Académique Documents
Professionnel Documents
Culture Documents
(TAS) is used by over 40 percent of life- fair price to individuals and corporations,
insurance companies in the US. to generate reasonable profit for share-
Towers Perrin-Tillinghast employs a holders (or dividends for mutual compa-
suite of stochastic financial-planning mod- nies), and to maintain capital adequacy
els to provide guidance to managers of and risk management so that the company
pension plans and insurance companies. remains solvent in the long run.
The system simulates asset-and-liability Evaluating pension plans and insurance
decisions across a long-term, multiperiod companies is complicated because of the
planning horizon—typically five to 15 long time horizons, the number of parties
years. The system has been under devel- with diverse interests who are affected,
opment since 1991 and is now employed and extensive government regulations. In
in over 16 countries, mostly European and most developed countries, regulations re-
English-speaking countries. It is the most quire an annual actuarial analysis. Actuar-
extensive global investment system for ac- ies evaluate the soundness of a pension
tuarial studies; its distinguishing features plan or an insurance company and can re-
result from a collaborative eight-year ef- quire cash contributions from the com-
fort between an operations researcher pany for a pension plan in deficit or from
(Mulvey) and Towers Perrin’s worldwide stockholders or outside investors for an
consulting staff. undercapitalized insurance company. Such
The asset-and-liability management contributions (or their absence) can have a
(ALM) system applies to two major do- major impact on a company’s profits or
mains: (1) designing and managing pen- losses. For example, several aerospace
sion plans, and (2) integrated financial risk companies have improved their profits by
management for insurance companies. In 40 percent over the past five years through
both cases, the system simulates an asset improved pension management. The actu-
policy with liability decisions to maximize arial profession closely monitors its mem-
the company’s wealth or pension-plan sur- bers and certifies them through periodic
plus, while maintaining a safe level of tests, educational requirements, and other
operations. The optimization module se- means.
lects the best dynamic policies for the in- The Towers Perrin stochastic investment
surance company or the pension plan. system is employed widely. Several hun-
For defined-benefit pension plans, the dred large US and international companies
primary goals are to design the plan and run the system for pension plans, approxi-
to manage the pension surplus so that the mately 40 percent of US life insurance
company will meet its obligations to its companies use the related Tillinghast TAS
employees, to minimize the company’s system, and Tillinghast has a major pres-
contribution over time, and to maximize ence in property-casualty insurance. Mul-
the growth in the plan’s surplus. tinational companies, such as Unilever and
For insurance companies, the three pri- IBM-Europe, employ the global system for
mary goals are to help the insurance com- modeling their worldwide pension plans
panies to provide insurance products at a on a consistent basis.
January–February 2000 97
MULVEY ET AL.
INTERFACES 30:1 98
TOWERS PERRIN
January–February 2000 99
MULVEY ET AL.
are calculated by actuaries who estimate There are numerous candidates for the ob-
the life spans of employees, their salaries jective function. In addition, we impose
at retirement dates, and myriad other fac- constraints on the process, for example, by
tors. We omit these details. This approach limiting the stock-asset ratio, addressing
goes by the title—accumulated benefit ob- transactions costs whenever assets are
ligations (ABO). A second approach for bought or sold, and taking advantage of
determining a surplus is to assume that investment opportunities. Our goal is to
the company will continue to exist over find a feasible decision policy that maxi-
time, adding new employees while others mizes a temporal objective function. Since
resign, retire, and so forth. This method we are dealing with temporal uncertainty,
requires additional assumptions regarding the optimal solution, like all feasible solu-
the company’s future; the resulting sur- tions, encompasses a set of paths or trajec-
plus is determined by the projected benefit tories for the pension-plan surplus across
obligations (PBO) method. These calcula- each of the scenarios. To give an idea of
tions provide a more robust definition of the model’s structure, we list two basic
the company’s ability to meet its future equations for the flow of funds: Equation
goals. However, the ABO is appropriate (1) for the jth asset category:
for determining company contribution.
