TRANSACTIONS OF SOCIETY OF ACTUARIES

1994 VOL. 46

THE PROJECTED UNIT CREDIT M E T H O D W I T H BENEFITS

A P P O R T I O N E D BY INTEREST-ADJUSTED SM.ARY

GERAI.D LEE GIESECKE

ABSTRACT

This paper adopts the framework whereby pension benefits are allo-
cated one exit point at a time in some fashion to the years of service of
those persons expected to leave at each exit point. Of particular interest
is an allocation by interest-adjusted salary. Within the same framework,
the paper considers allocations based on the implicitly perceived value
of the benefit to employees. While impractical, these perceived value
allocations suggest the desirability of something more back-loaded than
allocating equal benefits to each year of service. Also considered are the
method of combining exit points, the determination of different normal
costs for different cost centers, the allocation of costs for persons trans-
ferring between cost centers, the problem of negative accruals, and why
the current employment benefit formula should not be used to allocate
the benefit at an assumed exit point.

INTRODUCTION
There are as many ways to fund a retirement benefit as there are to
finance a house. To say that one is wrong is analogous to saying that a
house cannot be financed with different payment schedules. To say that
one is best presumes a particular criterion or a particular weighting of
criteria. From the perspectives of accounting and economic theory, some
cost allocation methods make more sense than others. So far, however,
the ideal cost allocation method for accrual accounting purposes has not
been found. Two accountants (Hall and Landsittel [6]) proposed using
the projected unit credit method with the benefit allocated by salary as
something of an accounting standard. Their proposal was not particularly
well received. I With a minor change, however, a rather interesting fund-
ing method results. It is not, in my view, the only reasonable alternative;
however, its funding philosophy merits further attention.

~Reactions by members of the Pension Research Council are given at the end of the 1977
publication.

193
194 TRANSACTIONS, VOLUME XLVI

The rationale of the Hall and Landsittel proposal was simple enough.
They reasoned that "since services are rewarded by a wage or salary,
which is presumed to reflect the worth of the employee to the firm, the
accrual of a pension obligation should be measured in terms of the em-
ployee's compensation, irrespective of how the benefit is determined un-
der the terms of the plan."
Among the criticisms made of the Hall and Landsittel proposal are the
following:
(1) The proposal did not fully appreciate the fact that it was only one
of many possible mappings of the projected benefit onto individual
years of service.
(2) Imposing an arbitrary standard causes undue paperwork.
(3) The method has unacceptably high back-loading.
(4) The method would not be permitted by federal tax regulations
(1.412(c)(3)-1).
To this list I can add another criticism:
(5) Prorating by nominal salary ignores the effects of inflation.
Since 1977 dollars are not equivalent to 1993 dollars, allocating a ben-
efit in proportion to unadjusted salary leads to peculiar results. An ob-
vious solution is to allocate the projected benefit in proportion to infla-
tion-adjusted salary. Chris Doyle, a colleague at the Department of
Defense, wondered, Why not use interest-adjusted salary? This latter
suggestion is the one considered here. It would not, at present, be per-
mitted under IRS regulations, but it has an appealing funding philosophy
and considerably less back-loading than the original Hall and Landsittel
proposal.
Later in the paper, I also discuss what I call perceived value alloca-
tions. These assess the employee's perceived value of the retirement ben-
efit and allocate the benefit to individual years of service, taking these
perceptions under consideration. While perceived value allocations may
be impractical for ordinary uses, they have some implications for what
we ought to be doing.
At various places in the paper, I appeal to what makes the best sense
for accrual accounting. In doing so, I do not restrict myself to current
accounting conventions or regulatory requirements. In fact, what I rec-
ommend would not pass muster there. Nevertheless, we can hope that
arbitrary standards will be modified and that those not restricted by them
may be able to implement some of the ideas presented here.
 PROJECTED UNIT CREDIT METHOD 195

JUSTIFICATION FOR INTEREST-ADJUSTED SALARY

Philosophically the benefit allocation under the Hall and Landsittel
proposal is not without merit. Retirement benefits are almost always re-
lated to levels of compensation. So it seems counterintuitive for an al-
location scheme to ignore different compensation levels within an indi-
vidual's career. By eliminating the apples-and-oranges nature of unadjusted
dollars, we could get a reasonable salary-based allocation. At the same
time, we would eliminate some back-loading. This gives us a projected
unit credit method with the benefit allocated by inflation-adjusted salary.
In addition, because the Social Security benefit formula is based on an
average indexed wage, I think we must consider a related allocation method
as a reasonable alternative.
The argument for interest-adjusted salary is similar. We imagine our-
selves at an exit point and must somehow apportion the benefit back to
individual years of service. In doing this, we want to treat the retirement
benefit earned in any year as a form of undistributed compensation that
is related to salary. But this situation is similar to a series of investments
made by a firm or an individual. The future value is the values brought
forward by interest rates. So the benefit at retirement could be appor-
tioned by interest-adjusted salary.
The argument is appealing but not mathematically compelling. How-
ever, there are two further arguments in favor of apportioning the benefit
by interest-adjusted salary. The first argument is that the benefit builds
at a reasonable rate compared to the obvious alternatives. At one extreme
the projected benefit is allocated equally to all years of service. This
method is required by federal tax regulations (1.412(c)(3)-1) when level
dollar amounts or level percentages of pay are not being used for the
normal cost and it is not a career-average pay plan. 2 If an employee is
expected to work 20 years for a firm, t/2oth of the expected benefit would
be allocated to each year. This leads to a relatively fast benefit buildup.
At the other extreme would be a benefit apportioned by inflation-adjusted
salary. 3 This is essentially the Hall and Landsittel proposal using real
dollars rather than nominal dollars.

2The regulation states "The numerator of the fraction is the participant's credited years o f
service. The denominator is the participant's total credited years of service at the anticipated
benefit commencement date. Adjustments are made to account for changes in the rate o f
benefit accrual. An allocation based on compensation is not permitted."
3In a later section I describe a perceived value allocation that also seems reasonable. U n d e r
some assumptions this could be more heavily back-loaded than an inflation-adjusted salary
allocation.
196 TRANSACTIONS, VOLUME XLVI

The original Hall and Landsittel scheme, however, leads to a slower

benefit buildup than the projected unit credit method with inflation-ad-
justed dollars. It is not a reasonable alternative because of the apples-
and-oranges nature of the nominal dollars used in mapping the benefit.
More extreme in the other direction for plans with substantial early
attrition is the entry age normal method. This can lead to a benefit that
builds even more rapidly than the level benefit method. Again, however,
this is not a reasonable alternative for accounting purposes, since the
entry age normal method levies the same normal cost, regardless of the
probability of reaching retirement. This disadvantage of the entry age
normal method is discussed more fully later in this section.
The second additional argument in favor of apportioning by interest-
adjusted salary is the following funding philosophy. The entry age nor-
mal method funds by a constant percentage of salary. The projected unit
credit method with interest-adjusted salary is similar. A constant is used,
but it is applied to the product of salary and the probability of retirement
at a particular exit point. The constant is the same for all lengths of
service, assuming there is only one point of retirement. 4
To demonstrate this, we need to develop a few equations. Projected
unit credit methods assign the projected benefit at retirement to particular
years of service. For the level benefit method, if at the end of 20 years
of service, a retirement benefit has a lump-sum value of $300,000, we
would assign V20th of $300,000 (discounted from retirement back to the
service year) to each year of service. If the funding is identical to the
accrual amount, the normal cost contribution for one person for one year
would pay only for the apportioned piece of the projected benefit. The
normal cost contribution for person l in the j-th year of service would
then be:
$300,000 interest
= X x prob (reaching retirement).
NCjt 20 discount

To simplify matters, I assume that retirement occurs at the end of 20

years of service, that salary and normal cost contributions are paid at the
end of each year for the average number working during the year, and
that salary increases are given at the beginning of the year. If the number

