbiased estimates of [[beta].sub.2] and [[beta].sub.3]. First, assuming that the unobserved effects are time invariant, we apply the FEs model to the panel data to estimate Equation (4). Additionally, because pension sponsors choose whether to freeze or not to freeze an existing DB plan, we use the Heckman (1979) two-step correction model to address the potential self-selection bias. Specifically, in the first step, we estimate a probit model of Equation (5) to get consistent estimates of [beta]s and [gamma]s. The resulting consistent estimates of [beta]s and [gamma]s are then used to estimate the inverse Mill's ratio (i.e., [lambda]) as follows:
If D = 1, then [lambda] = [phi](X[beta] + Z[gamma])/[PHI](X[beta] + Z[gamma]), (6a)
If D = 0, then [lambda] = -[phi](X[beta] + Z[gamma])/(1 - [PHI](X[beta] + Z[gamma]), (6b)
where [phi] is the probability density function and [PHI] is the cumulative density function.
In the second step, we estimate the liquidity Equation (4) augmented with [lambda] assuming that the error terms ([[epsilon].sub.it], [u.sub.it]) in Equations (4) and (5), respectively, follow a bivariate normal distribution. Moreover, it is plausible that liquidity and the DB freeze are jointly determined. That is, when firms believe that the growing burden of DB contributions is draining their liquidity and hampering their ability to undertake value-enhancing projects, they resort to pension freezes. To address this endogeneity concern, we use IV regression to estimate the investment equation. Note that the endogenous variable, D, is a dummy variable. Because the natural instruments for the DB-DC shift dummy variable are also those included in the investment equation, we are careful to select valid instruments [i.e., variables in the set Z in Equation (5)] that highly correlate with the DB-DC shift decision, but that do not belong to the investment equation. We follow a procedure suggested by Wooldridge (2002) (similar to that used by Campa and Kedia, 2002) in estimating the IV regression with the endogenous dummy variable via two-stage least squares (2SLS). The detailed estimation procedure is included in the Online Appendix A4. As in the case of 2SLS, the generated variance-covariance matrix is incorrect as the estimate of the error term variance is computed using the residuals calculated with predicted [D.sub.it] instead of actual [D.sub.it]. To correct standard errors for potential correlation of residuals across firms and across time, we follow Shivdasani and Stafanescu (2010) to use bootstrapped standard errors for statistical inference.
As the first step in the Heckman (1979) two-step correction approach, we use a probit model to estimate the probability of freezing DB plans as a function of the beginning-of-period firm characteristics, contemporary industry features, and year effects. In addition to the exogenous variables from the investment equation, we identify a number of firm characteristics and industry features that are used in the literature to predict the probability of a firm to sponsor or freeze a DB plan (Munnell and Soto, 2007; Beaudoin et al., 2010; Shivdasani and Stefanescu, 2010, among others). The probit model has the following form:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (7)
The dependent variable is the DB freeze dummy as defined earlier. Other variable definitions are provided in the Appendix. Expected signs on the regression coefficients based on the prior literature are reported in the first column and the probit estimates with firm and year FEs can be found in Column (2) of Table IV.
It is not straightforward to interpret the coefficients of the probit model, so we report the marginal effects of an increase of a one-standard deviation in the relevant variable above its sample mean on the DB-DC shift probability in Column (3) of Table IV. The likelihood ratio (LR) Chi-square (425.56, not tabulated) rejects the null hypothesis that the coefficient estimates are jointly zero. The pseudo R-square (18%) indicates that the variables used in the probit model have significant predictive power of the DB-DC shifting probability. Most of the coefficient estimates have the expected signs and those on Active Participant Ratio, Firm Age, Industry's DB Ratio, and Industry's Unionization Rate are negative and statistically significant. Relating firm characteristics to the DB-DC shifting probability, we note that a one-standard deviation increase above the sample mean in the Active Participant Ratio (Firm Age) decreases the freeze probability by 2.84% (2.07%). With respect to industry features, labor unions seem to hold great bargaining power because the higher the industry's unionization rate, the less likely the firm shifts from DB to DC plans (the related marginal effect is -0.91%). In addition, the higher the ratio of firms with DB plans in an industry, the lower the probability that a firm in that industry freezes DB plans (marginal probability of -0.94%). The coefficients on funding status and MC are not significant in this probit specification probably due to their collinearity. When we include either of them, while controlling for other firm and industry factors in the probit model, they have the expected signs and are statistically significant.
The net working capital (current assets minus current liabilities, scaled by the beginning-of-period book value of assets) regression results are reported in Panel A of Table V. The first two columns present firm FEs estimates and the next two columns report the Heckman (1979) two-step correction and IV regression estimates, respectively. Column (1) is the baseline regression, Columns (2)-(4) are baseline specifications augmented with the DB-DC shift dummy and its interaction with MC. The Hausman (1979) tests of FEs versus random effects validate the use of the FEs model in Columns (1) and (2). In Column (3), the Wald test results firmly reject the null hypothesis that the probit equation and the liquidity equation are independent, thus supporting the use of the Heckman (1979) two-step correction model to address endogeneity of the freeze decision. We are careful in selecting instruments for the IV regression in Column (4) that meet the over-identifying restriction requirement, so we conduct a battery of over-identification tests and finally select Lag Active Participant Ratio and Industry DB Ratio as instruments for D. The Hausman (1979) test of endogeneity in Column (4) rejects the null hypothesis of no difference between the 2SLS and OLS estimates, justifies endogeneity concerns about the DB-DC shift, and supports the use of IV regressions to obtain unbiased estimates. The chi-square statistic indicates that the coefficient estimates in the second stage of the IV regression are jointly significant. The test of weak instruments (robust F-stat = 52.35) firmly rejects the null hypothesis of weak instruments, and the isolated power of the generated instrument in explaining the DB-DC shift, after controlling for other variables, is fairly good (partial R-square = 3%). The coefficients on MC across the models are negative and significant at the 1% level. The magnitude of the coefficient estimates suggest that, on average, a one dollar increase in MCs before the freeze drains net working capital by $1.02-$1.43. It is worth pointing out that these estimates suggest a far stronger negative impact of required contributions on corporate liquidity when compared with the insignificant effects documented by Rauh (2006) for the 1990s. The coefficients on D are not statistically different from zero. However, in untabulated results using alternative specifications with smaller sample sizes, we find that the coefficient on D is negative and significant, suggesting that the shift from DB to DC tends to reduce the average level of liquidity. The coefficients on the interaction term MC*D are all positive (ranging from 0.848 to 1.575) and statistically significant. These interaction test results suggest that for firms facing financial constraints, the DB freeze tends to neutralize the negative effect of MCs on net working capital over the following three years, which is consistent with our expectations.
Panel B replicates the above net working capital regression of the firm FE model for two subsamples, high (Hi-FC) and low (Lo-FC) financial constraints. The Hi-FC subsample includes firm-year observations with WW index values above the sample median index value in a given year. The Lo-FC subsample includes the remaining observations. The coefficients of control variables are suppressed for brevity. As expected, the estimated coefficient on MCs is significant and negative only in the Hi-FC subsample, but not for the Lo-FC peers. Similarly, the interaction term, D*MC, is significant and positive, but only in the Hi-FC subsample. However, the differences between the two sets of coefficient estimates are not statistically significant, as indicated by the p-value (Hi-FC-Lo-FC). Further, the coefficient estimates on D are negative and often significant for both subsamples, but the differences in estimates between the two subsamples are insignificant. In summary, these findings suggest that MCs have a depressing effect on the liquidity of firms facing financial constraints. The pension freeze tends to neutralize the negative liquidity effects over the following three years. In addition, we find weak evidence regarding a drop in the average level of liquidity following the pension freeze.
To examine the effect of the DB freeze on the linkage between financial leverage and MCs, we modify the net working capital MC specification by using financial leverage as the outcome variable and adding relevant regressors selected form the capital structure choice literature. Specifically, the leverage ratio is the ratio of book debt to the book value of assets (i.e., book leverage) or of book debt to the sum of book debt and the market value of equity (i.e., market leverage). Control variables include lagged Tobin's Q, NPC, lagged FS, asset tangibility, year FEs, industry FEs (or firm FEs) and size. We report the book leverage regression results in Panel A of Table VI (market leverage regression produces qualitatively that control for industry FEs or firm FEs, respectively, but do not include the DB-DC shift dummy. Columns (3)-(6) include the DB-DC shift dummy and its interactions with MC. Column (3) controls for industry FEs, Column (4) controls for firm FEs, Columns (5) and (6) are the second stage of the Heckman (1979) self-selection model and IV regression, respectively. The Hausman tests (in Columns (2), (4), and (6)) and the Wald test (in Column (5)) validate the use of the respective models. The test of weak instruments in Column (6) validates the relevance of the selected instruments in the IV regression. The estimates of all conventional control variables are consistent with those documented in the literature and are suppressed for brevity. The coefficients on MCs are positive, varying from 0.157 to 1.181, and significant in five out of six specifications, consistent with the notion that before the freeze, firms with underfunded DB plans borrow to fund their pension promises. In sharp contrast, the estimates on the interaction term MC*D are all negative, ranging from -2.779 to -0.107, and highly significant in the Heckman (1979) correction and IV regression models. For example, the IV coefficient estimates suggest that a one dollar increase in MCs before the freeze increases book leverage by $0.419, on average, but this marginal effect diminishes by $1.025 following the pension freeze. These results imply that financially constrained firms (i.e., those with high MCs) rely significantly less on borrowing to fund their required pension obligations after the freeze. Alternatively, the coefficients on D are positive (from 0.02 to 0.182) and significant suggesting an increase in the general level of book leverage after the freeze.
