MANAGEMENT SCIENCE

Vol. 25, No. 4. April 1»79

Fmudin

AGGREGATIONS OF SUBSIDIARY FIRMS FOR

MINIMAL UNEMPLOYMENT COMPENSATION
PAYMENTS VIA INTEGER PROGRAMMING!
HARVEY M. SALKINJ AND C . H . L I N §

In Ohio, as well as several other states, annual unemployment compensation payments paid
by a corp<»fttion can be minimized by solving a set partitioning problem which has all
possible nonzero binary columns. The scenario, along with the model and an illustration are
given. Some solution techniques and actual results, taken from a consulting effort, for a few
Northeast CHiio corporations are also mentioned.
(PROGRAMMING-INTEGER, APPLICATIONS; GOVERNMENT-TAX POLICY; IN-
DUSTRIES)

1. IntrodiKtion
In Ohio, as well as in several other states, the unemployment compensation system
is financed by a payroll tax levied on employers of one or more employees and is
administered by the Bureau of Employment Service (BES) [1]. The administrator of
the BES maintains a separate account for each employer, and credits each employer's
account with aU contributions which he has paid on his own behalf, the interest
obtained from his contributions, and charges to the account, the benefits based on
renumeration paid by the employers. The tax rate to the account is based on an
experience system, and increases in a stepwise fashion as the ratio (RT), of the
account balance to employer's three year average annual payroll, decreases. The tax
rate is also adjusted according to the value of the compensation fund in relation to a
given minimum safe level [4].
Suppose we let
P be the average annual payroll of an employer,
R be the balance in the reserve fund of the employer,
RT be the ratio of the balance to payroll expressed as a percent (i.e., RT = R/P),
fiRT) be the individual contribution rate (from a state supplied rate schedule),
CT be the contribution tax rate (i.e., CT-f(RT)), and
EP be the expected annual payroll.
Then the annual contribution due is equal to EP X CT.
As of July 1, 1974, the compensation fund in Ohio was such that the net tax rate
schedule was as in Table 1.
The law also permits the establishment of joint accounts by a group of employers in
a common industry [1]. The calculation of the joint contribution is the same as the
calculation of an individual account computed from an aggregate average annual
payroll and balance. This option is especially important if the employers are sub-
sidiaries of a parent corporation wishing to minimize its total tax payment. The total
payment amount is dependent on ^ e company's accounting conventions for purposes
of the Fund. In particular, each may contribute separately, all together, or in various

* AB Notes are rrfereed.

'*' ^xe]^ied by P^» Koiesw, fotmet DqMutmental Eitttor; received AptQ 1978. This paptt hta been with
the anttors 2 iwnflis tot I reviaon.
'Caw Westem Reserve
* Staiidcrd O& Con^airy,
405
0025-1909/^/2504/0405101 J5
9. Tiw tattitut of Masacnoeat Scieaoct
 406

TABLE 1
Indtd^ml Rate Schedule for Ohio in 1974
Contribution Contribution
Reserve Ratio (%) Rate(%) Reserve Ratio (%) RateC%)
RT /(/jr) RT f(RT)
Non-rated Erq^yers 3.2
RT<, - 1 . 0 4.0 6.0 <.RT< 6.5 1.8
-i.o</{r<o.o 3.9 6.5 < RT< 7.0 1.6
o.o</{r<i.o 3.8 7.0 <RT< 7.5 1.4
i.o</jr<i.5 3.7 7.5 <RT< 8.0 U
\.S<RT<1.Q 3.6 8.0 < /fr < 8.5 1.1
2.0 <RT< 2.5 3.4 8.5 <RT< 9.0 1.0
2.5</!r<3.0 3.2 9.0 <RT< 9.5 0.9
3.0</Jr<3.5 3.0 9.5 <i?r< 10.5 0.8
3.5</jr<4.0 2.8 10.0 < /!r< 11.0 0.7
4.0 < * r < 4.5 2.6 10.5 </jr< 11.5 0.6
4.5 <iRT< 5.0 2.4 U.O < RT < 12.0 0.5
5.0</ir<5.5 2.2 U.5<i RT 0.4
5.5<i{r<6.0 2.0 12.0 < RT 0.3
EXAMPLE
P -$4,533,643.00
R'$ 422360.81
£/>-$5,900,000.00
RT-R/P-9.76%
-0.5% (from Table 1)
and CT-f(RT)-O.i%
The total annual contribution is EP X Cr-$47,200.00.

aggregations. Therefore, the problem is to determine how the subsidiaries should be

aggregated so that the corporation's total annual contribution is minimized. The
problem can be formulated as a set partitioning problem [2] which contains a row for
each subsidiary and all nonzero binary columns.
2. H K Model
Let the corporation's subsidiaries be numbered as {1,2, ..., m), and let S
=» ( S j , . . . , S^}, be the set of all possible combinations of the m subsidiaries, where
n »= 2" - 1. Corresponding to each Sj in S, there is an nt vector (column) a, with
components a^ == 1 or 0 depending on whether the ith subsidiary is in Sj (a,j == 1) or
not (Oy " 0); associated with S is a cost c. Thus, we have the set partitioning problem
(SP):
n
Minimize z -= 2 'y^y (the total cost)

subject to (guarantees each subsidiary is in exactly one aggregation)

and Xj = 0 or 1 (y = 1, 2 , . . . , n),
where e is an m vector of all ones, and a, — (a,-,) is an m binary column.
EXAMPLE (Ohio Corporation, 1974)
Average Payroll Reserve Fund Estimated Payroll
Subsidiary (P) (R) {EP)
1 $4,533,643.00 $422,360.81 $5,900,000.00
2 9,900.34 200,000.00
3 84,791,00 9,552.46 105,000.00
4 6«),855.75 100,984.94 780,000.00
 NOTE 407