4The extension to multiple exit points is considered later, but the gist of it is that for every-
one "expected ~ to terminate at a particular exit point, we must set aside the same percentage
of salary. Different amounts would be set aside for each potential exit point.
 PROJECTED UNIT CREDIT METHOD 197

reaching the end of the j-th year is nj and we assume a level distribution
of losses during the year, the average number working (nj_~ +nj)/2 will
equal the mid-year value of ni_o.5. The normal cost contribution for each
year is the sum of the normal costs for all 20 cohorts. Assuming a con-
stant interest rate i, the normal cost contribution for the cohort in the j-th
year of service is:
$300,000 1 n2o
NCj = nj-o.5 20 (1 + i) 2°-j nj-o.5" (1)

Equation (1) gives the normal cost contribution in dollars. To get a

percentage, Equation (1) and the subsequent equations in this paper need
to be divided by the payroll in the j-th service year.
To make Equation (1) more general, assume a benefit worth Bz at the
end of z years and an apportioning ratio offj. Then
1 n,
NCj = nj-o.5 Bzfj
(1 + i) ~-j nj-0.5"
The generalization to different possible years of retirement with dif-
fering probabilities and benefits is straightforward and need not concern
us here. Similarly, we might use months rather than years or assume
different interest rates in each year.
What is of interest here is the apportioning ratio, )~. The first two equa-
tions assumed 1/20th, or 1/z. The apportioning could also be done by
salary, interest-adjusted salary, and so on.
By using interest-adjusted salary in the f ratio, the normal cost for
service year j becomes:
Salj(1 + i) ~-j 1 nz
NCj = ny-0.5 Bz z . (2)
Salk(1 + i) ~-* (1 + i) z-i nj-0.5
k= I

where Sali=the salary of a typical individual paid at the end of the j-th
year.
If inflation-adjusted salary were wanted, the apportioning ratio in
Equation (2) is all that would have to be modified. To get the formula
for the entry age normal method, we proceed as follows. In Equation
(2) we apportioned the benefit by the work (as measured by interest-
adjusted salary) of those who retire. For the entry age normal method,
198 TRANSACTIONS, VOLUME XLVI

we want to include the work of those who will not make it to retirement.
Each individual who makes it to retirement in this formulation could be
thought of as having a leverage in year j such that the contribution is
effectively increased by the ratio nj-o.5/n,.. The normal cost contribution
then becomes

ny-o.5 Salj(1 + i) ~-i

nz 1 nz
"~" = . (3)
NC~ nj-o.5 B~ ~ nk-0.5 (1 + i) '-i nj-0.5
2 Salk(1 + i) ~-k
k=l nz

By simplifying and repositioning nj-0.5 and n~, Equation (3) becomes

nj-o.5 Salj
NC7 TM = ndB, ~ (4)
2 nk-o.5 Salk(l + i) ~-~
k=l

If the nzBz in Equation (4) is divided by n0(1 +i) ~, it represents the

present value of future benefits at the point of entry. When the sum-
mation is divided by the same term, it becomes the present value of
future salaries at the point of entry. Thus we have
PV(future benefits)
NC~" = ni_o. 5 Salj. (5)
PV(future salary)
Equation (5) is a more familiar version of the entry age normal formula
with an assumed exit point. It is distinguished from the projected unit
credit method in Equation (2) only by the benefit-apportioning ratio. This
relationship between the entry age normal and the projected unit credit
methods has also been demonstrated by Anderson [1, p. 153]. The term
with the summation in the denominator in Equation (3) is the benefit-
apportioning ratio for the entry age normal method.
The entry age normal method has the peculiar interpretation that it
apportions the benefit to people who are not expected to retire, as well
as those who will. This may be seen more directly by rearranging the
terms of Equation (3) as follows:
nj-o.5 Salj(1 + i) ~-j 1
NC~ '~" = nzB~ z
(1 + i)z-J "
2 nk-0.5 Salk(1 + i) z-k
k= 1
 PROJECTED UNIT CREDIT METHOD 199

Because of the above peculiarity, the entry age normal method ap-
portions too much benefit to the early years of service. Consequently,
for accrual accounting purposes, it leaves something to be desired.
Of particular interest here is an alternative development of Equation
(2). It begins with the entry age normal method. However, instead of a
uniform percentage contribution on all salaries, a uniform contribution
is made on the product of salary and the probability of reaching retire-
ment. This is given by
PV(benefit) nz
NCj = nj-o.5 Salj
PV[salaries × prob (retirement)] nj-0.5
or

nz Bz
(1 + i ) ~ n~
NC i = ni_o.s Salj - - . (6)
~ nk-0.5 Salk n= n/-o.5
k=l (1 + i)k nk-0.5
Simplification shows that Equation (6) is equivalent to (2). So an al-
ternative interpretation of (2) is that it uses a uniform percentage applied
to the product of salary and the probability of reaching retirement. An-
other way of looking at this is to say that we contribute a uniform per-
centage of salary for the n~ persons in a cohort who will reach retirement.
This avoids a problem of the entry age normal method, that is, making
those who leave pay for those who stay.

A COMPARISON OF BENEFIT ALLOCATIONS FOR EQUATION 2

WITH OTHER METHODS
Table l presents the benefit-apportioning ratio by service year for reg-
ular enlistees in the military services of the U.S. under the five different
cost methods. The first three columns show the benefit allocation under
the projected unit credit method with the benefit being allocated by un-
adjusted salary, inflation-adjusted salary, and interest-adjusted salary, re-
spectively. The fourth column gives a level benefit allocation. The' final
column shows the benefit allocation under the entry age normal method.
All methods assume the net loss rates for regular enlistees in the U.S.
Armed Forces given in the valuation report [3] and ignore other losses
or transfers to other statuses. The five columns also assume retirement
at 20 years of service, an entry age to the nearest birthday of 19, an
200 TRANSACTIONS, VOLUME XLVI

interest rate of 7 percent, an inflation rate of 5 percent, an annual pay

scale increase of 5.75 percent, and the promotion and longevity increases
given for regular enlistees. The annual pay scale increase is in addition
to promotions and length-of-service increases.
Table 1 looks at a cohort retiring at the 20th year of service and asks
ih which year various pieces of the retirement benefit should have been
accounted for. Alternatively, it looks at those expected to retire in an
entry cohort and apportions their retirement benefit to a year of service.

TABLE 1
PROPORTION OF RETIREMENT BENEFIT EARNED BY REGULAR ENLISTEES IN THE MILITARY SERVICES
BY YEAR OF SERVICE AND COST METHOD
Projected Unit Credit MethodwithBenefit Allocated by
Year of Inflation-Adjusted Interest-Adjusted Level-Benefit Entry-Age
Service Salary Salary Salary Method NormalMethod
1 0.015 0.026 0.032 0.05 0.107
2 0.017 0.029 0.035 0.05 0.106
3 0.020 0.032 0.038 0.05 0.096
4 0.023 0.035 0.041 0.05 0.074
5 0.026 0.038 0.044 0.05 0.057
6 0.029 0.040 0.045 0.05 0.050
7 0.032 0.043 0.047 0.05 0.046
8 0.035 0.045 0.048 0.05 0.041
9 0.039 0.047 0.049 0.05 0.038
10 0.042 0.049 0.050 0.05 0.036
11 0.047 0.051 0.052 0.05 0.035
12 0.051 0.053 0.053 0.05 0.034
13 0.056 0.056 0.054 0.05 0.034
14 0.061 0.058 0.055 0.05 0.034
15 0.067 0.061 0.057 0.05 0.034
16 0.073 0.063 0.058 0.05 0.034
17 0.080 0.065 0.059 0.05 0.035
18 0.087 0.067 0.060 0.05 0.035
19 0.095 0.070 0.061 0.05 0.036
20 0.103 0.072 0.062 0.05 0.036
Total 1.00 1.00 1.00 1.00 1.00

Table 2 shows the normal cost percentages for each of these cost meth-
ods, using the same assumptions as for Table 1.