Panel B replicates the above leverage regression model with industry and year FEs, but no firm FEs for two subsamples, Hi-FC and Lo-FC. The Hi-FC subsample includes firm year observations with WW index values above the sample median index value in a given year. The Lo-FC subsample provides the remaining observations. The coefficients of control variables are suppressed for brevity. The slope estimates indicate that, on average, Hi-FC firms borrow significantly more than Lo-FC firms to pay for MCs before the pension freeze. However, the highly significant interaction terms suggest that after the freeze, Hi-FC firms use significantly less leverage than their Lo-FC counterparts to fund MCs. In addition, the estimated coefficients of D (i.e., intercept changes) suggest that Hi-FC firms use a significantly higher level of debt than Lo-FC firms following a pension freeze.
In summary, the above tests suggest that financially constrained firms tend to increase the average level of leverage after freezing DB plans supporting our conjecture that the shift from DB to DC plans increases the incentive to borrow to recoup lost tax subsidies and to take advantage of lower borrowing rates following the transfer of pension risk to employees. In addition, the interaction tests suggest that constrained firms resort to significantly less leverage to pay for MCs following a pension freeze. This is consistent with the notion that the hard freeze is likely triggered by the poor financial health of sponsors making it difficult for constrained firms to borrow to fund MCs over the subsequent years despite the reduced pension burden, diminished pension risk, and lower interest rates.
C. DB Freeze and Investment
Table VII reports the investment regression results with scaled annual investment as the dependent variable. Columns (1) and (2) include firm FEs, whereas Columns (3) and (4) are the second stage of the Heckman (1979) self-selection and IV regression models, respectively. Following Rauh (2006), Column (1) reports the baseline regression without the DB-DC shift dummy. Columns (2)-(4) include the DB-DC shift dummy and its interactions with MC. The Hausman tests (Columns (1), (2), and (4)) and the Wald test (Column (3)) support the use of the respective models. The test of weak instruments in Column (6) indicates that the selected instruments in the IV regression are valid.
The coefficients on all conventional control variables (NPC and Lagged Q) across all specifications have the signs and significance levels consistent with those reported in the literature and are suppressed for brevity. The coefficients on MCs are negative (ranging from -0.209 to -0.045) and statistically significant at the 5% and 1% levels across all specifications. They indicate that a dollar increase in required pension contributions lowers capital expenditures by $0.05-$0.21, on average. This evidence is consistent with the findings of Rauh (2006) that MCs have a negative impact on the investment of constrained firms, although our MC coefficient estimates are far smaller than the typical 0.60 in absolute value reported by Rauh (2006). A possible explanation could be that firms are less constrained and/or external financing is more accessible in the 2000s than in the 1990s.
The coefficients on D are negative (ranging from -0.024 to -0.021) and significant in the Heckman (1979) self-selection correction and IV regression models. These estimates indicate that the average level of investment drops over the three years following the freeze, perhaps suggesting that the cost savings are not large enough to offset the adverse investment effects due to the deteriorating pension funding status. However, in untabulated results using alternative specifications with smaller sample sizes (discussed in the robustness checks section below), we find that the coefficient on D is positive and significant suggesting that the shift from DB to DC tends to raise the average level of capital expenditures. In contrast, coefficients on the interaction term MC*D are positive, varying from 0.035 to 0.301. Yet they are only marginally significant in the Heckman (1979) self-selection correction model (Column (3)) indicating weak evidence that the freeze attenuates the harmful investment effects of MCs for firms with financial constraints. The coefficient on [lambda] is positive (0.008) and highly significant implying that the firm's private information, which is likely to affect the DB freeze decision, is positively correlated with investment.
Rauh (2006) reports that the negative investment effects of MCs are particularly pronounced among firms facing financial constraints. To further investigate whether the pension freeze is especially beneficial to financially constrained sponsors, we construct two subsamples: 1) Hi-FC firms that rank above the median on the WW financial constraints index in a given year in the full DB sample and 2) Lo-FC firms which include the rest. (8) Then, we rerun the FE, Heckman (1979) two-step correction, and IV regressions. The estimates, reported in Tables A1 and A2 in the On-line Appendix, confirm that the negative investment effects of MCs are significant, but only for the Hi-FC group. Further, the lack of significance of the interaction term (MC*D) suggests that the adverse impact of MCs on investment persists following the freeze in the constrained subsample.
Overall, our estimates, derived from a sample of DB pension plans drawn from 2001 to 2008, suggest that the negative marginal effect of statutory pension contributions on investment is significantly smaller than that reported by Rauh (2006) for the 1990s. Moreover, we uncover little evidence to suggest that the negative effect of MCs on investment is mitigated over three years following the regime shift from DB to DC plans. (9)
D. Robustness Checks
DB plan funding status may have asymmetric effects on liquidity, financial leverage, and investment because an arbitrary level of underfunding status triggers MCs under the pension rules. To account for the potential asymmetric effects of the funding status, we split our sample into two subsamples covering firms with overfunded (i.e., funding status is greater than or equal to zero) and underfunded plans (where funding status less than zero) and re-estimate the above multivariate regressions separately for the two subsamples. However, controlling for the asymmetric effects of the funding status does not appear to materially change the results. We also account for the nonlinear effects of the funding status by augmenting the multivariate regressions with the squared and cubic terms of the funding status, but our results are qualitatively similar. Moreover, it is possible that the effects of MCs and the DB plan freeze are concentrated in the underfunded subsample. Therefore, we re-estimate the regressions with only those firms with underfunded plans, but, again, the results are essentially unchanged (results are not tabulated for brevity, but available upon request).
When firms freeze DB plans, they usually offer DC plans as replacement or enhance existing DC plans. To account for the possible joint effects of a decrease in DB plan contributions and an increase in DC plan contributions on firm performance, we re-estimate our liquidity, financial leverage, and investment regressions that are augmented with DC plan contributions. It is noteworthy that DC, due to its flexibility feature, is more likely to be an endogenous regressor. As such, we use each of the following alternative measures as a proxy for DC in our regressions: 1) the lagged DC, 2) the lagged change in DC, and 3) the contemporaneous two-digit SIC industry median DC. Our estimation results indicate that DC does not have a significant impact on liquidity, financial leverage, or investment. Our earlier findings are qualitatively unchanged (results are not reported to save space, but are available upon request).
Rauh (2009) suggests that the funding status of pension plans reflects the financial health of the plan sponsors. Therefore, we use the DB plan funding status to control for a firm's financial condition in the probit model (i.e., Equation (7)). However, to address a possible concern that the funding status cannot fully capture the financial condition or distress of a DB sponsoring firm, we augment our probit model with the Altman Z-score and re-estimate it. We then use this augmented probit specification in the self-selection correction models of liquidity, financial leverage, and investment and find that although the sample size for estimation decreases due to missing data of some variables used in the Altman Z-score calculation, our results are basically unchanged (results are available upon request). (10)
Because we gauge the impact of the DB plan freeze on firm performance over a subsequent period of about three years, there is a concern that DB freeze firms may embark on other corporate restructuring activities or expansion programs in our postfreeze period. To address concerns about potentially confounding firm-specific events, we conduct univariate analyses of the change in employee growth rates (to account for possible labor layoffs), mergers and acquisition (M&A), and research and development expenditures (R&D) and report the results in Table VIII. In Columns (1)-(4) of Panel A, we compare a DB freeze firm's employee growth rates (measured as (number of employees in the current year number of employees in the prior year)/number of employees in the prior year) in the freeze year and in each of the three years following the freeze with the growth rate in the year preceding the freeze. In Column (5), we compare the average of the annual employee growth rates in three years postfreeze with the average over three years prefreeze. The t-tests (of the means) and Wilcoxon signed-rank tests (of the medians) indicate that the employee growth rates either increase (Columns (2)-(4)) relative to the base year or do not change significantly (Columns (1) and (5)).
We collect M&A data from Thomson One Banker for the analysis reported in Panel B of Table VIII. Because M&As are big and lumpy investments, we analyze the changes in the average of the scaled M&A transaction values for the DB freeze firms within three years before the freeze and three years after the freeze. Firms can make payment for M&A deals in cash, stocks, or both. To ensure the robustness of our results, we scale M&A transaction values by the lagged book value of assets and lagged market value of assets (the sum of book value of debt and lagged market value of equity). Our analysis sample includes all DB freeze firms regardless as to whether they did or did not engage in M&As over the prefreeze and/or postfreeze periods. We fail to find significant changes in the M&A deal values from the prefreeze to the postfreeze period. It is possible that firms that had not engaged in any M&A in the prefreeze period belong to a subgroup of those that never conduct M&As and including them in the sample would bias the results toward zero. To address this concern, in an additional analysis, we focus on a subsample of firms that conduct M&As in both the prefreeze and postfreeze periods, but do not find any significant changes in M&A deal values following the freeze (results not reported for brevity).
Our analysis of R&D expenditures is reported in Panel C of Table VIII. The results indicate no evidence of a change in R&D following the freeze. In addition to the univariate analyses tabulated in Table VIII, we conduct similar analyses on net debt issuance, net equity issuance, stock repurchase, asset sales, return of assets (ROA), and returns on equity (ROE) of DB freeze firms, but fail to find significant changes except some improvement in ROA and ROE following the freeze (the results of these additional analyses are not reported to save space, but are available upon request). The changes in ROA and ROE are consistent with our documented positive short-term announcement abnormal returns. (11)
In short, although we cannot exclude the possibility that our earlier findings could be somewhat confounded, in part by the combined effects of several small/insignificant changes in firm performance following the DB plan freeze, our combined univariate and multivariate results suggest that the regime shift from DB to DC plans itself is more likely the driver of our results in the postfreeze period.