In either case, the estimated cash flows xj,tⳭ1 ⳱ (xj,t
s
⳯ rj,t
s
) ⳮ pj,t
s
all scenarios s1 and s2 that inherent a com- cases, risk increases as a function of proba-
mon past up to time t. In words, scenarios bility. An improved alternative for evalu-
that share a common path will have non- ating risks for an insurance company is to
anticipatory constraints for variables oc- estimate the entire probability distribution
curring on the shared path. Any financial- of shareholders equity at each time period,
planning system must address these along with other measures of financial
conditions, either explicitly or implicitly. well-being. In the case of a pension plan,
Special purpose algorithms are available we measure the discounted contribution
for solving the resulting convex or non- and the plan’s surplus at the end of the
convex optimization model [Ziemba and planning horizon.
Mulvey 1998]. Typically, we equate reward with ex-
Pension-Planning Objectives pected value. We have particular interest
A major element of asset-and-liability in the pension plan’s surplus at the
management is to trade off risks and re- horizon:
wards at various temporal junctures. It is
natural to expect that investments possess- Expected surplus ⳱ 兺 ps * zs,
s僆S
ing volatility will generate greater ex-
pected returns over time than assets with where ps is the probability of scenario s, zs
lower levels of volatility. The temporal is- is the pension surplus under scenario s,
sue complicates the decision process since time T, and S is a set of representative sce-
long-term horizons provide a cushion to narios. We are interested in risk at the ho-
recoup losses; thus volatile assets may be, rizon and include a number of alternative
in fact, safer in terms of contextual risks measures, including semi-variance, down-
than less volatile assets. An example is the side risks, and surplus variance. In addi-
stock-cash comparison: stocks provide tion, we can show the trajectory paths for
higher expected returns but are more vola- the pension plan under any of the gener-
tile than cash. We must consider the time ated scenarios.
horizon when measuring contextual risks, Policy Rules
that is, the chances of meeting the inves- As a distinguishing feature, the Towers
tor’s goals. An integrated ALM system Perrin system establishes a set of policy
provides an ideal method for understand- rules for managing investments, liability
ing the temporal risk-reward trade-offs payments, and pension contributions. We
and for evaluating the probability of meet- define a policy rule as a specified formula
ing company goals. or set of rules for all the decisions—as a
There are many metrics for financial function of the state of the world—at each
risks, just as there are alternative measures time period and scenario. As a simple ex-
of profitability or return. For an insurance ample, we might put a fixed proportion
company, we might consider the chance of (say 70 percent) of the pension surplus into
a loss over the next year. Or we might set an S&P500 index fund with the remaining
a profitability target and evaluate the assets assigned to a fixed-income bond
probability of missing the target. In both portfolio that mimics the pension liabili-
ties. Thus at each time period, the system same time, pension planners must satisfy
must calculate pension surplus based on the prudent-man rule. The company must
discounted cash-flow projections, deter- follow numerous regulations scrupu-
mine the excess of assets over market lously—government, tax, accounting, and
value of liabilities, and rebalance the in- so forth. Clearly, there are a variety of per-
vestment portfolio to achieve the desired spectives on the risk-reward trade-offs.
target ratios. We call this particular policy To address these issues, Towers Perrin
rule fixed proportional surplus (FPS). Nu- developed the Retirement Plan Financial
merous alternative policy rules are avail- Management approach. It combines an as-
able for testing. sessment of the plan liabilities and assets
The optimization module compares pol- to help determine benefit, funding, invest-
icy rules and determines the best combina- ment, and accounting policies that are
tion of rules and parameter settings. For consistent with a company’s overall goals.
example, in the FPS rule the parameter de- This methodology applies to all compo-
fining equity proportion will be found via nents of a company’s retirement pro-
optimization. Policy optimization provides gram—including pension, savings, and re-
a natural approach for analyzing long- tiree benefits. Some of the key financial
term investment problems. The rules can measures that are typically included in a
be as complex as the investor likes. Policy company’s statement of objectives are:
rules satisfy the nonanticipatory con- —Future cash-contribution requirements;
straints in an implicit sense. In addition, —Long-term portfolio return relative to
the approach simplifies the stochastic pro- interest-crediting rates (for cash-balance
gramming model since it reduces the plans);
number of decision variables over tree- —Possible reductions to shareholder
based stochastic programming methods equity;
and it can readily parallelize the optimiza- —Recovery of plan costs in rates (for regu-
tion problem. In addition, senior managers lated companies);
can readily grasp policy rules. As a disad- —Future cash-flow needs for benefit
vantage, the resulting model is likely to be payments;
nonconvex. —Funded status; and
Pension Planning Examples —Financial expense.