WEIGHTED NORMAL COST PERCENTAGE

As the bottom row of Table 2 suggests, the normal cost percentage at
each length of service can be weighted to get a weighted normal cost
percentage. This is defined as the normal cost percentage, which if ap-
plied to all salaries would be equivalent to the length-of-service specific
 PROJECTED UNIT CREDIT METHOD 201

TABLE 2
NORMAL COST PERCENTAGES FOR REGULAR ENLISTEES BY YEAR OF SERVICE AND COST METHOD
Projected Unit Credit Method with Benefit Allocated by
Year of Inflation-Adjusted Interest-Adjusted Level-Benefit Entry-Age
Service Salary Salary Salary Method Normal Method
1 6.4 11.3 13.9 21.7 46.4
2 7.5 12.7 15.3 21.9 46.4
3 9.6 15.5 18.2 24.1 46.4
4 14.4 22.2 25.7 31.3 46.4
5 21.3 31.1 35.4 40.6 46.4
6 26.5 36.9 41.2 46.0 46.4
7 32.9 43.6 47.7 50.9 46.4
8 39.7 50.1 53.8 56.2 46.4
9 47. I 56.6 59.7 60.3 46.4
10 54.4 62.3 64.4 64.1 46.4
11 61.7 67.3 68.3 66.1 46.4
12 69.0 71.7 71.3 67.5 46.4
13 76.4 75.6 73.9 67.8 46.4
14 83.7 78.9 75.7 68.2 46.4
15 91.1 81.8 77.0 67.5 46.4
16 98.7 84.4 77.9 67.4 46.4
17 106.6 86.8 78.6 66.5 46.4
18 114.8 89.0 79.1 65.9 46.4
19 123.2 91.0 79.4 65.0 46.4
20 131.8 92.7 79.4 64.1 46.4
Weighted 49.3 I 48.5 48.1 47.6 46.4

normal cost percentages applied to the salaries at each length of service.

In this case the salaries used are what we would see after 20 successive
equally sized cohorts, assuming all assumptions were met. The weighted
normal cost would be expected to vary, however, as force composition
changed. The formula for the weighted normal cost percentage is as
follows:
2

(NCPj Salj.,)
j=l
Weighted NCP, = z
E Salj.,
j=l
where
J = length of service
t = period for which a normal cost contribution is being made
Salj,, = salaries earned for all persons in theirj-th year of service dur-
ing period t.
202 TRANSACTIONS, VOLUME XLVI

The weighted normal costs in Table 2 differ by funding method, since

the more front-loaded methods will build a larger fund prior to reaching
equilibrium. Although the present value of future benefits is the same in
all the cost methods, the present value of future normal cost contributions
differs. In effect, when we choose a cost method, we assume a present
value of past normal cost contributions. If we pick a method with smaller
assumed past normal cost contributions, the future normal cost contri-
butions must be larger.
The motivation for getting a weighted normal cost is to simplify mat-
ters administratively. Ultimately, our goal is to have weighted normal
costs that differ for different cost centers. For example, we might have
a different normal cost for each branch of the military services, with
separate normal costs for officers and enlistees. To do this, we have to
extend Equation (2) to more than one point of retirement and to allow
transfers between cost centers.

EXTENSION TO MULTIPLE EXIT POINTS

The extension to multiple points of retirement is straightforward. For
each possible point of retirement, we apply Equation (2). The n, in Equa-
tion (2) now means those who retire at z years, a smaller number than
those who complete z years. Then for each length of service we add the
appropriate normal costs. For example, if a person with 20 years of ser-
vice could retire at any time in the next 10 years, there would be 10
normal costs, one for each of the 10 possible points of retirement. So
the normal cost for the person with 20 years of service would be a sum
of l0 numbers. The normal cost for a person with 19 years of service
would be a sum of 11 numbers, and so forth. The normal cost equation
for period j is then
~_~ fz.j B~nz
NCi = (1 + i) '-j

where f,,j=part of B, attributable to period j.

Administratively, this is not prohibitively complicated, since for each
cost center we can use a weighted normal cost percentage, which is just
one number.
 PROJECTED U N I T C R E D I T M E T H O D 203

The unfunded liability is given by

( n~,j B z .k

urn=EEl. z y
~+i)- j /-Assets

where nzj=number at length of service j expected to retire at exit

point z.
The normal costs here are assumed to be paid at the end of the period
and included in the assets. Otherwise there is nothing special in the treat-
ment of gains. If, for example, we assume an amortization payment on
an earlier unfunded liability is paid at the beginning of period t, then the
gain at the end of the period is given by
Gain, = UFL,_~(I + i) - UFL, - am. payment,(1 + i).
Note also that the allocation of benefits within exit points should have
no effect on pension benefits guaranteed for particular exit points.

ALLOWING TRANSFERS BETWEEN COST CENTERS

Transfers between cost centers are more complicated. To do this, we
must imagine a network of nodes. There would be one node for each
completed length of service at each cost center. The nodes represent the
end of each length of service. The connecting segments represent the
periods of service.
A procedure for allocating benefits by cost center is available for any
allocation scheme. This can be done by separating the network into all
possible paths. For example, in Figure 1, AH is a path; AI is a path;
and so forth. But each path can be handled by Equation (2). An allo-
cation can be applied to the path to get t h e f t s for the path. So for each
of these paths we know not only fj's but also the cost center for each ~.
To get the normal cost for cost center r and length of service j, we must
sum the normal costs for all exit points and paths, but include only nor-
mal costs associated with a cost center.
The normal cost is

NCr,j,z Y. [fr.'.. 1
"" L (l + i)-'
P J
204 TRANSACTIONS, VOLUME XLVI

where
NCr,j,z = the current normal cost for cost center r, length of service
j , for the benefit at exit point z
fr,j.~.p = the allocation formula for cost center r, length of service j,
exit point z, and path p
n~.j,,,p = the number currently at cost center r, length of service j,
who reached where they are by path p and will exit at z by
path p
p = all paths that end at z.
The subscript r in f~j,z,p and nr,j,z,p is added for clarity. Once we are
given exit point z and path p, j will either be in cost center r or not. The
formula assumes thatfr,j,z,p and n~j,~,p will be zero if path p does not pass
through cost center r at length o f service j.

FIGURE 1
NETWORK ALLOCATION

Benefit=20,O00 Benefit=lO,000
( )
H .8 I ~ . 1 K9 Benefit=2,000
(
A
( 7
n =100 n = 300 n = 600

Pay=lO0,O00 Pay =50,000 Pay =20,000

 PROJECTED UNIT CREDIT METHOD 205

The total normal cost for a cost center is then

= E
z j

For an example of a network allocation, let's do an interest-adjusted

salary allocation for the network in Figure 1. The number entering at
nodes 1, 2, and 3 is given, as well as the starting pay. Each segment is
identified by a letter and shows the probability of moving along that
segment. We assume salary and normal costs are paid at the end of the
year and that transfers occur at the end of the year after salary and normal
costs have been paid.
The normal costs for segments E and G are simplest. At the end of
the first year, the normal cost for E is given by 300(0.1)(2000).
No interest adjustment is needed for segments E and G. For other
segments we need to allocate by the interest-adjusted salary. Let's as-
sume 7 percent interest and a 10 percent increase in the second period.
The number of employees, the salaries, benefits, and normal cost per-
centage for each of the remaining paths are as shown in Table 3.

TABLE 3
NORMALCOST PERCENTAGESIN A NETWORKALLOCATION

Ist-Year 2nd-Year NormalCost

Path Number Salary Salary Exit Benefit Percentage
i i i i
AH 64 100,000 110,000 20,000 9.22
AI 16 100,000 110,000 10,000 4.61
BJ 2 100,000 55,000 20,000 12.35
BK 18 100,000 55,000 10,000 6.17
CH 24 50,000 110,000 20,000 12.23
CI 6 50,000 110,000 10,000 6.12
DJ 24 50,000 55,000 20,000 18.43
DK 216 50,000 55,000 10,000 9.22
FJ 12 20,000 55,000 20,000 26.18
FK 108 20,000 55,000 10,000 13.09

The normal cost percentage for path AH, for example, is computed
from
20,000/[(100,000)(1.07) + 110,000].
To get the normal cost, we would apply the 9.22 percent to the salaries
of persons believed to be on path AH.
While this method can be done with any funding philosophy, shortcuts
would be needed for most applications, since the number of paths can
206 TRANSACTIONS, VOLUME XLVI

become large rather quickly. For a level benefit allocation, a nice sim-
plification results when all who reach a particular node have the same
number of years of service, n. The proportion of the individual person's
benefit paid for by the segment that feeds that node is 1/n, while (n-1)In
is passed down to comprise the benefit at lower nodes. The lower node
benefit equals the passed-down benefits from the connecting next higher
nodes, with each passed-down benefit reduced by interest and multiplied
by the segment probability.