As these are four subsidiaries, m =* 4, and 2" - 1 = 15, meaning that 15 binary
columns can be generated corresponding to all possible aggregation schemes. Notice
that (e.g.) the column a^ =° [1010] means that subsidiary 1 and 3 will be aggregated
when Afg = 1.
For each a, its cost Cj is computed using Cj =^ EP x CT, where CT=^f(RT) and
RT = R/P. The terms CT, RT, and EP may, of course, represent aggregated figures.
For example, the calculation for c^ gives
R =442,360.81 + 9,552.46 = 451,913.27,
P =4,533,643.00 + 84,791.00 = 4,618,434.00,
/?r = /?/P = 0.09785,
CT = f{RT) = 0.008 (from Table 1),
EP = 5,900,000.00 + 105,000.00 = 6,005,000.00,
C 6 « £ ' i ' x c r = $48,040.
Similar calculation gives the remaining payment figures listed below.

c, Cl Ci CA Cf, ^8

47,200 2,800 525 2,340 48,800 48,040 46,760 3,050 2,940

^10 ^12 ^13 ^14

2,655 49,640 48,160 47,495 3,250 48,895

3. Solution Tedmiqiies and Computations

In the classical set partitioning problem with m rows, the number of columns is
usually less than the maximum number or 2" - 1. However, in this particular
formulation we have all the columns and the program is termed a complete set
partitioning problem [4].
It has b « n reported [3] that the complete set partitioning problem can often be
solved as a linear programming problem using the simplex method [5]. Several smaller
problems representing data from Ohio corporations were solved as part of a consult-
ing effort by substituting the constraints jc^ > 0 for x, = 0 or 1 (y = 1 , . . . , n). Each
computer run required less than a second of UNI VAC 1108 computer time using the
SCA4 code [6], and the largest problem solved had 10 subsidiaries. Every optimal
solution tumed out to have only integer values for the variables. Naturally, the
simplex method will not, in general, produce integer solutions.
A highly efficient, specialized integer programming algorithm for the complete set
partitioning problem has been developed [4]. Problems, reflecting data from several
Ohio corporations having as many as 20 subsidiaries, have been solved optimally in at
most two minutes of UNIVAC 1108 computer time. In those case studies, it tumed
out that every solution resulted in an actual reduced tax payment, with as much as a
20% reduction (or maximum of $60,000) in total corporate tax payments, when
compared with the previous year. All results mentioned here and discussed in [3] and
[4], were an out^owth of a consulting study. As far as the authors ioiow, the optimal
a^regations, given to the corporations, were used.'

' Early contributions to the sttufy by Burton V. Dean, Dq»rtmental Editor, and Stuumu Morito are

1. Ctao, Ett^^ment Serdee, Oim Reviied Code.

X GmmaxL, R. MOJ MENHW^K, G., "TTje Set Itetitimung Problem: Set Covering with Equality
C(»Hndats," (^enaions Ito., VtA. 17, No. 5 (1969), i^. 8^-856.
 NOTE

3. SAUUN, H . , DBAS, B. AND LIN, C , "^^grq^atiag Corporation Subsidiaries Under a Given Tax
Structure," Technical Memorandum No. 360, Department of Operations Research, Case Western
Reserve University, May 1975.
4. IJN, CRIEN-HUA. "Corporate Tax Structure and a Special Class of Set Partiticming Problems," E>octoral
Thesis, Case Westem Reserve University, Qeveland, Ohio 44106, June 11, 1975.
5. DANTZIO, G . D . Linear Programming aid Extensions, Prentice-Hidl, Englewood Cliffs, N. J., 1960.
6. SALKIN, H . Integer Progiwnming, Addison-Wesley, Reading, Mass., 1975.

MANAGEMENT SOENCE
VoL 23. No. 4. April 1979
PruutdiH U.SA.

Liters should he addressed to the Editor, Graduate School of Business, 401 Uris HaU,
Cdumbia University, New York, New York 100i7

An Ex^Kd<w erf Gi&m's Metfaodcrfogy tar Designing Abortioa Oiidcs

In [1] Gitlow presented a methodology for determining the optimal design of an
abortion clinic, combining a simulation model with a steepest ascent algorithm. In his
conclusion the author stated as a main limitation of his methodology: " . . . the
necessary number of clinic ccmfigurations increases exponentially (2"), rendering the
methodology impractical when considering a large number of variables". However, I
disagree completely with the author's modesty! A large number of variables in
so-called Response Surface Methodology (RSNf) which includes the steepest ascent
search algorithm, can be handled as follows: Ek) not simulate all 2" combinations, but
select a much smaller subset, e.g. only n + 1 combinations, applying statistical
methodology of experimental design. Actually experimental design is a standard part
of RSM; see [2], [3]. Experimental designs are also discussed in Gitlow's own
references. Hicks and Mendenhall.

1. Gnww, H. S., "A Metiioddc^ for Determining the Optimal D e s ^ ol a Free StaiKUng Abwticm
Clisk," Managmem Sd., Vol. 22, No. 12 (Ai^iot 1976), pp. 1289^1298.
2. KuEUNBN, J. p. C Statistical Tecknupm in Simidtaim (in two v<dnmesX Marcel D d c t o , New Y < ^
1974-1975.
3. MEYKS, R. H., Re^mae Surfatx Methodology, AUyn and Baccw, Bostcm, Mass., 1971.
J. P. C. KLEUNIN
KathoHeke Hogeschool Tilburg
Hogesehoollaan 225
5000 LE Tilbtirg
Netherlands

0Q25-1909/79/29M/O4QWI.25