IV. Abnormal Returns Following the Pension Freeze
The above analyses suggest that the DB freeze is partially beneficial to the sponsoring firm as it tends to neutralize the negative effects of MCs on net working capital while improving financial leverage. However, the freeze may have an adverse impact on employee morale posing a real challenge to the firm in maintaining productivity and attracting and retaining talent, resulting in possible adverse long-term financial consequences. Therefore, we are interested in assessing the net verdict of the market on the decision to freeze a DB pension plan. Franzoni and Marin (2006) argue that the market is slow in impounding the effect of pension-related information into stock price, and Franzoni (2009) documents a negative impact of MCs on long run abnormal returns from 1990 to 1998. Further, it is likely that the value effects (arising primarily from the investment effects) of a hard freeze would require several years to materialize. Therefore, we investigate both the short run and long run implications of the DB-DC shift on stock returns in this section.
A. Pension Freeze Announcement Abnormal Returns
Form 5500 does not provide DB freeze announcement dates that can be used for the short run event study, so we hand-collected the news about the freeze from The Wall Street Journal, the Lexis-Nexis database, websites of the Center for Retirement Research at Boston College and Pension Rights Center, and websites of the related firms. We check each freeze announcement carefully to ensure that those DB freeze announcements included in the announcement sample are stand-alone events and free from other potential confounding effects. We consider two events relating to the DB freeze: 1) the announcement of the DB freeze decision, and 2) when the freeze becomes effective. For some firms, both events fall on the same date, but for others the effective date is usually a few months following the announcement. If the market is efficient and there is less uncertainty about the freeze after its announcement, most of the effect should be recorded on the announcement date. We are able to locate announcement dates for 82 firms, of which 67 firms have sufficient daily stock return data to run the event study. We obtain daily stock returns from Center for Research in Security Prices (CRSP), the Fama-French (1993) factors from Kenneth French's website, and other firm-specific annual financial data from Compustat. In comparison with the average DB freeze sample, the event study sample firms are about three times larger in asset value ($53.7 billion vs. $17.6 billion), but comparable in most other characteristics (descriptive statistics for the event study are not reported for brevity).
We use the market model to estimate the abnormal returns (AR) and cumulative abnormal returns (CAR) during the event windows. The day on which the firm announces a DB-DC shift is the event day (t = 0) (or the effective date is the event day if it is considered instead). The estimation period starts 255 days and ends 46 days before the event day. In line with the literature, we consider CAR for the (0, +1), (-1, 0), and (-1, +1) windows. Taking into consideration potential information leakage before the official announcement of the DB-DC shift and under reaction of the market, we further consider other windows such as (-30, -2), (-10, -2), (+2, +5), and (+2, +10).
Panel A of Table IX reports the event study results that are estimated by the market model using the CRSP equally weighted index and the DB freeze announcement subsample. The CAR during the event window (0, +1) is positive (0.36%) and statistically significant at the 1% level. The CAR over the window (-1, +1) is also positive (0.35%) and significant at the 5% level, but the CAR over the window (-1, 0) is not significant. The CAR in the (-30, -2) window is negative and significant, whereas CARs in other windows before and after the announcement are either negative or not significant. The positive CAR in the event window (0, +1) supports the hypothesis of value creation from the DB-DC shift. The economic significance of the CAR figure is important. Given the average market value of equity of $53.7 billion of the announcement firms in the sample, a CAR of 0.36% in the (0, +1) window is equivalent to an approximately $196 million increase in market value. For a robustness check, we replace the CRSP equally weighted index with the CRSP value-weighted index, but the results still hold. Alternatively, we use the Fama-French (1993) three-factor model to estimate CARs of the announcement subsample. CARs during the event window (0, +1) continue to be positive and statistically significant. In addition, to address the issue of clustered events, we use the Seemingly Related Regression (SUR) to estimate the abnormal returns but the results are qualitatively similar (not reported to save space). CARs results using the effective DB-DC shift dates for event windows (0, +1) or (-1, +1) are negative, but not statistically significant (not reported for brevity) consistent with the argument that the market incorporates the DB freeze effect immediately when the freeze is announced.
We perform a cross-sectional regression to scrutinize the correlation between lagged MC and other firm characteristics with CARs, and use the Heckman (1979) two-step model to correct for the potential self selection bias of the freeze decision. Independent variables include MC, Funding Status, Size, Book-to-Market Ratio, Stock Price Run-Up, financial constraints (Hi-FC) dummy, and Corporate Governance Dummies (Hi GOV). All right-hand side variables, except stock price run-up, are measured at the end of December of the year preceding the freeze. MC and Funding Status are scaled by the market value of equity. The cross sectional regression is estimated by OLS with robust standard errors with year FEs, and the results are reported in Panel B of Table IX. The small number of available observations limits the explanatory power of the regression, but at least, as expected, the coefficients on (lagged) MC are all positive and significant in Columns (1), (3), (4), and (5) indicating that the freeze-induced decrease in required contributions and the improvements in liquidity and leverage created by the DB freeze are particularly valuable to firms that incur high MCs before the freeze. The coefficients on [lambda] are not significant in any specification suggesting that the self-selection bias has little effect on CARs. Moreover, the results in Table IX are qualitatively unchanged when we use one-step OLS estimations of the cross sectional regression (i.e., without including [lambda] in the cross sectional regression).
B. Long Run Abnormal Returns
We follow Franzoni (2009) to investigate the effect of MC on stock returns in a long run event study, multivariate regression framework. The dependent variable is long run abnormal returns (LAR) which is measured as either annual abnormal returns calculated by the market model or annual industry-adjusted returns. Variables are constructed as in Franzoni (2009). Specifically, annual stock returns are the compound returns computed from monthly stock returns from the beginning of July of year t+1 to the end of June of year t+2. The market beta is estimated using the preceding 36 months of stock returns using the CRSP value-weighted index as a proxy for market returns. Industry-adjusted returns are the deviation of stock returns from the median returns of the same two-digit SIC code industry in the same year. MC is the test variable in the regression. Besides the funding ratio, we follow the relevant literature and include momentum, size, and book-to-market ratio, and a set of year dummies (the market model abnormal return regressions, additionally include a set of industry dummies using two-digit SIC codes) as control variables. (12) All variables are measured in December of year t except Momentum and Size, which are measured in June of year t+1. In this setting, it is natural to scale key variables by the market value of equity, so we scale MC and funding status of year t by the market value of equity in June of year t+1.
We estimate the model with FE, the Heckman (1979) two-step correction, and IV regressions. The FE regression results based on the industry-adjusted market model LARs are reported in Table X (the Heckman (1979) two-step correction and IV regression results are not tabulated to save space). The Hausman test (not reported) rejects the null hypothesis of no difference between the FE and OLS estimates suggesting that the FE estimates are consistent. The coefficients on MC are -0.21 and -0.17 and statistically significant at 5% and 1% levels, respectively. Because MC is scaled by the market value of equity, the coefficient estimates indicate that a one dollar increase in MC reduces firm value by $0.17-$0.21. These results are consistent with those reported by Franzoni (2009) for the 1990s, but are far smaller than the average drop in value of $0.99 he documents. The coefficient estimates on the DB-DC shift dummy are not statistically significant. The coefficient on the interaction term, MC*D, is positive (0.298 and 0.136), but insignificant. In a separate analysis, we estimate LARs of the DB freeze subsample in an event study framework using the calendar time portfolio method (Fama, 1998) and buy-and-hold abnormal returns method (Barber and Lyon, 1997; Lyon, Barber, and Tsai, 1999). Our estimates indicate that DB freeze firms have marginally positive LAR for one-year and two-year periods following the freeze (results are not reported for brevity).
Moreover, we test the conjecture that, all else being equal, the reduced impact of MC on LAR will be more pronounced for financially constrained firms following the DB freeze. As we did with corporate investment earlier, we further divide the samples into Hi-FC and Lo-FC subsamples and run regressions of LAR on MC, D, MC*D, and other control variables for each subsample using FE, the Heckman (1979) two-step correction, and IV regressions. However, we do not find reliable evidence to suggest that the freeze mitigates the negative impact of MC on LAR for Hi-FC firms following the freeze (results are reported in Table A3 of the On-line Appendix).
In summary, we find evidence suggesting the announcement of a DB freeze creates value for firms in the short-term, but the negative effect of MC on firm value does not appear to be alleviated over three years following the freeze. These results seem to be somewhat contradictory and puzzling. A possible explanation for the positive announcement abnormal returns is that the freeze announcement attracts the attention of investors who immediately embrace the visible positive impact on liquidity, financial leverage, and the funding status of DB plans. However, the market may underestimate the depth of financial constraints or distress that triggered the freeze or other long-term adverse effects on the human capital. This line of reasoning is consistent with those in Coronado, Coronado, and Sharpe (2003) and Franzoni and Marin (2006) who suggest that the market is slow in impounding pension information into value, possibly due to a lack of transparency in pension accounting and/or the practice of using pension items to manipulate earnings (Bergstresser, Desai, and Rauh, 2006).
V. Summary and Concluding Remarks
Using a recent and large data set, we present evidence to indicate that required contributions under the DBs pension plans have a negative impact on corporate liquidity, investment, and firm value and an upward pressure on financial leverage. In our full sample, covering both frozen and nonfrozen DB plans from 2001 to 2008, our point estimates (corrected for endogeneity bias) indicate that before the pension freeze, a one dollar increase in MCs reduces liquidity by $1.203, increases debt by $0.419, reduces capital expenditures by $0.209, and depresses shareholder value by $0.213. As compared with a $0.60 reduction in investment reported by Rauh (2006) and $0.99 drop in value reported by Franzoni (2009) based on a 1990-1998 sample, these point estimates represent a substantial decline in the marginal negative investment and firm value effects, despite the sharp market declines in recent years. Our evidence regarding the diminished investment and value effects is suggestive of management's success in formulating new DB pension schemes and reshaping existing ones such that they are more harmonious with corporate policy on investment and shareholder value. Nonetheless, even the diminished investment effects are not inconsequential in light of the average positive investment cash flow coefficient of 6% for firms sponsoring DB plans.