NEGATIVE OR SMALL ACCRUALS

In many retirement systems, when a person reaches advanced age and
service, the present value of the benefit declines from year to year. In-
creased salary and length-of-service multipliers do not offset the declin-
ing number of years over which the benefit is payable. However, none
of the funding methods given in Tables 1 and 2 would show a negative
accrual; The traditional unit credit method would. If we want our ac-
counting to reflect negative or small accruals, it makes good sense to
switch to the unit credit method at some point. For example, the pro-
jected unit credit method using interest-adjusted salary might be used for
the first 25 years, using the false assumption that all who complete 25
years would then retire. Thereafter, the unit credit method would be used
to show the small or negative accruals beyond the 25th year.
But at what point should we switch? To resolve this, we need to con-
sider perceived value allocations, which are discussed in the next section.

PERCEIVED VALUE ALLOCATIONS

Except for negative or small accruals, mentioned in the previous sec-
tion, we have followed a single idea for allocating costs. Each exit point
allocates its costs among the people who are expected to reach that exit
point. Perceived value allocations adopt this same framework. We as-
sume we are at one of the exit points and want to allocate the benefits
at that exit point back to individual years of service of those people who
are expected to reach the exit point. How should this be done?
If we could show that the retirement benefit has a motivational equiv-
alent of cash in particular years, then it would make good accounting
sense to allocate each year's cash equivalent to that year. To do this, we
need to know the employee's implicit allocation of the benefit vj, at an
assumed exit point z to each year of service and the employee's perceived
 PROJECTED UNIT CREDIT METHOD 207

value of the benefit at z. The perceived value is the present value of

(Bz× vj), discounted back to time j using personal discount rates. The
personal discount rates and the employee's allocation are clarified below.
Personal discount rates are like interest rates, which tend to be higher
for young people than market interest rates (Gilman [5]). To find a per-
sonal discount rate, we must first find the cash amount at which a person
is indifferent to whether he or she receives the cash now or a deferred
amount later. Then we use the cash amount as the present value in a
present value equation and solve for the personal discount rate. ~
To find the employee's allocation, we ask how much it would take to
lure an employee away from a company now, assuming no retirement
benefits were retained. An interest-adjusted difference between the pay-
off amounts for two successive periods will reveal the part of the benefit
the employee implicitly allocates to a given period. This is given by
Payoffj(l + i) ~-j - Payoffj-i(1 + i) z-j+1
vj = (7)
Bz
where
payoffj = the amount an employee leaving at the end of period j would
accept in lieu of receiving a retirement benefit
vj = the employee's implicit allocation of the benefit to period
j. It's like the f1 shown earlier, but the sum of the vj's need
not equal 1.
i = the employer's investment rate of return.
A complete set of vj's is the perceived value allocation. Each vj is the
difference between two interest-adjusted payoff amounts, divided by the
benefit at exit point z. If the benefit at z, Bz, is a lump sum, then (ig-
noring tax considerations) payoffz would equal Bz and the v/s would sum
to 1. However, payoff, need not always equal B~.
Finding the payoff amounts would require a good deal of empirical
research. However, if we assume an employee at the margin is contem-
plating outside employment, we can develop the relationship b e t w e e n
the vj's, the personal discount rates, and the characteristics of the retire-
ment benefit in outside employment. We assume the salary and benefits
in outside employment are the same as that of current employment. We

'Gilman in his 1976 paper actually found the personal discount rates by fitting probit regres-
sions to joiners and nonjoiners in pension plans and getting maximum likelihood estimates for
the coefficients of the independent variables, including the personal discount rates.
208 TRANSACTIONS, VOLUME XLVI

do not require that the outside retirement plan be identical to that of the
current plan. However, we do require that persons retiring at z in outside
employment have the same benefit at z, Bz, if they had entered outside
employment at time 0.
Let us next define B°j to be the benefit that accumulates in outside
employment for a person who leaves current employment after j and
stays in the outside employment until z. Then the part of the benefit
attributable to potential outside employment during period j is given by
BOj_ ! - - B z o. j
f; = (8)
Bz
I f j is 1, Bz°o_! is identical to Bz. I f j is z, then B°i is zero. So thef~'s
sum to 1.
If we know the payoff amounts, we can solve for personal discounts
in the following equation, by starting at high values of j and working
down. Or knowing the personal discount rates, we can solve for the
payoff amounts.

Kj f j o B~
Kj(Bz - B°j)
P a y o f f j - (1 + d X -j - (1 + dj) z-j " (9)

where Kj is a scaling factor giving 1 if B~ is a lump sum and otherwise

is the ratio of two annuities at the age at z, the numerator evaluated with
dj and the denominator evaluated with i.

,-')
Substitute this back into the definition of vj. We obtain
J

vj = Bz(1 + d S -J - B~(1 + dj_~),-j+~

This can be rearranged as follows:

_f Kj(l + i) Z-J J-' [Kj(1 + i) "-j Kj_,(1 + i)'-J+']

Lo . (lO)

Equation (10) shows that the vj has two pans. The first part is the
f~' discounted by the ratio of 1 +i over 1+dj and the ratio Ks. The second
 PROJECTED UNIT CREDIT METHOD 209

part is a correction for the sum of earlierf~'s, which now must be dis-
counted for one less year with Kj and dj instead of Kj-I and dj_l. Also,
if dj is always equal to i, then vj = f'~.
To illustrate, consider the interest-adjusted salary allocation shown in
Table 1. This column can be constructed by accumulating the interest-
adjusted salary for a person exPected to reach exit point z and dividing
the accumulated sum from each period by the total accumulation. As a
consequence, the numbers in the column will give the f~'s for the case
in which the outside benefit is a defined-contribution plan. Similarly, the
level benefit column in Table 1 would give thef~'s when the alternative
in outside employment is a final-pay plan that uses a constant length-of-
service multiplier. Table 4 shows the vj's for each of these, assuming a
constant market rate of return of 7 percent and the employee discount
rates shown. Table 4 also assumes the same salary growth as Table 1.

TABLE 4
Two PERCEIVED VALUE ALLOCATIONSFOR YOUNG EMPLOYEES EARNING $25,000
(BASED ON INTERESTOF 7% AND OTHER ASSUMPTIONSOF TABLE 1)
Outside Benefit Is a Outside Benefit Is a
Final-Pay Plan Defined-ContributionPlan
Yearof Personal Discount
Service Rate (%)
1 16.3 0.05 0.006 0.032 0.004
2 15.6 0.05 0.009 0.035 0.006
3 14.9 0.05 0.012 0.038 0.009
4 14.3 0.05 0.016 0.041 0.013
5 13.6 0.05 0.022 0.044 0.018
6 13.0 0.05 0.027 0.045 0.023
7 12.4 0.05 0.034 0.047 0.030
8 11.9 0.05 0.039 0.048 0.034
9 11.4 0.05 0.045 0.049 0.041
10 10.9 0.05 0.053 0.O5O O.O48
11 10.4 0.05 0.060 0.052 0.057
12 10.0 0.05 0.062 0.053 0.060
13 9.6 0.05 0.067 0.054 0.066
14 9.2 0.05 0.072 0.055 0.073
15 8.8 0.05 0.077 0.057 0.079
16 8.4 0.05 0.080 0.058 0.085
17 8.2 0.05 0.067 0.059 0.074
18 7.9 0.05 0.076 0.060 0.084
19 7.6 0.05 0.076 0.061 0.086
20 7.4 0.05 0.068 0.062 0.079
Total 1.00 0.969 1.00 0.969

