21. Generally, when a company freezes its DB plan, it discloses information related to the
freeze for several years subsequent to the decision. For instance, it is possible to read a
2004 footnote disclosure that references a decision to freeze a pension plan in 2000,
2001, 2002, or 2003.
22. We present the sample sizes used for various tests in the respective tables.
23. It is often difficult to distinguish between soft and partial freezes. For example, a firm
may institute a freeze that has characteristics of both freeze types (e.g. reduced benefits
for current employees and no benefits for new employees). Therefore, we classify soft
and partial freezes together as distinct from hard freezes.
24. We also examine corporate governance as a control variable and find it insignificant,
both in univariate tests and logistic regression analyses. Ultimately, we eliminate
corporate governance from our analyses as it results in additional missing
observations and does not materially affect the inferences for any of our models.
25. Though high correlation between FVA and PBO in this model cause variance inflation
factors to increase beyond a level of 10, the condition indices diagnostic remains below 30,
and we are confident that our inferences hold. Models including only the individual asset
or liability variables yield nearly identical results. We also use another measure,
FUNDRATIO, defined as FVA/PBO, which is the proportion of the pension liability that is
funded. The FUNDRATIO variable is also highly significant and negative (not tabulated).
26. We also expand our balance sheet models to include a net periodic pension cost
variable (not tabulated). The net periodic pension cost variable is never significantly
associated with the freeze decision in any model, and none of our inferences change by
its inclusion. We investigate pension accounting income statement effects on the DB
plan freeze decision in their own right as well in supplemental tests, the results of
which we present in Table VI.
27. The GAO (2008) study reports that both plan contributions as a percentage of overall
current cash flows and the magnitude and volatility of plan contributions are potential
reasons for firms freezing their DB plans. Therefore, we also examine firm contributions
to the plan in their own right as well as benefits paid to retirees and actual returns on
plan assets as alternate variables in place of our main cash flow variable in levels,
changes, and average formats. In addition, we replace our main cash flow variable with
average cash flow from operations, cash flow from operations at time t 1, and cash
flow from operations at time t 1 along with the change in cash flow from operations
from time t 3 to time t 1. While we do not tabulate any of the results related to our
alternative specifications in this regard, we consistently find that none of these cash flow
measures are significantly associated with the freeze decision.
28. By 2007, in aggregate, the S&P 500 was once again overfunded with greater pension
assets than liabilities (Standard & Poor’s, 2008).
29. We also use an alternate specification to better distinguish between the influence of the
level of return on assets and the change in return on assets on the freeze decision.
Specifically, we employ the return on assets at time t 1 along with the change in
return on assets from time t 3 to time t 1. The levels variable remains significant,
but the change variable is insignificant in this model (not tabulated).
30. Mittelstaedt et al. (1995) find that contracting costs related to closeness to debt
covenant restrictions are significantly related to benefit reductions. We attempt to use a
similar contracting cost variable in our tests (SFAS158_EFFECT (net of tax) scaled by
lagged assets and interacted with LEV). However, we find no significance for this
incremental leverage effect variable, and all other inferences remain unchanged when
we expand our models to include it (not tabulated).