A defined-benefit pension plan for a A pension-plan design should address
large US company encompasses issues the four areas—benefit design, funding
that affect a range of stakeholders. The policy, investment strategy, and account-
company must set its benefit policy so that ing policy—that affect the company’s fi-
both retired and working employees are nancial health as measured by the state-
properly compensated and fairly treated. ment of objectives. These financial “levers”
The terms properly and fairly are subject link together and link to the company’s
to alternative interpretations. The invest- broader business objectives (Figure 3).
ment policy must consider the long-term Benefit design includes the types of plans
nature of the planning horizon. At the utilized and the level of benefits each plan
surplus;
—Develop funding, accounting, and in-
vestment policies to minimize the proba-
bility of future reductions in shareholder
equity;
—Determine the financial and cash-flow
impact of offering lump sums to pension-
plan participants at retirement;
—Shape the financial-expense patterns to
Figure 3: The four elements of pension-plan compress spikes and reduce year-to-year
design should together fit the company’s volatility; and
goals. Each element should assist in achieving
the company’s business objectives. —Analyze the probable impacts on the
financial statement of actuarial
provides. assumptions.
The funding policy covers the amount of The following three real-world case
capital the company contributes in ad- studies contain details that are similar to
vance to support promised benefits. many pension plans.
The investment strategy controls the in- Example A
vestment of the capital allocated by the Company A’s retirement plan has been
funding policy. fully funded for many years. As a result, it
The accounting policy recognizes the an- has not made cash contributions in recent
nual corporate expenses and balance-sheet memory. Because of several acquisitions,
reserves. the retirement plan will likely no longer
A fundamental objective is to minimize be fully funding in a few years. In fact, a
or at least manage cash contributions, fi- simple deterministic forecast projected
nancial expense, or otherwise-defined fi- that a cash contribution in excess of $60
nancial risk using these levers. Adopting a million would be required in the year
systematic process helps the company ful- 2000. While the company accepted the fact
fill its financial duties to shareholders and that it would have to contribute cash, it
its fiduciary duties under its regulatory wanted to minimize the probability that it
guidelines, such as the Employee Retire- would have to contribute more than $60
ment Income Stability Act (ERISA). Re- million in any single year of the next five.
cently, our clients have used the financial- Following our analysis, the company
management process to: adopted a new funding policy to shape its
—Shape funding requirements to reduce future cash contributions. The new policy
excessive funding requirements in any sin- reduced the probability that the cash con-
gle year; tribution would exceed $60 million in any
—Evaluate the need to cap the interest of the next five years from 58 percent to 18
crediting rate in a new cash balance plan; percent. The new funding policy included
—Determine the optimal asset allocation two key changes from the current policy:
aimed at maximizing return on the plan’s the asset-smoothing method was reset so
Figure 5: Adding a cap to the pension plan will have little impact on Company B’s financial
health. It will detract from the perceived value of the plan to the participants.
Figure 7: The surplus efficient frontier shows the optimal combination of risk and reward for
the pension-plan surplus. Company C’s current policies are inefficient with respect to its sur-
plus.
sonable range of outcomes, depending many cases, a set of 500 scenarios has
upon the investor’s needs. We minimize proven adequate for investors interested
the number of scenarios by employing in standard risk-reward analysis, such as
variance-reduction methods including surplus efficient frontiers. Additional sce-
antithetic variables. In addition, we can narios can be generated as needed when
evaluate the model’s recommendations by there is particular interest in rare events.
running these through a process that we For instance, the Tillinghast property casu-
call stratified filtered search [Ziemba and alty system requires a large number of sce-
Mulvey 1998]. Here, the goal is to generate narios (over 10,000) due to its emphasis on
a series of scenarios that minimizes the rare catastrophes arising from large hurri-
sampling errors and maximizes confidence canes and earthquakes.