In addition, we find that, on average, the freeze neutralizes the drain on corporate liquidity, relieves the pressure to borrow to pay for MCs, and leads to an increase in financial leverage, particularly for firms facing financial constraints. However, we find little evidence to indicate that hard-freezing DB plans and replacing them with DC plans can help mitigate the negative effects of MCs on the investment of financially constrained firms up to three years after the freeze. These findings are robust to alternative corrections for endogeneity of the freeze decision. Finally, the stock market seems to favor the shift from DB to DC plans as evidenced by the positive short run abnormal returns on freeze announcements, although the long-term effects of MCs on firm value continue to persist unabated up to three years following the freeze.
Overall, our findings suggest that the benefits of freezing DB plans to financially constrained firms are primarily in the form of financial rather than real (investment) effects and are concentrated in the improved funding status of underfunded pension plans, diminished drain on liquidity, and debt relief. The recent sustained market declines and increased volatility have aggravated the funding status of many DB plans, posing a real threat to their viability. We believe the comprehensive analysis of the impact of a regime shift in pensions presented in this study serves to inform the ongoing debate regarding optimal corporate policy on pension plans and regulatory policy with respect to labor and welfare.
Appendix: Variable Definitions Variable Definition Cash flows Net Income + Depreciation and amortization + Pension expenses - Total pension contributions. Nonpension Net Income + Depreciation and amortization + cash flows Pension expenses. Tobin's Q (Market value of equity + book value of assets - book value of equity - deferred taxes)/book value of assets. Funding status (Firm-level plan assets - Firm-level plan (FS) liabilities)/book value of assets. Funding ratio Pension assets/current pension liabilities. Mandatory Max(MFC, DRC), where MFC is the minimum funding contribution contribution and DRC is the deficit reduction ($) contribution. MFC = the normal cost + 10% of previous underfunding status. DRC = max(0.30, [0.30 - 0.40(funding status - 0.60)]) of the underfunding amount. Mandatory Mandatory contribution/book value of assets. contribution ratio (MC) Voluntary DB pension contribution - mandatory contribution. contribution Investment Capital expenditure/book value of assets. Cash flow The standard deviation of the past five-year volatility (EBITDA/book value of assets). Size Ln(book value of assets). (in investment regressions) Active Number of active DB participants/total employees participant of a firm. ratio Firm age Number of years that a firm has been included in the Compustat database. Industry DB Number of firms that sponsor DB plans in a ratio two-digit SIC industry/total number of firms in that two-digit SIC industry. Industry RND Average (research and development investment/the intensity book value of assets) in a two-digit SIC industry. Industry Two-digit SIC industry's unionization rate unionization ([c] 2010 by Barry T. Hirsch and David A. Macpherson rate at Georgia State University). Whited-Wu (WW) = 0.938407 - 0.091CF - 0.062DIV POS + 0.021 TLTD index - 0.044LNTA + 0.1021SG 0.035SG, in which CF is the ratio of cash flow to book value of assets, DIVPOS is a dummy variable equal to one if the firm pays dividends, and zero otherwise, SG is own firm real sales growth, ISG is the three-digit industry sales growth, and LNTA is the natural log of total assets. Net working (Current assets - current liabilities)/book value capital of assets. Book leverage Book value of debtsibook value of assets. Asset (Net property, plant, and equipment)/book value tangibility of assets. Size (in Ln(sales). leverage regression) Size (in CAR Ln(market value of equity). regressions) Size (in LAR Ln(market value of equity in June of year t+1). regression) Stock price Forty-day stock returns during the (-45, -5) run-up window before the DB freeze announcement. Hi Gov Dummy variable equal to one if a firm's Gompers- Ishii-Metrick (Gompers et al., 2003) index is greater than the contemporaneous sample median GIM index, and zero otherwise. Momentum Past 12-month returns. (in LAR regression)
Alderson, M.J. and B.L. Betker, 2009, "Were Internal Capital Markets Affected by the 'Perfect' Pension Storm?" Journal of Corporate Finance 15, 257-271.
Alti, A., 2003, "How Sensitive is Investment to Cash Flow When Financing is Frictionless?" Journal of Finance 58, 707-722.
Almeida, H. and M. Campello, 2010, "Financing Frictions and the Substitution Between Internal and External Funds," Journal of Financial and Quantitative Analysis 45, 589-622.
Almeida, H., M. Campello, and M.S. Weisbach, 2011, "Corporate Financial and Investment Policies When Future Financing is Not Frictionless," Journal of Corporate Finance 17, 675-693.
Bakke, T. and T.M. Whited, 2012, "Threshold Events and Identification: A Study of Cash Shortfalls," Journal of Finance 67, 1083-1111.
Barber, B.M. and J.D. Lyon, 1997, "Detecting Long-Run Abnormal Stock Returns: The Empirical Power and Specification of Test Statistics," Journal of Financial Economics 43, 341-372.
Beaudoin, C., N. Chandar, and E.M. Werner, 2010, "Are Potential Effects of SFAS 158 Associated with Firms' Decisions to Freeze Their Defined Benefit Pension Plans?" Review of Accounting and Finance 9, 424-451.
Bergstresser, D., M. Desai, and J. Rauh, 2006, "Earnings Manipulation, Pension Assumptions and Managerial Investment Decisions," Quarterly Journal of Economics 12 l, 157-195.
Blanchard, O.J., F. Lopez-de-Silanes, and A. Shleifer, 1994, "What Do Firms Do with Cash Windfalls?" Journal of Financial Economics 36, 337-360.
Campa, J.M. and S. Kedia, 2002, "Explaining the Diversification Discount," Journal of Financial Economics 57, 1731-176l.
Coronado, J.L. and S.A. Sharpe, 2003, "Did Pension Plan Accounting Contribute to a Stock Market Bubble?" Brookings Papers on Economic Activity 1, 323-371.
Dasgupta, S., T.N. Noe, and Z. Wang, 2011, "Where Did All the Dollars Go? The Effect of Cash Flows on Capital and Asset Structure," Journal of Financial Quantitative Analysis 46, 1259-1294.
Erikson, T. and T.M. Whited, 2000, "Measurement Error and the Relationship Between Investment and q," Journal of Political Economy 108, 1027-1057.
Fama, E.F., 1998, "Market Efficiency, Long-Term Returns, and Behavioral Finance," Journal of Financial Economics 49, 283-306.
Fama, E.F. and K.R. French, 1993, "Common Risk Factors in the Returns on Stocks and Bonds," Journal of Financial Economics 33, 3-56.
Fazzari, S.M., G.R. Hubbard, and B.C. Petersen, 1988, "Financing Constraints and Corporate Investment," Brookings Paper on Economic Activity 1, 141-195.
Fee, C.E., C.J. Hadlock, and J.R. Pierce, 2009, "Investment, Financing Constraints, and Internal Capital Markets: Evidence from the Advertising Expenditures of Multinational Firms," Review of Financial Studies 22, 2361-2392.
Feldstein, M. and S. Seligma, 1981, "Pension Funding, Share Prices, and National Savings," Journal of Finance 36, 801-824.
Franzoni, F., 2009, "Underinvestment vs. Overinvestment: Evidence from Price Reaction to Pension Contribution," Journal of Financial Economics 92, 491-518.
Franzoni, E and J.M. Marin, 2006, "Pension Plan Funding and Stock Market Efficiency," Journal of Finance 61, 921-956.
Gomes, J., 2001, "Financing Investment," American Economic Review 91, 1263-1285.
Gompers, P., J. Ishii, and A. Metrick, 2003, "Corporate governance and Equity Price," Quarterly Journal of Economics 118, 107-155.
Heckman, J.J, 1979, "Sample Selection as a Specification Error," Econometrica 47, 153-161.
Hill, M., G.W. Kelly, and M. Highfield, 2010, "Net Operating Working Capital Behavior: A First Look," Financial Management 39, 783-805.
Lamont, O., C. Polk, and J. Saa-Requejo, 2001, "Financial Constraints and Stock Returns," Review of Financial Studies 14, 529-554.
Lyon, J.D., B.M. Barber, and C. Tsai, 1999, "Improved Methods for Tests of Long-Run Abnormal Stock Returns," Journal of Finance 54, 165-201.
Marchica, M. and R. Mura, 2010, "Financial Flexibility, Investment Ability, and Firm Value: Evidence from Firms with Spare Debt Capacity," Financial Management 39, 1339-1365.
McFarland, B., G. Pang, and M. Warshawsky, 2009, "Does Freezing a Defined-Benefit Pension Plan Increase Company Value? Empirical Evidence," Financial Analysts Journal 65, 47-59.
Milevsky, M.A. and K. Song, 2010, "Do Markets Like Frozen DB Pensions? An Event Study," Journal of Risk and Insurance 77, 893-909.
Mitchell, M.E. and J.H. Mulherin, 1989, "The Stock Price Response to Pension Terminations and the Relation of Terminations with Corporate Takeovers," Financial Management 18, 41-56.
Munnell, A.H., F. Golub-Sass, M. Soto, and F. Vitagliano, 2006, "Why are Healthy Employers Freezing Their Pensions?" Center for Retirement Research at Boston College, No. 44.
Munnell, A.H. and M. Soto, 2004, "The Outlook for Pension Contributions and Profits in the US" Journal of Pension Economics and Finance 3, 77-97.
Munnell, A. and M. Soto, 2007, "Why are Companies Freezing Their Pensions?" Research report prepared for the 9th Annual Joint Conference of the Retirement Research Consortium "Challenges and Solutions for Retirement Security," Washington, DC.
Petersen, M.A., 1992, "Pension Reversions and Worker-Stockholder Wealth Transfers," Quarterly Journal of Economics 107, 1033-1056.