In both examples in Table 4, the perceived value allocation is more

back-loaded than thef~'s. Using the discount rates of Table 4, a level-
benefit allocation is insufficiently back-loaded, even when the outside
2 l0 TRANSACTIONS, VOLUME XLVI

benefit is a final-pay plan. This assumes the perceived value allocation

is used as an ideal standard. The projected unit credit method using in-
terest-adjusted salary is also insufficiently back-loaded here, although
it's closer to the perceived value allocation.
Also, note in Table 4 that the v/s do not sum to 1. This occurs because
d2o was equal to 7.4 percent, somewhat higher than the market interest
rate of 7.0 percent. As mentioned earlier, the K's would each equal 1,
if the benefit were a lump sum. For purposes of determining K20 in this
example, I assumed the benefit was paid in the form of a 25-year annuity
due.
The extent of back-loading in perceived value allocations is particu-
larly sensitive to the personal discount rates. Higher discount rates lead
to more back-loading. The discount rates used here are numbers that are
not out of the question. They were taken from Gilman's estimate for
persons 20 to 39 earning $25,000 per year, in 1975 dollars. Gilman's
estimates are expressed as continuous rates of real interest. To get the
rates used here, I converted using the formula
1 + dj = (l.05)e rj
where
rj = Gilman's estimate of the continuous real interest rate for the age
associated with length of service j
dj = the rate appearing for certain lengths of service in Table 4.
Gilman's results gave r / s for all ages below 20, for even ages between
20 and 30, and thereafter only for ages evenly divisible by 5. The re-
maining d / s were obtained by interpolation.
Gilman also gave discount rates for persons earning $5,000 and $15,000
per year. Table 5 was constructed by using the discount rates for persons
20-39 who earned $5,000 per year.
The v/s in Table 5 are also back-loaded. But what is most remarkable
about Table 4 is how little of the benefit gets allocated by the vj's. This
suggests that the perceived value allocation acts. more like a constraint
on the employer's allocation. To be consistent with the perceived value
allocation, each fj should equal or exceed the corresponding vj. The em-
ployer might satisfy this by setting t h e f t s to v/s that have been rescaled
so they sum to 1.
As I mentioned, the v/'s will equal thef~'s when the personal discount
rates equal the market interest rates. In this event a level benefit allo-
cation will be the perceived value allocation when the outside benefit is
 PROJECTED UNIT CREDIT METHOD 211

TABLE 5
TWO PERCEIVED VALUE ALLOCATIONSFOR LOW-AGE, LOW-INCOMEEMPLOYEES
(BASED ON INTERESTOF 7% AND OTHER ASSUMPTIONSOF TABLE 1)
Outside Benefit Is a Outside Benefit Is a
Final-Pay Plan Defined-Contr bution Plan
Year of Personal Discount
Service Rate (%)
1 31.8 0.05 0.000 0.032 0.000
2 31.0 0.05 0.001 0.035 0.000
3 30.2 0.05 0.001 0.038 0.001
4 29.5 0.05 0.001 0.041 0.001
5 28.8 0.05 0.002 0.044 0.002
6 28.1 0.05 0.003 0.045 0.003
7 27.4 0.05 0.005 0.047 0.004
8 26.8 0.05 0.006 0.048 0.005
9 26.2 0.05 0.008 0.049 0.007
I0 25.6 0.05 0.011 0.050 0.010
11 25.1 0.05 0.014 0.052 0.013
12 24.6 0.05 0.018 0.053 0.017
13 24.2 0.05 0.022 0.054 0.021
14 23.7 0.05 0.028 0.055 0.027
15 23.3 0.05 0:034 0.057 0.033
16 22.8 0.05 0.042 0.058 0.042
17 22.6 0.05 0.046 0.059 0.047
18 22.3 0.05 0.057 0.060 0.058
19 22.0 0.05 0.067 0.061 0.070
20 21.7 0.05 0.078 0.062 0.083
Total 1.00 0.446 1.00 0.446

a constant multiplier final-pay plan. Under the same assumption of ij

equals d~, projected unit credit using interest-adjusted salary will be the
perceived value allocation when the outside benefit is a defined-contri-
bution plan.
Normally, however, there will be a mixture of outside alternatives.
This problem can be solved by restating Equation (9) probabilistically:

E P,~ fkot KjBz

ij
Payoff~ = (1 + di)z-J (11)
where
lj indexes all the outside alternatives available at j
Plj gives the probability of each outside alternative at j
f°kl is defined like f~' in Equation (8) but for alternative l.
212 TRANSACTIONS, VOLUMEXLVI

The probability in Equation (11) deserves some discussion. We as-

sume we are allocating the benefits of someone who will leave the cur-
rent job at exit point z. However, at point j this person has the percep-
tions of similarly situated persons who may leave and accept employment
elsewhere. So PI i gives the probability of accepting employment along
path lj at length of service j for a person who temporarily is planning to
leave at j. The Plj's sum to 1. Note also that the probability of being
offered employment may be smaller along the more back-loaded paths.
However, the probability of accepting employment, given that it is of-
fered, may be increased for the more back-loaded plans.
Substituting Equation (11) back into Equation (7), we have
j j-| ]
v;=Ee,
lj
A-EP,,-,
t~_,
L j-
This can be rearranged as

vj = Pt;fjt (1 + d;) z-j

,-,

Equation (12) again has two parts. The first part, in line 1, gives the
expected fraction of the benefit (based on outside employment charac-
teristics) for period j, multiplied by a discounting factor. The remainder
of Equation (12) is a correction term giving y e a r j ' s expected amount of
the benefit accumulated by the end of j - 1 and yearj's discounting, less
year j - l ' s expected amount of benefit accumulated by the end of j - 1
and year j - l's discounting.
Table 6 illustrates Equation (12) for middle-aged (35-54), higher-than-
average income ($25,000) employees. The Pl's are constant, giving equal
probabilities of the outside benefit being a final-pay or a defined-
contribution plan. The personal discount rates above age 40 are set to 7
 PROJECTED UNIT CREDIT METHOD 2 13

percent. Gilman's results showed lower rates at those ages, but here I'm
assuming a 2 percent spread between inflation and the interest rates. To
be consistent with this assumption, I have restricted real interest rates to
no lower than the difference between 5 percent inflation and 7 percent
interest. To use a lower spread here would be comparable to assuming
that someone would be willing to borrow at the market rate and loan at
a lower rate. In this example, the v/s sum to 1, because d20 is equal to
the market interest rate.
Using the perceived value allocation in Table 6 as a standard, a level-
benefit allocation would be slightly too front-loaded and an interest-adjusted
salary allocation would be slightly too back-loaded.

TABLE 6
PERCEIVED VALUEALLOCATIONSFOR MIDDLE-AGEDEMPLOYEESEARNING $25,000
(BASEDON 7% Im'ERESTAND OTHER AssuMr'rlOt~OF TABLE 1)
Outside Benefit Has an Equal Chance
of Being a Final-Pay or a Def'med-ConUSbution Plan
Personal Discount
Year of Service Ra~ (%) Weighted f7
1 8.4 0.041 0.029
2 8.2 0.043 0.033
3 7.9 0.044 0.041
4 7.6 0.046 0.048
5 7.4 0.047 0.051
6 7.1 0.048 0.060
7 7.0 0.049 0.054
8 7.0 0.049 0.049
9 7.0 0.050 0.050
10 7.0 0.050 0.050
II 7.0 0.051 0.051
12 7.0 0.052 0.052
13 7.0 0.052 0.052
14 7.0 0.053 0.053
15 7.0 0.054 0.054
16 7.0 0.054 0.054
17 7.0 0.055 0.055
18 7.0 0.055 0.055
19 7.0 0.056 0.056
20 7.0 0.056 0.056
Total ! .00 ! .00