in the resulting efficient frontier and asset To validate the model, we spend a great
allocation. deal of time evaluating scenarios for rea-
A second issue is the errors that arise in sonableness. We often look at individual
sampling scenarios within the optimiza- scenarios to understand the connection be-
tion module. We address this issue by ad- tween variables, for example, the level of
justing the number of scenarios sampled interest rates and the accompanying level
depending upon the investor’s needs. In of inflation (Figure 8). Although interest
Figure 9: The outcomes of three strategies over the standard 500 scenarios and three stressful
scenarios sets. The current policy has the best risk-reward relationship.
ing Travelers and Citicorp (with Smith tion with Tillinghast, has developed an in-
Barney and Salomon Brothers as divi- tegrated risk-management system [Lowe
sions). European companies are well along and Stanard 1996].
the same path. This trend will accelerate Financial companies have traditionally
as Congress replaces depression-era regu- been managed as decentralized businesses,
lations, such as the 1933 Glass-Steagall with occasional meetings between divi-
Act. These conglomerate financial organi- sional executives to set policy and targets.
zations need to integrate their decisions. The situation is reminiscent of manufac-
Companies must understand the risk- turers and retailers in the 1960s and 1970s
adjusted returns for each operation. A fi- with their production, inventory, and
nancial company must also be able to transportation subsystems. Integrated lo-
identify risk factors that span the organi- gistics is the norm for these companies to-
zation, giving rise to correlated risks. The day. A similar change will take place in
Towers Perrin ALM system can be ex- the management of financial organiza-
tended to cover business-related activities. tions. The Towers Perrin system provides
As a consequence, the property casualty a prominent example of the advantages of
(PC) division of Tillinghast—Towers Per- integrating asset and liability decisions
rin has developed an integrated risk- within a common framework.
management system for the PC industry. APPENDIX
Also, Renaissance Reinsurance, in conjunc- The scenarios generated by CAP:Link
contain key economic variables, such as cade format. Modules above and equal to
price and wage inflation, interest rates for that level can affect each submodule
different maturities (real and nominal), within the system. Briefly, the first level
stock dividend yields and growth rates, consists of short and long interest rates,
and exchange rates through each year for and price inflation. Interest rates are a key
a period of up to 40 years. We model re- attribute in modeling asset returns and es-
turns on asset classes and liability projec- pecially in coordinating the linkages be-
tions consistent with the underlying eco- tween asset returns and liability invest-
nomic factors, especially interest rates and ments. To calculate a pension plan or an
inflation. The model simultaneously deter- insurance company’s surplus, we must be
mines economic variables for multiple able to discount the projected liability cash
economies within a common global frame- flows at a discount rate that is consistent
work. Long-term asset-and-liability man- with bond returns, under each scenario.
agement is the primary application. Also, since dynamic relationships are es-
The global CAP:Link system forms a sential in risk analysis, the interest-rate
linked network of single-country modules. model forms a critical element.
The three major economic powers—the The second level entails real yields,
United States, Germany (now the EU), and currency-exchange rates, and wages. At
Japan—occupy a central role, with the re- the third level, we focus on the compo-
maining countries designated as home or nents of equity returns: dividend yields
other countries. We assume that the other and dividend growth. Returns for the re-
countries are affected by but do not affect maining asset classes form the next level,
the economies of the three major coun- with fixed-income assets reflecting the
tries. The basic stochastic differential equa- term structure of interest rates and other
tions are identical in each country, al- mechanisms. We project each economic
though the parameters reflect unique variable by means of a stochastic differen-
characteristics of each particular economy. tial equation, relating the variable through
We can readily include additional coun- time and with the stochastic elements of
tries in the framework by increasing the the equation and to other variables and
number of other countries. factors at the same or higher levels in the
Within each country, the basic economic cascade.