Pulvino, T. and V. Tarhan, 2006, "Cash Flow Sensitivities with Constraints," Northwestern University, Working Paper.
Rauh, J.D., 2006, "Investment and Financing Constraints: Evidence from the Funding of Corporate Pension Plans," Journal of Finance 61, 33-71.
Rauh, J.D., 2009, "Risk Shifting versus Risk Management: Investment Policy in Corporate Pension Plans," Review of Financial Studies 22, 2687-2733.
Shivdasani, A. and I. Stefanescu, 2010, "How Do Pensions Affect Corporate Capital Structure Decisions?" Review of Financial Studies 23, 1287-1323.
Stewart, G.B., III, 2003, "Pension Roulette: Have You Bet Too Much on Equities?" Harvard Business Review 81, 104-109.
Whited, T.M. and G. Wu, 2006, "Financial Constraints Risk," Review of Financial Studies 19, 531-559.
Wooldridge, J., 2002, Econometric Analysis of Cross Section and Panel Data, Cambridge, MA, The MIT Press.
The following supporting information is available for this article:
Table A1. Investment and Financial Constraints-FE Model.
Table A2. Investment and Financial Constraints--Heckman (1979) Two-Step Correction and IV Regressions.
Table A3. Long Run Abnormal Returns and Financial Constraints-FE Model.
Appendix A4: Heckman Self-Selection Correction and IV Regression Estimation.
Supporting Information may be found in the online version of this article.
Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.
(1) Although Milevsky and Song (2010) report positive pooled cumulative announcement abnormal returns, none of these abnormal returns are statistically significant (see Panel A of Table III).
(2) We thank an anonymous referee for suggesting a possible negative impact of the DB plan freeze on the level of liquidity. This argument is also similar in spirit to Hill, Kelly, and Highfield (2010).
(3) As Munnell and Soto (2007) note, the Employee Retirement Income Security Act of 1974 introduced the maximum funding limit of 100% of actuarial pension liability with respect to tax deductible contributions. The maximum funding limit was significantly tightened under the Omnibus Budget Reconciliation Act of 1987 (OBRA87) by lowering the limit to the lesser of 100% of actuarial liability or 150% of current liability. Current liability is generally less than the actuarial liability as it does not include the effect of future salary increases on the value of pension rights already earned. The Pension Plan Protection Act of 2006 provides that for tax years beginning in 2006 and 2007, the maximum deductible contribution amount for a multi-employer defined benefit plan is 140% of the plan's current liability, minus the value of plan assets. For more information, see http://www.irs.gov/publications/p560/ch04.html#en_US_publink10008917.
(4) We thank an anonymous referee for suggesting this risk-shifting argument. The diminished pension risk would likely lower firms' borrowing rates and result in higher debt levels, which could be used to fund the remaining mandatory contributions and other operating and capital expenditures.
(5) We have a total of 275 firms with frozen DB plans in year t in conducting the univariate tests, but the sample drops to 179 firms in multivariate tests performed later as those tests require observations on all regressors over the postfreeze (i.e., year t+1 onwards) window.
(6) Note that Rauh (2006) uses similar specifications for investment and liquidity regressions. We adopt Rauh's (2006) specifications in this research to maintain consistency and comparability.
(7) Time fixed effects control for the potential correlation between investment and required contributions due to market wide phenomena, such as a recession, and absorb the correlation in error components within a given year. Firm fixed effects control for arbitrary within-firm serial correlation in the error terms. To scrutinize the argument that MCs affect liquidity only through their effects on the overall DB contributions (which cover both the exogenous MCs and endogenous voluntary contributions), we use the instrumental variables (IV) regressions via two-stage least squares. Specifically, in the first stage, we estimate pension contribution as a function of Tobin's Q, nonpension cash flow, mandatory contribution, and funding status. In the second stage, we use the predicted pension contribution in the first stage as an instrument for the pension contribution in the liquidity regression. The result is qualitatively similar, so we do not tabulate the results to save space.
(8) We also try Kaplan-Zingales financial constraints index as used by Lamont and Saa-Requejo (2001) and get qualitatively similar results, but at lower significance levels. Therefore, we report the results based on the WW index and use the financial constraint index for the rest of the paper.
(9) We do not control for firm fixed effects in the main equations of the Heckman (1979) two-step correction model and IV regression model in the tabulated results to avoid potential over specification problems. However, our findings are virtually unchanged if we additionally control for firm fixed effects in the main equations of these models.
(10) Because the Altman Z-score does not meet the exclusionary requirement for an instrument in IV regressions, we do not re-estimate our IV regressions. Alternatively, we attempt to use S&P long-term credit ratings as a proxy for a firm's financial distress, but note that many firm-year observations in our sample do not have credit ratings. Therefore, it is infeasible, in our case, to use S&P credit ratings as a proxy for financial distress.
(11) We thank two anonymous referees for suggesting some of these robustness checks.
(12) Following Franzoni (2009), we include Tobin's Q to control for investment opportunities. Although Tobin's Q is highly correlated with book-to market (B/M) ratio, the results are qualitatively unchanged.
We appreciate the helpful comments from Bill Christie (Editor), three anonymous referees, Assaf Eisdorfer, Carmelo Giaccotto, Joseph Golec, John Harding, Sigitas Karpavicius, Jung-Min Kim, Sanjay Kudrimoti, Alfred Liu, Marcel Prokopczuk, Susan Thorp, as well as session participants at the 2009 International Conference at Iqfai Business School (Bangalore), 2009 and 2010 Financial Management Association International Annual Meetings, 2009 Eastern Finance Association Annual Meeting, 2009 Southwestern Finance Association Annual Meeting, 2011 FINCON at Management Development Institute (Gurgaon). and seminar participants at the University of Connecticut, Massey University, University of Adelaide, and Victoria University of Wellington. A previous version of this paper was titled "Impact of Change in Retirement Benefit Plans on Firm's Investment, Value, and Risk. "All errors remain the sole responsibility of the authors.
Hiev V. Phan and Shantaram P. Hegde *
* Hieu V. Phan is an Assistant Professor of Finance at the University, of Massachusetts at Lowell, MA. Shantaram P. Hegde is a Professor of Finance at the University of Connecticut in Storrs, CT.
Table I. Funding Ratio of DB Plans The table reports the distribution of firms that freeze at least one DB pension plan across time and industries and the funding ratios of DB plans reported in Form 5500 filed with the IRS from 2000 to 2007. The funding ratio is measured as the ratio of pension plan assets to pension plan current liabilities. Year No. of Top 15 DB Freeze Industries No. of Freezes Freezes Panel A. Temporal Distribution of DB Freezes and Top 15 Industries with DB Freezes 2002 3 Depository institutions 37 2003 129 Chemical and allied products 27 2004 56 Industrial and commercial machinery 25 and computer equipment 2005 59 Electronic, electrical equipment, 20 and components 2006 79 Transportation equipment 19 2007 62 Business services 17 Total 388 Food and kindred products 13 Primary metal industry 13 Communications 13 Insurance carriers 12 Apparel and other finished products 11 Furniture and fixtures 11 Fabricated metal products 11 Measuring, analyzing, and controlling 11 instruments; Photographic, medical and optical goods; watches and clocks Oil and gas extraction 10 Panel B. Funding Ratio of All DB Plans Number of Plans Mean 25th Pctl Median 75th Pctl 2000 10,372 1.08 0.88 1.00 1.20 2001 13,105 1.04 0.84 0.95 1.12 2002 12,487 1.05 0.88 0.98 1.14 2003 12,034 0.96 0.80 0.91 1.05 2004 10,341 0.99 0.83 0.94 1.08 2005 11,574 0.94 0.80 0.90 1.03 2006 10,825 0.93 0.80 0.90 1.01 2007 8,598 0.95 0.82 0.91 1.02 Panel C. Funding Ratio of Frozen Plans in the Freeze Year Number of Plans Mean 25th Pctl Median 75th Pctl 2002 183 1.04 0.81 0.91 1.02 2003 709 0.89 0.76 0.86 0.97 2004 358 0.91 0.81 0.90 0.98 2005 478 0.90 0.77 0.87 0.98 2006 407 0.89 0.80 0.86 0.96 2007 346 0.94 0.81 0.88 0.96 Table II. Descriptive Statistics of DB Sponsoring Firms The DB sample covers 1,071 firms that sponsor defined benefit pensions from 2001 to 2008. Means (equally weighted), medians, and percentiles of variables are reported. All ratios are constructed by scaling the associated variables by