But the big question is whether we should treat the perceived value
allocation as a standard. To answer this question, reconsider Equation
(7). It is hard to quarrel with the payoff amounts in Equation (7), at least
not until we give more details. A payoff amount, in effect, tells us how
much an employee has allocated to the past. Also, the interest adjustment
214 TRANSACTIONS, VOLUME XLVI

and division by Bz in Equation (7) are straightforward. If an employee

invested adjacent payoff amounts, the difference between their accu-
mulations at time z as a ratio to Bz yields the vfs.
A potential problem emerges when we consider the payoff amounts
for an employee who is considering employment elsewhere, but who we
nevertheless assume will remain until the assumed exit point with the
current employer. The hypothetical employee at the margin is different
from an employee who has no thought of leaving. But this is generally
true in pricing. Employees who are not at the margin are paid more than
is necessary to retain them. So it seems reasonable to base our "prices"
on this hypothetical person at the margin.
I also see no disadvantages (other than practical ones) in using personal
discount rates. If we accept Equation (7), we must ask how it should be
evaluated. The use of personal discount rates seems quite reasonable.
Gilman's equations were developed by examining the extent of partici-
pation in pension plans in which there were matching employer contri-
butions as well as tax advantages. This examination of actual behavior
seems a sounder basis than would a survey eliciting respondents' stated
payoff amounts under hypothetical circumstances.
Equation (9) makes three further assumptions, however. First, it as-
sumes the current and outside salaries are equal. Second, it assumes the
pension benefits at z are equal for current and alternative employment,
given entry to either at time 0. Third, it assumes our hypothetical person
at the margin evaluates the payoff amount by using the alternative of
staying in outside employment until exit point z. While these are strong
assumptions, they are not quite so strong for a person we already assume
will be staying in current employment until z. If the outside benefits were
larger, then the employee should already have left, other things being
equal. If the outside benefits were less, then the employee is not at the
margin. If the employee is considering outside employment but not stay-
ing until z, it still seems reasonable to evaluate the alternative as if the
employee would stay until z in outside employment.
But to find a payoff for a hypothetical employee considering outside
employment, why not use a hypothetical internal payoff? The internal
payoff would remove the accumulated years and salary from the benefit
at z, but there would still be a benefit at z based on subsequent years
and salary. The difference between the external and internal payoffs is
 P R O J E C T E D UNIT C R E D I T M E T H O D 215

that a person really could leave, while an internal payoff is entirely hy-
pothetical. If a person considers leaving, he or she will consider what
is being lost by doing so. This leads back to estimating payoff amounts.
What about the person who considers leaving, when the overall ben-
efits outside equal the internal benefits (in present value terms using mar-
ket interest rates), but the pension/salary mix is different? Consider, for
example, a larger salary but a smaller retirement benefit. I suggest that
the extra salary increments be treated as an early retirement benefit. Neg-
ative salary differentials could be treated as an early negative pension
stream. This generalization would complicate our equations, but would
not erode the underlying rationale. A similar comment could be made
about employment paths that include periods of unemployment but
nevertheless yield the same overall benefit.
There is also a question about how free we are to set the f ] ' s , if we
fix Bz and require that the internal and external salaries be equal. Let us
define an internal allocation, f~, corresponding to f~ and based on the
internal benefit formula. Next, let us define g~'s to correspond to the
f~'s but based on salaries to date, rather than projected salaries. For a
career-average plan, the g~'s would equal the f~'s, but for a final-pay
plan, they would not. Now we may ask, What happens if
j ' j

k=l
Es ?
k=l

This would imply that an employee could leave current employment

at j and get a total benefit at z larger than Bz based on the current em-
ployment until j and outside employment after j. For such an employee,
the perceived value would be more likely to be based on the sum of the
g~'s than on the sum of the f~'s. So we should require that
J J

E g~ ~
k=l
Eft"
k=l
(13)

Where the current employment has a career-average plan, this is not a

difficult requirement. Because of the back-loading inherent in career-
average plans, the sum of the early g~'s will be small. For a final-pay
plan, this is normally not an imposing constraint, since the early g~'s will
be less than thef~'s. In any case Inequality (13) requires that we treat
216 TRANSACTIONS, VOLUMEXLVI

outside alternatives where the inequality is not met in a special way,

namely, to substitute the g~'s for thef~'s in Equation (11) for those lengths
of service not meeting the !nequality. 6
But then why not have a constraint that prevents someone from leaving
to get a smaller total benefit? In other words, why not make Inequality
(13) an equality? Well, that would discard one of the advantages of per-
ceived value allocations. During the early years of employment, an em-
ployee may feel he or she has earned not only benefits dictated by the
formula, but also some share of later benefits, since the later years' ben-
efits are not possible without the earlier ones. The question is, How
much? Equations (9) and (11) answer the question by asking how much
an employee would have to be paid off to jump into alternative em-
ployment at mid-career.
There are two theoretical disadvantages of the perceived value allo-
cation, however. First, the reader may have noticed a shift in the mean-
ing of "payoff" from Equation (7) to Equation (9). In Equation (7), it
isn't clear what "payoff" means. In Equation (9) we operationally define
"payoff," but in a way that may not entirely correspond to the employ-
ees' feelings on the matter. For example, if we computed the Equation
(9) payoffs when the current employment had a defined-contribution plan,
the payoff need not correspond to the accumulated investment value. The
employee may see the accumulated investment value (less discounts based
on the fact that the employee cannot immediately get the money) as closer
to what the person thinks he or she has. Second, the allocation depends
on the mix of outside alternatives. As a consequence, the "perceived
value" can change as the mix varies. While the resulting allocation has
these two theoretical weaknesses, there is still likely to be an overlap
between the payoff amounts under Equations (7) and (9). It is this over-
lap that provides the justification for perceived value allocations.
The perceived value allocation is designed to answer this question: "If
a particular retirement system were eliminated, with what cash outlays
could the employer replace it so that an employee does not feel a gain
or a loss?" For the person who really is at the margin and is really con-
sidering outside employment until z, a perceived value allocation ap-
proximates this ideal. For others, it seems reasonable to base "prices"
on this hypothetical person at the margin.

6When no outside alternative satisfies the inequality, then the g~'s are all that are needed
for that length of service.
 PROJECTED UNIT CREDIT METHOD 217

There are, however, some practical problems. The parameters needed

for a perceived value allocation are substantial. Personal discount rates
are needed for all ages and income levels as well as for other charac-
teristics that may be important. In addition, the mixture of benefits avail-
able in outside employment and the probabilities associated with each
must be known. Because all these can change, the number of parameters
necessary becomes unwieldy.
Because of this parameter estimation problem, perceived value allo-
cations cannot be used routinely. I suggest that they be used only in a
broad way to determine how to allocate benefits. In particular, Tables 4
and 5 suggest that something more back-loaded than a level benefit al-
location is appropriate for young or relatively poor employees. While a
level benefit allocation is appropriate for older, somewhat wealthier groups
when the outside benefit alternative is a final-pay plan, this represents
one extreme. In other cases, something more back-loaded is more
appropriate.