structure is such that variables at the top A critical feature for a global scenario
of the structure influence those below, but generator is the currency module. Several
not vice-versa (Figure 10). This approach issues complicate modeling currency-
eases the task of calibrating the model’s exchange rates. First, currencies must en-
parameters. The ordering does not reflect force the arbitrage free condition among
causality between economic variables but spot exchange rates and among forward
rather captures significant comovements. rates with differential interest rates. The
Linkages across countries occur at various second issue concerns symmetry and nu-
levels of the model—for example, interest meraire independence. We must create a
rates and stock returns. These connections structure in which the distribution of cur-
are discussed by Mulvey and Thorlacius rency returns from country A to country B
[1998]. Roughly, the economic conditions has the same distribution as returns from
in a single country are more or less af- B to A. Both issues limit the form of the
fected by those of its neighboring coun- currency-exchange models, especially
tries and by its trading partners. The de- when integrating three or more currencies.
gree of interaction depends upon the To avoid these problems, we focus on the
country. The structure is based on a cas- strength of each country’s currency. Ex-
Figure 10: The stochastic differential equations are arranged in a cascade structure. Factors at
the higher levels of the cascade affect those below, but not vice versa. Each country has the
identical structure with its unique parameters.
Report, No. 97-02, Princeton, New Jersey. ferent reinsurance treaties that the insurer
Mulvey, J. M.; Morin, F.; and Pauling, B. 1999,
“Calibration of stochastic scenario generators
might buy to protect itself from adverse
for DFA,” Casualty Actuarial Society Forum, claim experience.
summer, and Annals of Operations Research, 3. We have analyzed the impact of alter-
Vol. 85, pp. 249–266. native debt/equity ratios on risk and re-
Mulvey, J. M. and Thorlacius, E. 1998. “The
Towers Perrin global capital market scenario turn for an insurer interested in the effi-
generation system,” in Worldwide Asset and ciency of its capital structure.
Liability Modeling, eds. W. T. Ziemba and “We expect the use of tools such as
J. M. Mulvey, Cambridge University Press,
Global CAP: Link to explode in the next
Cambridge, UK.
Ziemba, W. T. and Mulvey, J. M., eds. 1998, few years, and to fundamentally alter the
Worldwide Asset and Liability Modeling, Cam- way insurers manage risk and capital. Our
bridge University Press, Cambridge, UK. clients are excited by the results we have
obtained; most importantly, they are act-
Stephen P. Lowe, FCAS, MAAA Princi- ing on the recommendations stemming
pal, Chief Actuary Tillinghast-Towers Per- from them.”
rin, Forestal Centre, 175 Powder Forest J. Stanford Willie, PhD, Vice President,
Drive, Weatogue, Connecticut 06069-9658, US WEST Investment, Management Com-
writes: “Insurance companies are increas- pany, 7800 East Orchard Road, Suite 290,
ingly recognizing the need to build dy- Englewood, Colorado 80111, writes: “One
namic financial models to support the of the most unique features of the Towers
management of risk and capital. A corner- Perrin System is the linkages between the
stone of this process is the generation of asset and liability components of the mod-
the economic scenarios. We are using els underlying the system. Many systems
Global CAP:Link for this purpose, with available in the marketplace focus only on
great success. Global CAP:Link provides the asset side of the problem and simply
scenarios for the key economic variables determine a mix of assets that attempts to
that drive the behavior of the assets and li- maximize expected return and/or mini-
abilities. Equally important is that the sce- mize volatility of return. They ignore the
narios extend across multiple periods, importance of the liability performance in
which is critical when modeling long- the evaluation of total risk. Even in the
term, non-tradable insurance contracts. cases where these systems take the liabili-
“We are using Global CAP:Link, in con- ties into account, the forecasted perfor-
junction with our TAS financial modeling mance is often inconsistent with the simu-
software to assist insurers in looking at a lated asset performance and is, we believe,
number of critical issues. unreliable, especially over multi-periods.
1. We have analyzed alternative invest- With the Towers Perrin System, there are
ment strategies, given an insurer’s liabili- direct linkages between the various model
ties. These projects entail the construction components underlying the system. These
of asset/liability efficient frontiers. linkages give us confidence that the out-
2. We have analyzed alternative portfo- comes are reasonable and truly reflect the
lio hedging strategies, in the form of dif- risk/reward trade-offs we are making in