the beginning- of-period book value of assets. Investment represents annual capital expenditures. Cash Flow is computed as (Net Income + Depreciation & Amortization + Pension Expense - Total Pension Contribution). Nonpension Cash Flow is equal to (Net Income + Depreciation & Amortization + Pension Expense). Leverage is the ratio of the book value of debt to the book value of total assets. Profitability is the ratio of net income over the beginning-of-period book value of assets. Pension Expense Ratio is measured as accounting pension expense divided by the beginning-of-period book value of assets. Tobin's Q is measured as the ratio of the sum (market value of equity + book value of assets - book value of equity - deferred taxes) to the book value of assets. Funding Status is measured as the difference between aggregated pension assets and pension liabilities scaled by the book value of assets. Mandatory Contribution is aggregated to firm level from individual DB plans. Voluntary Contributions represent the dollar amount that a firm contributes to its DB plans on top of MCs and is also aggregated from individual plans. DB Contributions represent the actual dollar amount that a firm contributes to its DB plans, which is equivalent to the sum of MCs and voluntary contributions. The Total Pension Contribution of a DB sponsoring firm aggregates the DB and DC contributions (if any) that a firm makes in a year. Firm Age is approximated by the number of years that the firm is included in the Compustat database. Variable Firm-Year Mean 25th Pctl Observations Assets ($M) 5,531 17,578 549 Investment ratio 5,531 0.046 0.018 Cash flow ratio 5,531 0.083 0.036 Nonpension cash flow 5,531 0.088 0.041 ratio Net working capital ratio 4,789 0.170 0.030 Tobin's Q 5,512 1.569 1.068 Leverage 5,526 0.271 0.138 Profitability 5,531 0.030 0.008 Pension expense ratio 5,531 0.009 0.002 Firm age 5,531 32.007 16.000 Total pension 5,081 49.222 2.517 contributions ($M) Total pension contribution 5,081 0.009 0.002 ratio Funding status 5,531 0.002 -0.007 Mandatory contributions 5,531 6.501 0.000 ($M) Mandatory contribution 5,531 0.002 0.000 ratio DB contribution ratio 5,531 0.005 0.000 DC contribution ratio 5,081 0.003 0.001 Variable Median 75th Pctl Std Dev Assets ($M) 1,986 7,633 90,587 Investment ratio 0.034 0.060 0.044 Cash flow ratio 0.079 0.125 0.105 Nonpension cash flow 0.085 0.132 0.083 ratio Net working capital ratio 0.149 0.284 0.226 Tobin's Q 1.292 1.739 0.926 Leverage 0.247 0.369 0.194 Profitability 0.034 0.068 0.083 Pension expense ratio 0.006 0.012 0.015 Firm age 34.000 47.000 16.900 Total pension 8.685 30.847 290.021 contributions ($M) Total pension contribution 0.005 0.011 0.022 ratio Funding status 0.000 0.007 0.039 Mandatory contributions 0.000 0.515 62.773 ($M) Mandatory contribution 0.000 0.001 0.014 ratio DB contribution ratio 0.001 0.007 0.009 DC contribution ratio 0.002 0.004 0.004 Table III. Univariate Analysis of Pension Freeze Effects The table presents univariate analysis of changes in self- adjusted, industry-adjusted, and control group-adjusted mandatory contributions (MC), defined contributions (DC), and funding ratio from the prefreeze year to three years following the freeze for firms that freeze one or more DB plans from 2001 to 2008. Variable definitions are presented in the appendix. The freeze year is defined as year t. Reported values are changes in the key variables from the base year t-1. Industry-adjusted (control group-adjusted) value is calculated as the difference between a firm's value and the contemporaneous median of the two-digit SIC code firms (control group). The control group consists of firms in the same industry with the values of respective variables in the range of 70% to 130% of the corresponding values of the DB freeze firm in year t-1. In Panel A, DB freeze firms, as well as firms in the same industry or control group, are required to have positive MCs in year t-1. Statistical inference is based on the t-test for means and Wilcoxon signed-rank test for medians. N denotes the number of firms. Self-Adjusted Change t t+1 t+2 t+3 Panel A. MCs of DB Freeze Subsample with MCs > 0 in Year t-1 Mean 0.0016 -0.0012 -0.001 -0.0016 p-value 0.12 0.24 0.37 0.07 Median 0.0000 -0.0001 -0.0001 -0.0003 p-value 0.16 <.0001 0.00 0.01 N 112 80 62 43 Panel B. DC of DB Freeze Subsample Mean 0.0000 0.0004 0.0003 0.0001 p-value 0.99 0.03 0.32 0.74 Median 0.0000 0.0001 0.0000 0.0000 p-value 0.73 0.09 0.38 0.81 N 214 168 116 73 Panel C. Funding Ratio of PreFreeze Undef funded DB Firms Mean 0.0005 0.0270 0.0398 0.0508 p-value 0.95 0.16 0.10 0.01 Median -0.0040 -0.0042 0.0153 0.0347 p-value 0.42 0.78 0.10 0.01 N 157 110 77 52 Industry-Adjusted Change t t+1 t+2 t+3 Panel A. MCs of DB Freeze Subsample with MCs > 0 in Year t-1 Mean 0.0017 -0.0013 -0.0012 -0.0020 p-value 0.13 0.33 0.40 0.05 Median -0.0001 -0.0001 -0.0003 -0.0003 p-value 0.14 0.00 0.00 0.01 N 104 72 53 36 Panel B. DC of DB Freeze Subsample Mean -0.0001 0.0002 0.0001 -0.0003 p-value 0.41 0.32 0.75 0.25 Median 0.0000 0.0000 0.0000 -0.0002 p-value 0.34 0.81 0.99 0.18 N 211 167 114 73 Panel C. Funding Ratio of PreFreeze Undef funded DB Firms Mean 0.0263 0.0510 0.0768 0.0788 p-value 0.00 0.01 0.00 0.00 Median 0.0090 0.0111 0.0457 0.0744 p-value 0.05 0.03 0.00 0.00 N 155 107 75 50 Control Group-Adjusted Change t t+1 t+2 t+3 Panel A. MCs of DB Freeze Subsample with MCs > 0 in Year t-1 Mean 0.0011 -0.0022 -0.0024 -0.0048 p-value 0.39 0.29 0.31 0.09 Median -0.0001 -0.0002 -0.0004 -0.0004 p-value 0.05 0.01 0.00 0.01 N 77 53 42 29 Panel B. DC of DB Freeze Subsample Mean 0.0011 0.0009 0.0010 0.0006 p-value 0.00 0.01 0.02 0.24 Median 0.0002 0.0001 0.0002 0.0001 p-value 0.00 0.10 0.10 0.40 N 163 128 92 60 Panel C. Funding Ratio of PreFreeze Undef funded DB Firms Mean 0.0237 0.0638 0.1005 0.0756 p-value 0.16 0.05 0.02 0.01 Median 0.0146 0.0266 0.0715 0.0549 p-value 0.31 0.03 0.00 0.00 N 107 71 49 32 Table IV. Probit Regression of DB Plan Freeze The table reports estimates of the following probit model: [D.sub.i,t] = [beta] [Z.sub.it], + [u.sub.it,] where D is a dummy variable that takes a value of one for firm years following the pension freeze and zero otherwise, Z is a vector of the explanatory variables including lagged funding status (lagged FS), MC, non-pension cash flow (NPC), five-year cash flow volatility, firm size, firm age, and active participant ratio all measured at the beginning of the period, the two-digit SIC code industry's unionization rate, DB ratio, and R&D intensity. Five-year cash flow volatility is measured as the standard deviation of the past five-year ratios of annual earnings before interest, tax, depreciation, and amortization (EBITDA) to the book value of assets. R&D intensity is measured as R&D investment scaled by the book value of assets. Size is proxied by the natural logarithm of the book value of assets. Active participant ratio is the ratio of the number of active DB participants to total employees of the firm. Firm age is the number of years that a firm has been included in the Compustat database. The z-statistics are given in parentheses. Expected Coefficients Marginal Sign effects Lagged FS - -0.8 -0.30% (-0.79) MC + -0.889 -0.13% (-0.55) Lagged Tobin's Q + (0.04) -0.38% (-1.07) NPC - -0.002 -0.05% (-0.15) Lagged size ? 0.016 -0.33% (0.89) Lagged active participant ratio - -1.436 *** -2.84% (-10.64) Firm age ? -0.014 *** -2.07% (-7.00) Five-year cash flow volatility + 0.156 -0.30% (1.12) Industry DB ratio - -0.540 *** -0.94% (-2.65) Industry R&D intensity - 0.001 -0.11% (0.33) Industry union rate - -0.013 ** -0.91% (-2.17) Intercept 0.754 (3.43) Year fixed effects Yes Firm fixed effects Yes Firm-year observations 4,875 Pseudo R-sq 0.18 *** Significant at the 0.01 level. ** Significant at the 0.05 level. * Significant at the 0.10 level. Table V. Freeze Effects on Net Working Capital Panel A reports net working capital regression results of the firm fixed effect model and the second step of the Heckman (1979) correction for self-selection and the IV models of the following form: [NWC.sub.it] = [alpha] + [[beta].sub.1] [MC.sub.it] + [[beta].sub.2] [D.sub.it] + [[beta].sub.3] [D.sub.it] * [MC.sub.it] + [gamma] [X.sub.i,t] + [[epsilon.sub.it] MC is the scaled mandatory pension contribution and D is a dummy variable that takes a value of one for firm years following the pension freeze and zero otherwise. X is a vector of control variables that includes nonpension cash flow, lagged Tobin's Q, and lagged funding status. a is the inverse Mill's ratio. The FE versus RE Hausman test examines the null hypothesis that the RE (and OLS) estimates are consistent against the alternative hypothesis that the fixed effect model estimates are consistent. The Wald test of independent equation examines whether the error terms of the first and second equation of the treatment effect model are independent. The Hausman endogeneity test examines whether the OLS estimates are consistent with the alternative hypothesis