NEGATIVE OR SMALL ACCRUALS RECONSIDERED

In the section on negative accruals, there is an apparent contradiction
between my advocacy of the projected unit credit method with multiple
exit points and my advocacy of switching to the traditional unit credit
method when negative or small accruals appear for long-tenure employ-
ees. How do perceived values help resolve this seeming contradiction?
Suppose a person is retirement-eligible and earning no additional ben-
efit under the formula. Then this person should not accept a payoff that
is less than what he or she would get by leaving. So let's expand Equa-
tion (9) as follows

Payoffj = Max
8z°j) ]
(I + ~)'~-J ' Kj,jBj

where
Kzj is the same as our earlier Kj, the ratio of two annuities at z, the
numerator evaluated using the personal discount rate at j and the
denominator evaluated using the market interest rate
is defined similarly but for annuities beginning at j.
To simplify the argument, assume the employer wants to rescale the
vj's so that they sum to 1. In this case, the K's would equal 1. When
the Bi's begin to decline, Bj will be larger than the discounted B~-B°j.
218 TRANSACTIONS, VOLUMEXLVI

So the payoff is simply Bj, and the difference between two successive
interest-adjusted payoff amounts is
Bj+, - Bj(I + i).
But this is the normal cost under the traditional unit credit method,
without the mortality and continued service factors. 7 So switching to unit
credit at some point more closely follows the perceived value of the
benefit to an employee.
Most fundamental is to allocate to the past the amount needed to lure
an employee away from a company now, assuming no pension benefits
are preserved. If an employee has reached retirement eligibility, the amount
allocated to the past should not be less than the full value of the benefit.
It could, however, be more in the case of a plan that is back-loaded by
virtue of the formula, by virtue of late career salary growth, or by an
interaction of the two.
So the switch to unit credit occurs when two tests have been met. First,
the employee must be retirement-eligible. Second, the present value of
future normal cost contribution under the traditional unit credit method
must have reached the point at which it is less than the present value of
future normal cost under the projected unit credit method. Not switching
to unit credit at this point implicitly allocates an insufficient part of the
benefit to the past.
This can lead to negative normal costs. Some may question whether
these should be permitted. An alternative would be to pay no normal
cost at that point and later recognize a gain. However, ignoring any reg-
ulatory constraints, it seems better to recognize now what will occur if
all our assumptions are met. That would imply that it is better to rec-
ognize a negative normal cost.

USE OF CURRENT EMPLOYMENT BENEFIT FORMULA

The preceding sections have already presented the main ideas of this
paper. However, an interesting related issue is the use of current em-
ployment benefit formulas for allocating the benefit to individual years
of service. Is this use of the benefit formula here deceptive?

7For perceived value allocations, we allocate the benefit for persons expected to reach z,
so the mortality and continued service factors are present indirectly.
 PROJECTED UNIT CREDIT METHOD 219

The chief reason for ignoring the characteristics of the plan is that they
are largely irrelevant. 8 We are already allocating one exit point at a time
and using the benefit formula for the benefit at each exit point. This
approach takes into account the probabilities of leaving at each exit point.
When exit points are combined, ~ome other allocation could reemerge,
but that would depend on the probability of leaving at each exit point
and on the benefit at each exit point. A similar view on combining sep-
arate exit points was expressed by Atteridg et al. [2], 9 who suggested
that we "allocate equal portions of the benefit associated with each exit
age to the employee's period of service up to that age." It is possible
that, when using a single assumed or typical exit point, the benefit for-
mula may be a good proxy for a reasonable accumulation under multiple
exit points. However, this would have to be demonstrated.
A second reason for not using the benefit formulas within exit points
goes back to perceived value allocations. The employee's perception of
the benefit accumulation need not correspond to that of the benefit for-
mula. In the notation of the perceived value section, the v/s need not
equal the g~."s
j • For an employer, the employee's perceptions become
important when trying to find a cash value for the motivating influence
of the pension plan.
Another approach to the same question is to ask, What would happen
in a divorce pension case? Consider the first column of Table 1. Because
this allocation is based on salary unadjusted by inflation, its allocation
is the same as the benefit formula under a career-average plan. But could
you convince a judge that only 1.5 percent of the benefit was earned in
the first year, as shown by Table 1? (An ex-spouse married to someone
for the last 19 years of a 20-year career is someone who could benefit
from such an allocation.)
The superficial answer is that the benefit formula dictates this. The
peculiar result is caused by the peculiar retirement benefit. We should
not blame the accounting procedure for the benefit design.
The counterargument is that during the first year the retiree earned not
only a share of the first-year's benefit, as dictated by the benefit formula,

sit is, of course, relevant for determining the benefit at each exit point and for satisfying
Inequality. (13), as discussed under perceived value allocations.
9Atteridg et al. were not, however, recommending that any consideration be given to per-
ceived values or that the allocation take into account anything but service time.
220 TRANSACTIONS, VOLUME XLVI

but also a share of subsequent years' benefits, since a second year would
not be possible without a first year.
Suppose we have cliff vesting at 20 years? Surely we wouldn't suppose
that spouses who precede the spouse at the time of vesting have no right-
ful share? Similarly, we can see strict application of the benefit formula
as leading to a series of small cliffs, or steps. We should, I think, give
ex-spouses some share of those later steps, which would not be possible
without the earlier steps.
But the benefit allocation for an employer should not be all that dif-
ferent from that used in divorce. Perhaps a divorce court should go fur-
ther than an employer in apportioning part of the benefit to lean years
during which a business or skill was being developed. For a divorce
court, perceived values may not be relevant. But both the divorce court
and the employer should ideally consider exit points one at a time. Once
we assume a particular exit point, the operation of the benefit formula
plays only a small role.

DISCUSSION
Perceived value allocations provide a justifiable basis for allocating
the benefit at an assumed exit point. As discussed earlier, a switch to
unit credit may be needed for high-tenure employees, but in that case
unit credit is consistent with the underlying rationale for perceived value
allocations.
Because of the difficulties in carrying out a perceived value allocation,
however, I doubt whether this method will be used except in the rarest
of circumstances. We may, however, work backwards. We may find a
perceived value allocation that is appropriate for a comparable situation.
Then we may find some other allocation that is reasonably consistent
with the perceived value allocation in terms of back-loading.
An allocation based on inflation-adjusted dollars has some appeal in
this respect. While the back-loading for this method seemed high in Ta-
ble 1, it is really quite reasonable in view of the perceived value allo-
cations shown in Tables 4 and 5.
An allocation based on interest-adjusted salary was insufficiently back-
loaded at times, but did a reasonably good job for the examples given
in Tables 4, 5, and 6. Its back-loading is perfect when the outside al-
ternative is a defined-contribution plan and the personal discount rates
equal the market rates. An added selling point is the funding philosophy.
 PROJECTED UNIT CREDIT METHOD 221

All who are expected to reach a particular exit point by a particular path
pay the same percentage of salary. While this method is more compli-
cated than the level-benefit method, few additional inputs would be re-
quired once we are dealing with packaged software. In most situations
this method would not be legally permitted; however, we hope that this
could change.
On the other hand, a level allocation has the advantage of simplicity
and a longer history. It did a reasonably good job for the example given
in Table 6. Its back-loading is perfect when the outside benefit is a final-
pay plan with a constant length-of-service multiplier and the personal
discount rates equal the market interest rates. In addition, a simplified
node-by-node approach can be used for network allocations.

SUMMARY
1. Perceived value allocations can be developed by using personal dis-
count rates and the characteristics of alternative pension plans in
outside employment. This results in a justifiable basis for allocating
pension benefits at an assumed exit point. Such allocations may be
impractical because of their unwieldy parameter estimation require-
ments, but they may be used in a crude way to evaluate other al-
locations. They do not always allocate all the benefit, but in that
case they give a minimum percentage of the benefit to be allocated
to various lengths of service.
2. The projected unit credit method with the benefits allocated by in-
terest-adjusted salary has a particularly appealing funding philoso-
phy. A constant percentage of salary is applied to all persons ex-
pected to retire at a particular exit point. In terms of back-loading,
it does a reasonably good job in many situations. Its allocation i s
ideal when the outside benefit alternative is a defined-contribution
plan and personal discount rates are identical with market interest
rates.
3. The projected unit credit method with the benefits allocated by in-
flation-adjusted salary also has a somewhat appealing funding phi-
losophy. In terms of back-loading, it seems superior to the interest-
adjusted s~ilary allocation for young or low-income employees.
4. The projected unit credit method with the benefits allocated equally
to all lengths of service has a straightforward funding philosophy.
It gives a reasonably good allocation for employees who are not too
222 TRANSACTIONS, VOLUME XLVI

young or poor. Its allocation is ideal when the outside benefit al-
ternative is a constant-multiplier final-pay plan and personal dis-
count rates are identical with market interest rates. In addition, a
simplified network allocation is possible.
5. Each o f the four projected unit credit methods discussed here allo-
cates one exit point at a time. The allocations must be combined to
get a single normal cost for each length of service. However, a sin-
gle percentage of pay, a weighted normal cost, can be developed
for different cost centers.
6. When there are transfers between cost centers, a method is available
to allocate benefits by using the same funding philosophy.
7. Most funding methods will not show negative accruals for employ-
ees at high lengths of service. Switching to unit credit at high lengths
of service makes it possible to recognize negative or small accruals.
8. There are good reasons for not using the current employment benefit
formula for allocations within exit points.