and that the IV regression is appropriate. The test of weak instruments examines whether the generated instrument is valid. The Heckman (1979) two-step correction model is estimated with heteroscedascity-robust standard errors clustered by firms. The IV regression is estimated with bootstrap standard errors. Panel B replicates the above net working capital regression model with industry and year fixed effects, but no firm fixed effects for two subsamples, high (Hi-FC), and low (Lo-FC) financial constraints. The Hi-FC subsample includes firm year observations with Whited-Wu (2006) (WW) index values above the median index value in a given year. The Lo-FC subsample includes the remaining observations. In Panel B, the coefficients of the other control variables are suppressed for brevity. The t-statistics are given in parentheses. Panel A. Freeze Effects on Net Working Capital FE1 FE2 Self- IV Model Selection MC -1.026 *** -1.175 *** -1.431 *** -1.203 *** (-3.17) (-4.78) (-10.22) (-6.27) MC*D 1.575 *** 0.848 * 1.480 ** (3.07) (1.72) (1.99) D -0.014 0.04 -0.075 (-1.27) (0.65) (-1.42) Lagged FS 0.074 0.102 0.006 -0.01 (0.62) (0.88) (0.05) (-0.08) NPC 0.568 *** 0.573 *** 0.001 *** 0.013 *** (6.84) (6.90) (6.67) (4.33) Lagged Tobin's Q 0.028 0.028 0.011 0.047 *** (1.33) (1.33) (1.57) (6.71) Intercept 0.058 0.058 0.008 0.116 *** (1.57) (1.57) (0.31) (4.00) [lambda] -0.031 (-0.89) Firm-year 4,799 4,799 4,311 3,042 observations Firm fixed effects Yes Yes No No Year fixed effects Yes Yes Yes Yes [R.sup.2] within 0.08 0.08 Adj [R.sup.2] 0.64 0.64 F-stat 36.21 14.78 Prob > F 0.00 0.00 Hausman test FE vs. RE: [chi square] 17.67 19.3 Prob> [chi 0.06 0.08 square] Wald test of independent eqns.: 2.66 [chi square] Prob> 0.10 [chi square] Hausman test of endogeneity: [chi square] 8.93 Prob> 0.000 [chi square] Test of weak instrument: F-stat 52.35 Prob>F-stat 0.00 Partial R-sq 0.03 Panel B. Freeze Effects on Net Working Capital Based on Financial Constraint Rankings Hi-FC Lo-FC p-value (Hi-FC - Lo-FC) MC -1.170 *** -0.231 0.42 (-3.86) (-0.54) D -0.027 -0.022 * 0.80 (-1.23) (-1.83) MC * D Firm-year observations 2,399 2,400 [R.sup.2] within 0.17 0.05 Adj. [R.sup.2] 0.75 0.6 Hausman test FE vs. RE: [chi square] 24.41 13.54 Prob > [chi square] 0.01 0.26 Hi-FC Lo-FC p-value (Hi-FC - Lo-FC) MC -1.264 *** -0.496 0.55 (-5.16) (-1.29) D -0.038 * -0.025 ** 0.54 (-1.73) (-2.08) MC * D 1.214 *** 1.355 0.90 (3.21) (1.39) Firm-year observations 2,399 2,400 [R.sup.2] within 0.17 0.06 Adj. [R.sup.2] 0.75 0.60 Hausman test FE vs. RE: [chi square] 26.12 13.01 Prob > [chi square] 0.01 0.37 *** Significant at the 0.01 level. ** Significant at the 0.05 level. * Significant at the 0.10 level. Table VI. Freeze Effects on Financial Leverage Panel A reports financial leverage regression results of the firm fixed effects model, industry fixed effects model, and the second step of the Heckman (1979) correction for self-selection and the IV model of the following form: Book [leverage.sub.it] = [alpha] + [[beta].sub.1] [MC.sub.it] + [[beta].sub.2] [D.sub.it] + [[beta].sub.3] [D.sub.it] * [MC.sub.it] + [gamma] [X.sub.i,t] + [[epsilon].sub.it] .MC is the scaled mandatory pension contribution and D is a dummy variable that takes a value of one for firm years following the pension freeze and zero otherwise. X is a vector of control variables that includes lagged Tobin's Q, nonpension cash flow, lagged funding status, asset tangibility, and size. [lambda] is the inverse Mill's ratio. The FE versus RE Hausman test examines the null hypothesis that the RE (and OLS) estimates are consistent against the alternative hypothesis that the fixed effect model estimates are consistent. The Wald test of independent equation examines whether the error terms of the first and second equation of the treatment effect model are independent. The Hausman endogeneity test examines whether the OLS estimates are consistent with the alternative hypothesis and that the IV regression is appropriate. The test of weak instruments examines whether the generated instrument is valid. The Heckman (1979) two-step correction model is estimated with heteroscedascity-robust standard errors clustered by firms. The IV regression is estimated with bootstrap standard errors. Panel B replicates the above leverage regression model with industry and year fixed effects, but no firm fixed effects for two subsamples, high (Hi-FC) and low (Lo-FC) financial constraints. The Hi-FC subsample includes firm-year observations with Whited-Wu (2006) (WW) index values above the median index value in a given year. The Lo-FC subsample includes the remaining observations. In both panels, the coefficients of other control variables are suppressed for brevity. The t-statistics are given in parentheses. Panel A. Freeze Effects on Financial Leverage 1 2 3 MC 1.083 ** 0.157 0.265 ** (2.34) (1.44) (2.35) MC*D -0.405 (-1.00) D 0.024 ** (2.67) Lagged FS -0.379 *** -0.147 * -0.189 *** (-3.68) (-1.93) (-2.66) [lambda] Firm-year observations 5,506 5,506 5,506 Industry fixed effects Yes No Yes Firm fixed effects No Yes No Year fixed effects Yes Yes Yes [R.sup.2] within 0.10 Adj. [R.sup.2] 0.31 0.84 0.32 F-stat Prob>F-stat Hausman test FE vs. RE: [chi square] 72.62 Prob> [chi square] 0.00 Wald test of independent eqns.: [chi square] Prob > [chi square] Hausman test of endogeneity: [chi square] Prob > [chi square] Test of weak instrument: F-stat Prob > F-stat Partial R-sq 4 5 6 MC 0.165 * 1.181 *** 0.419 *** (1.65) (2.89) (3.22) MC*D -0.107 -2.779 *** -1.025 ** (-0.27) (-4.32) (-2.13) D 0.020 ** 0.182 *** 0.060 * (2.00) (4.55) (1.94) Lagged FS -0.151 ** -0.402 *** -0.311 *** (-1.99) (-3.62) (-4.09) [lambda] -0.079 *** (4.27) Firm-year observations 5,506 4,842 3,584 Industry fixed effects No Yes Yes Firm fixed effects Yes No No Year fixed effects Yes Yes Yes [R.sup.2] within 0.10 Adj. [R.sup.2] 0.84 F-stat 24.89 24.98 Prob>F-stat 0.0 0.0 Hausman test FE vs. RE: [chi square] 82.5 Prob > [chi square] 0.00 Wald test of independent eqns.: [chi square] 27.90 Prob > [chi square] 0.00 Hausman test of endogeneity: [chi square] 23.74 Prob > [chi square] 0.00 Test of weak instrument: F-stat 196.63 Prob >F-stat 0.00 Partial R-sq 0.09 Panel B. Freeze Effects on Financial Leverage Based on Financial Constraints Rankings Hi-FC Lo-FC p-value (Hi-FC- Lo-FC) MC 1.117 *** -0.833 * 0.00 (4.28) (-1.77) D 0.056 ** 0.015 0.12 (2.24) (1.50) MC*D Lagged FS -0.689 *** -0.273 *** 0.02 (-5.14) (-3.96) Year fixed effects Yes Yes Industry fixed effects Yes Yes Firm-year observations 2,753 2,753 Adj. [R.sup.2] 0.30 0.38 Hi-FC Low-FC p-value (Hi-FC Lo-FC) MC 1.308 *** -0.668 0.00 (5.03) (-0.98) D 0.085 *** 0.017 0.01 (3.27) (1.13) MC*D -3.253 *** -1.275 0.10 (-5.49) (-0.95) Lagged FS -0.747 *** -0.273 *** 0.03 (-5.45) (-2.84) Year fixed effects Yes Yes Industry fixed effects Yes Yes Firm-year observations 2,753 2,753 Adj. [R.sup.2] 0.31 0.38 *** Significant at the 0.01 level. ** Significant at the 0.05 level. * Significant at the 0.10 level. Table VII. DB Freeze and Investment The table reports the investment regression results of the firm fixed effects model and the second step of the Heckman (1979) two-step correction and the IV regressions models with the following form: [I.sub.it] = [alpha] + [[beta].sub.1] [MC.sub.it] + [[beta].sub.2] [D.sub.it] + [[beta].sub.3] [D.sub.it] * [MC.sub.it] + [gamma] [X.sub.i,t] + [[epsilon].sub.it]. I denotes scaled annual investment, MC is the scaled mandatory pension contribution, and D is a dummy variable that takes a value of one for firm years following the pension freeze and zero otherwise. X is a vector of control variables that include nonpension cash flow, lagged Tobin's Q, and lagged funding status. [lambda] is the inverse Mill's ratio. The FE versus RE Hausman test examines the null hypothesis that the RE (and OLS) estimates are consistent against the alternative hypothesis that the fixed effect model estimates are consistent. The Wald test of independent equation examines whether the error terms of the first and second equation of the treatment effect model are independent. The Hausman endogeneity test examines whether the OLS estimates are consistent with the alternative hypothesis and that the IV regression is appropriate. The test of weak instruments examines whether the generated instrument is valid. The Heckman (1979) two-step correction model is estimated with heteroscedascity-robust standard errors clustered by firms. The IV regression is estimated with bootstrap standard errors. The coefficients of other control variables are suppressed for brevity. The t-statistics are given in parentheses. FE1 FE2 MC -0.045 ** -0.054 ** (-2.25) (-2.35) D -0.003 (-1.00) MC*D 0.035 (0.32) Firm-year observations 5,531 5,531 Firm fixed effects Yes Yes Year fixed effects Yes Yes Adjusted R-sq 0.74 0.74 [chi square] Prob > [chi square] Hausman test FE vs. RE: [chi square] 55.88 171.56 Prob > [chi square] 0.00 0.00 Wald test of independent eqns.: [chi square] Prob > [chi square] Hausman test of endogeneity: [chi square] Prob > [chi square] Test of weak instrument: F-stat Prob > F-stat Partial R-sq Self-Selection IV Model Correction MC -0.173 *** -0.209 *** (-6.41) (-5.81) D -0.021 *** -0.024 ** (-7.00) (-2.00) MC*D 0.150 * 0.301 (1.65) (1.40) 0.008 (8.00) Firm-year observations 4,875 5,452 Firm fixed effects No No Year fixed effects Yes Yes Adjusted R-sq [chi square] 124.95 125.28 Prob > [chi square] 0.00 0.00 Hausman test FE vs. RE: [chi square] Prob > [chi square] Wald test of independent eqns.: [chi square] 15.57 Prob > [chi square] 0.00 Hausman test of endogeneity: [chi square] 7.08 Prob > [chi square] 0.01 Test of weak instrument: F-stat 38.31 Prob > F-stat 0.00 Partial R-sq 0.05 *** Significant at the 0.01 level. ** Significant at the 0.05 level. * Significant at the 0.10 level. Table VIII. Robustness Checks In the following analyses, year t is the freeze year. In self- adjusted analysis, year t - 1 is used as a benchmark for computing changes in the succeeding years t to t + 3. In industry-adjusted analysis, the differences between a firm and its two-digit SIC industry median ratio in year t, t + 1, t + 2, and t + 3 are compared with the respective differences in year t - 1 (diff in diff). In control firm-adjusted analysis, the differences between a firm and its matching firm in year t, t + 1, t + 2, and t + 3 are compared with the respective differences in base year t - 1 (diff in diff), where the matching firm is selected from the same two-digit SIC industry as the sample firm, and has a size and book-to-market value within the range of 70% to 130% of the sample firm in the year preceding the freeze. If more than one matching firm is available, we rank firms on the basis of the absolute value of the difference in book-to-market in comparison with the sample firm and select the firm with the lowest absolute value of difference as the matching firm. In Panel A, employee growth rate is calculated as the year-over-year percentage difference in the number of employees. In Panel B, the average of scaled M&A transaction values for the DB freeze firms over three years before the freeze is compared with the respective average over three years after the freeze. In Panel C, R&D expenditures are scaled by the lagged book value of assets. Statistical inference is based on the t-test for means and Wilcoxon signed- rank test for medians. Panel A. Change Employee Growth Rate Self-Adjusted (base year t - 1) Difference in the Average Three Year t t + 1 t + 2 t + 3 Annual Growth Rates b/w prefreeze and postfreeze Mean 0.0292 0.0656 0.1360 0.2052 -0.0017 p-value 0.00 0.00 0.00 0.00 0.88 Median 0.0000 0.0165 0.0727 0.1191 -0.0015 p-value 0.22 0.02 0.00 0.00 0.89 N 254 236 184 126 235 Panel B. Change in M&A M&A value/ M&A value/ M&A value/ lag-asset lag-MVA lag-MVE Mean 0.0130 0.0051 0.0137 p-value 0.21 0.15 0.25 Median 0.0000 0.0000 0.0000 p-value 0.01 0.00 0.00 N 252 252 252 Panel C. Change in R&D Self-Adjusted (base year t-1) t t+1 t+2 t+3 Mean 0.0054 0.0018 0.0001 0.0041 p-value 0.19 0.24 0.96 0.07 Median 0.0001 0.0003 0.0001 0.0002 p-value 0.06 0.13 0.44 0.22 N 111 107 84 59 Panel C. Change in R&D Industry-Adjusted (diff-in-diff) t t+1 t+2 t+3 Mean -0.0002 0.0001 -0.0035 -0.0027 p-value 0.89 0.97 0.13 0.40 Median 0.0000 0.0000 0.0000 0.0000 p-value 0.89 0.89 0.46 0.63 N 110 106 83 58 Control Firm-Adjusted (diff-in-diff) t t+1 t+2 t+3 Mean 0.0061 0.0016 0.0009 0.0039 p-value 0.18 0.33 0.58 0.19 Median 0.0000 0.0000 0.0000 0.0014 p-value 0.68 0.61 0.31 0.10 N 102 98 79 55 Table IX. Freeze Announcement Abnormal Returns and Cross-Sectional Regression In Panel A, CARS for the event window are computed using the market model with the return of the CRSP equally weighted portfolio proxies for market return. The p-value reported in the table represents the probability of a Type I error. Panel B reports results of the following cross sectional regression model: [CAR.sub.i] = [delta] + [[delta].sub.t] + [beta][X.sub.i] + [gamma][[lambda].sub.i] + [u.sub.i]. CAR; is stock is CAR calculated using the market model. X is a vector of firm characteristics that include MC, funding status, overfunding and underfunding status, Hi-FC, Hi Gov, stock price run-up, size, and book-to-market ratio. All right hand side variables are measured at the end of December of the previous year. MCs and funding status are scaled by the market value of equity. Hi-FC is a dummy variable that takes a value of one if a firm's Whited-Wu (WW) (2006) index is above the sample median WW index value, and zero otherwise. Higher WW index denotes more financial constraints. Hi Gov is a dummy variable that takes a value of one if a firm's GIM index is above the sample median GIM index, and zero otherwise. High GIM index values reflect more anti-takeover provisions. Size is proxied by the natural logarithm of market value of equity. Stock price run-up is the 40-day stock return during the period (-45, -5) before the DB freeze announcement. [lambda] is the inverse Mill's ratio calculated from the probit regression outputs in the first step of the Heckman (1979) two-step self-selection correction model. The model is estimated by OLS with heteroskedasticity-robust standard errors. Panel A. Freeze Announcement Abnormal Returns Window N CAAR Pos: Patell Neg Z (-30,-2) 67 -2.38% 24 : 43 -2.11 (-1,0) 67 0.18% 34 : 33 1.29 (0,+ 1) 67 0.36% 38 : 29 2.73 (-1,+1) 67 0.35% 39 : 28 2.03 (+2,+5) 67 -0.29% 31 : 36 -0.41 (+2,+10) 67 -0.55% 30 : 37 -0.25 Panel A. Freeze Announcement Abnormal Returns Window p-value Generalized p-value Sign Z (-30,-2) 0.03 -1.79 0.07 (-1,0) 0.20 0.66 0.51 (0,+ 1) 0.01 1.64 0.10 (-1,+1) 0.04 1.88 0.06 (+2,+5) 0.69 -0.08 0.94 (+2,+10) 0.80 -0.32 0.75 Panel B. Cross Sectional Regression Expected 1 2 3 Sign MC + 0.281 * 0.292 0.491 *** (1.65) (1.10) (6.14) FS +/- -0.01 (-1.00) Overfunding +/- status Underfunding +/- status Hi-FC - Hi GOV - Stock price +/- run-up Size +/- B/M - [lambda] - 0.015 0.041 0.036 (0.38) (1.17) Year fixed Yes Yes effects Number of 41 41 41 observations Adj. R-square 0.07 0.07 0.19 Panel B. Cross Sectional Regression 4 5 6 MC 0.497 *** 0.495 ** 0.53 (5.78) (3.81) (0.95) FS -0.021 -0.022 (-1.00) (0.76) Overfunding -0.002 status (-0.01) Underfunding 0.036 status (0.05) Hi-FC 0.004 0.003 (0.22) (0.16) Hi GOV 0.000 0.002 (0.01) (0.22) Stock price -0.264 -0.300 -0.283 run-up (-1.61) (-1.64) (-1.63) Size 0.003 0.005 0.005 (1.50) (1.25) (1.00) B/M 0.004 0.007 0.005 (0.57) (0.88) (0.63) [lambda] 0.012 -0.001 (0.88) (0.22) (-0.02) Year fixed Yes Yes Yes effects Number of 36 30 30 observations Adj. R-square 0.35 0.15 0.09 *** Significant at the 0.01 level. ** Significant at the 0.05 level. * Significant at the 0.10 level. Table X. DB Freeze and Long Run Abnormal Returns (LAR) Regression-FE Model The table reports results of the following regression model: [LAR.sub.it] = [alpha] + [[alpha].sub.1] + [[beta].sub.i] [MC.sub.it] + [[beta].sub.2] [D.sub.it] + [[beta].sub.3] [D.sub.it] * [MC.sub.it] + [gamma] [X.sub.i,t] + [[epsilon] of].sub.it]. [LAR.sub.it] is stock i's compound returns from July of year t + 1 to June of year t + 2, calculated using the market model (i.e., market model abnormal returns) or adjusted by the same two-digit SIC code industry returns (i.e., industry-adjusted returns). MC is e scaled mandatory pension contribution and D is a dummy variable that takes a value of one for firm years following the pension freeze and zero otherwise. X is a vector of firm characteristics. Momentum is proxied by the past 12-month returns. Size is proxied by the natural logarithm of the market value of equity in June of year t + 1. All regressor variables are measured in December of year t except for momentum and size, which are measured in June of year t + 1. MCs and funding status of year t are scaled by the market value of equity in June of year t + 1. The market-model LAR regression additionally includes two-digit SIC industry dummies. The t-statistics, based on heteroskedasticity-robust standard errors clustered by firms, are given in parentheses. Industry- Market Model Adjusted LAR LAR MC -0.213 ** -0.173 *** (-2.48) (-2.65) D 0.009 -0.021 (0.41) (-0.88) MC*D 0.298 0.136 (0.88) (0.20) Momentum -0.037 -0.116 ** (-0.86) (-2.00) Size 0.013 0.031 *** (1.18) (2.58) Book-to-market -0.014 -0.013 (-1.00) (-0.54) Intercept 0.077 -0.431 *** (0.47) (-2.61) Year fixed effects Yes Yes Firm fixed effects Yes Yes Firm-year observations 4,769 4,784 Adj. [R.sup.2] 0.43 0.11 Hausman test FE vs. RE: [chi square] 1,391.75 1,471.72 Prob > [chi square] 0.00 0.00 *** Significant at the 0.01 level. ** Significant at the 0.05 level. * Significant at the 0.10 level.
|Printer friendly Cite/link Email Feedback|
|Author:||Phan, Hieu V.; Hegde, Shantaram P.|
|Date:||Jun 22, 2013|
|Previous Article:||Emerging stars and developed neighbors: the effects of development imbalance and political shocks on mutual fund investments in China.|
|Next Article:||Pension policy and the value of corporate-level investment.|