REFERENCES
1. ANDERSON,ARTHURW. Pension Mathematics for Actuaries. Needham, Mass.:
Arthur W. Anderson, 1985.
2. ATI~ERIDG,JOHNW., ETAL., "The Projected Unit Credit Cost Method," The Pen-
sion Forum 6, no. 1 (September 1991): 1-27.
3. DEPARTMENTOF DEFENSE OFFICE OF THE ACTUARY. Valuation of the Military. Re-
tirement System, September 30, 1990. Arlington, Va.: Department of Defense
1991, pp. H-6 and H-12.
4. Federal Tax Regulations, 1991, "Regulation 1.412(c)(3)-1." St. Paul, Minn.:
West Publishing Co., 1991.
5. GILMAN,HARRYJ. "Determinants of Implicit Discount Rates: An Empirical Ex-
amination of the Pattern of Voluntary Pension Contributions of Employees in
Four Firms," (CNA)76-0984.10. Arlington, Va.: Center for Naval Analyses,
August 1976.
6. HALL, WILLIAMD., AND LANDSITTEL, DAVID L. A New Look at Accounting for
Pension Costs. Published for the Pension Research Council, Wharton School,
University of Pennsylvania by Richard D. Irwin Inc., 1977.
 D I S C U S S I O N OF P R E C E D I N G P A P E R

INGER M. PETTYGROVE:
I found Mr. Giesecke's paper both interesting and informative. In reading
it, however, a question occurred to me. The section on network allo-
cations mentions that shortcut methods would be needed if projected unit
credit with interest-adjusted salary were used. Does the author
have any suggestions along these lines? I look forward to reading the
response.

KENNETH A. STEINER:
Mr. Giesecke has written an interesting paper that illustrates his point
that "there are as many ways to fund a retirement benefit as there are to
finance a house." I agree with his statement that "to say that one is
wrong is analogous to saying that a house cannot be financed with dif-
ferent payment schedules." However, I do believe that criteria do exist
for selecting whether one actuarial cost method may be better than an-
other for meeting a plan sponsor's funding objectives for a particular
plan. For example, one criterion I use is whether the method produces
a reasonable relationship between the plan's actuarial accrued liability
(or "asset target") and the plan's actuarial present value of accrued ben-
efits when consistent assumptions are used to develop both values. In
the discussion that follows, I refer to the ratio of these two items as the
"asset target ratio."
Assuming the plan sponsor's funding objective is to accumulate suf-
ficient plan assets to cover plan termination liability at all times with
reasonable certainty but without overfunding the plan, a reasonable asset
target ratio might be 1.25-1.75. A narrower reasonable asset target ratio
range for a plan sponsor with this objective could be developed by ex-
amining such factors as the relationship between the ongoing plan in-
vestment return assumption and the interest rate used to determine plan
termination liability, volatility of plan assets and liabilities, plan de-
mographics, and so on. If an actuarial cost method does not produce a
reasonable asset target ratio, it is probably a good time for the actuary
to initiate discussions with the plan sponsor about whether the sponsor's
funding objectives can be better accomplished by using a different ac-
tuarial cost method.

223
224 TRANSACTIONS, VOLUME XLVI

Given current Internal Revenue Service regulations and Financial Ac-

counting Standards Board interpretations, Mr. Giesecke's inflation-
adjusted and interest-adjusted methods appear to have limited potential
application in the U.S. (funding for final average pay plans not governed
by Section 412 of ERISA and possibly career average plans). I do believe
that his cost methods can, under certain circumstances, produce reason-
able asset target ratios. However, since his methods fail to consider the
actual pattern of benefit accruals, they are much more likely to produce
unreasonable asset target ratios for specific situations than the projected
unit credit methods currently anticipated by the Internal Revenue Service
and the Financial Accounting Standards Board.

(AUTHOR'S REVIEW OF DISCUSSIONS)

GERALD LEE GIESECKE:
I thank the referees and Ms. Pettygrove and Mr. Steiner for their help-
ful comments.

Inger Pettygrove
Ms. Pettygrove asks about shortcut methods for network allocations
under the projected unit credit cost method with interest-adjusted salary.
One shortcut would be to do something similar to what I suggested for
level benefit network allocations. This essentially divides the benefit at
each node into two pieces. Part is allocated to the segment that precedes
the node. The rest is passed down to the feeding node where that segment
originates.
In a level benefit allocation, the fraction of the benefit 1/n is allocated
to the feeding segment, while (n-1)In is passed down the feeding node.
The benefit for reaching the feeding node equals the passed-down benefit
from the connecting next higher nodes, with each passed-down benefit
reduced by interest and multiplied by segment probability.
The same procedure can be applied to the projected unit credit cost
method with the benefit apportioned by interest-adjusted salary. In place
of l and ( n - l ) , we have the interest-adjusted salary of the preceding
segment versus the expected interest-adjusted salary accumulated at the
feeding node.
As an example of the shortcut method, consider Figure 1 in the paper.
For a person on segment H, we would pay:
 DISCUSSION 225

20,000 z 110,000
110,000 + 1.07 × {[80/(80 + 30)] × 100,000 + [30/(80 + 30)] × 50,000}
The numerator is the amount needed (20,000) multiplied by segment H
pay (110,000), which is the numerator of the apportioning fraction. The
110,000 is not multiplied by an interest adjustment, since it is paid at
the end of the year. The denominator is the 110,000 plus the expected
accumulated interest-adjusted salary from node 4. The shortcut method
gives a normal cost of $10,869.08 for segment H. This differs from the
weighted all-path normal cost of segment H, which equals $11,043.
As a variant of the shortcut method, certain mainstream paths could
be separated and calculated independently. This may be overkill, but the
result should more closely approximate the all-path method. With a level
benefit allocation, the shortcut method would equal the all-path method,
so isolating certain mainstream paths would make no difference.
/
Kenneth Steiner
Mr. Steiner suggests we use any latitude we have in selecting the ac-
tuarial cost method to get a desirable "asset target ratio." This seems
like a reasonable goal to me; however, it was not my goal.
My goal was to develop a scheme that makes pricing sense. In allo-
cating pension cost to years of service, we change the price for using an
employee's time. This can affect decisions on whether we have more
generals or more privates as well as on a host of other tradeoffs. One
can of course use a funding method that tries to isolate pension costs
from such considerations. But if you let pension costs influence person-
nel decisions, some actuarial cost methods make more sense than others.
By looking at perceived value allocations, I have tried to formalize
the notion that pension benefits act as more of an incentive to older work-
ers than they do to younger workers. This suggests that our accounting
be more back-loaded so that these incentives are allocated just as they
are perceived. More work is needed, however, since I assumed the pref-
erences of a hypothetical employee who was at the margin between leav-
ing and staying. Where this is not true, the right side of Equation (9) in
the paper would no longer hold. Equation (7)--which still holds--might
then suggest a more front-loaded benefit, particularly if the salary, pen-
sion, and other benefits available in outside employment are less than
those available in current employment.
226 TRANSACTIONS, VOLUME XLVI

Perceived value allocations still leave considerable latitude in allocat-

ing pension costs to years of service, in part because the perceived value
allocation does not always allocate all the benefit, in part because the
perceived value allocation can allocate some of the benefit to the time
of entry, and in part because we do not know the details of the perceived
value allocation. I have advocated an interest-adjusted salary allocation
as a reasonably good proxy for perceived value allocations in m a n y - -
but not all--situations. I am also tempted to use an interest-adjusted
salary allocation to distribute any residual, since it proportionately raises
the costs of all persons expected to leave at a particular exit point by a
particular path. But this point needs further study. So for the time being,
at least, I agree that there should be considerable latitude in how we
allocate the pension benefit.