Beruflich Dokumente
Kultur Dokumente
Contents
1
Introduction............................................................................................................. 1
Background ............................................................................................................. 3
Product Features...................................................................................................... 7
3.1
3.2
3.3
3.4
3.5
4.1
4.2
4.3
4.4
4.5
4.6
4.7
Managing Risks..................................................................................................... 16
5.1
5.2
5.3
5.4
5.5
5.6
5.7
Introduction........................................................................................................... 16
Variable Annuity Risk Factors.............................................................................. 16
Market Risk........................................................................................................... 16
Mortality/Longevity .............................................................................................. 17
Policyholder behaviour ......................................................................................... 17
Risk Mitigation ..................................................................................................... 17
Risk Management ................................................................................................. 18
6.1
6.2
Using Risk Geographies to find the Least Solvent Likely Event (LSLE) ............ 31
10
Acknowledgements ............................................................................................... 38
11
Bibliography.......................................................................................................... 39
Introduction
This paper is designed to give an introduction to variable annuities, an overview of their
origins and some insights into the techniques used to manage them.
In this paper we set out the key features of a variable annuity product and aim to give the
reader understanding of the risks involved. We also discuss the design and pricing of
variable annuities. Variable annuities (VA) have sold extensively in the United States
and parts of Asia. Now many multi-national companies have are launching variable
annuity programs across Europe. A reader who is new to variable annuities will be
introduced to the key product features.
As a starting point, what are variable annuities? Well, they arent really variable and are
not necessarily annuities.. They are actually unit linked savings contracts with
attaching guarantees.
These contracts are not completely new, they have been seen before in various guises.
There have been
These all have similarities with the latest incarnation of unit linked funds with guarantees.
What has hopefully changed, and we will be demonstrating this in the paper, is the
understanding of the value of the guarantees and the monitoring and management of
them.
The benefits can be referred to collectively as GMxBs and the following are the most
common benefits attaching to contracts.
The following table shows the various life stages of a consumer together with the
appropriate GMxB. The final column shows a possible version of the benefit that has
been provided in the past in the UK.
Life Stage
Protection of
Preretirement
Principal
Retirement
Return of Principal
Retirement Income
On Death
Estate
Type of
Guarantee
GMAB
UK equivalent example
GMWB
GMIB
Lifetime GMWB
GMDB
Drawdown
GAO
Maximum of return of
premium or fund value
We also discuss hedging as this is a key element in a variable annuity programme. The
guarantees in the product are largely investment guarantees therefore suitable assets for
hedging are required. Commonly used derivatives are equity options and interest rate
swaps. Other, more complex options are available, over-the-counter (OTC) on an
individual basis. The hedging program involves the calculation of the greeks, measures
of how the fund moves with the change in key economic drivers such as the price of the
underlying assets for equities, and interest rates for bonds. A general discussion of the
greeks is given as well as some practical advice for the outworking of such a programme.
This leads on to a discussion of the cost of hedging and cost of capital that may result if
such a programme was introduced in practice. A formulaic approach has been developed
which avoids the problem of calculating projected greeks and capital using monte-carlo
simulations or sims within sims, for calculating the cost of hedging to use in the pricing
of variable annuities. The formula is based on ideas developed by Davis, Panas and
Zariphopolu at Imperial College in the 1980s. It considers the portfolio hedge position as
a markov process that has a stationary distribution. This process can be used to find key
hedging variables such as the cost of the hedge per annum and the cost of capital after the
hedge is in place. Results derived from this method give reasonable results and are
demonstrated in graphical form. Once calculated, these variables serve as inputs to the
pricing process.
Background
2.1
US Annuity sales
400.0
$250,000,000,000
350.0
Sales
$200,000,000,000
300.0
250.0
$150,000,000,000
200.0
$100,000,000,000
150.0
100.0
$50,000,000,000
Inflation Factor
50.0
$0
0.0
1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Fixed
Variable
Inflation Factor
Source: 1 The 2005 Individual Annuity Market, LIMRA International, Consumer Price Index for All Urban
Consumers, US Department of Labor, Bureau of Labor Statistics.
2 Deloitte Debrief.
Note the Inflation Factor is the CPI rescaled such that calendar year 1975 has a value of 100.
2.2
Following the various challenges in the life and pensions market place, the insurance
industry is looking to make pensions products attractive again.
Given the general decline in the availability of final salary pension schemes over the
last 15 to 20 years, introducing products with manageable, client targeted guarantees
should fill a space left by With Profits and Final Salary not currently being provided
by unit fund products or by buy to let properties
If well designed and attractive to consumers, they may even end up by being more
bought than sold, such is the demand for good quality retirement savings schemes
They should be transparent in benefits and charges to the consumers and advisors
This could be the opportunity for the insurance industry to return to providing savings
products with investment and/or longevity protection. Given that most consumers, when
asked, reply that they would like equity style growth with no risk of loss, this should be
an excellent opportunity for the insurance industry.
2.3
Guaranteed annuity rates can look poor in a stable low interest rate environment, but
is that about to change given the recent credit crunch?
IFAs appear to have concerns about the value for money of the charges for the
guarantees as well as general concerns about a new product; gaining their
understanding and buy in for the product features and charges. They are
understandably concerned about potential misselling.
Given that the products have been available in the US since the 1990s, we want to ensure
that we learn from the challenges they encountered.
Uncovered guarantees: initially some companies were only hedging against small
changes in price (delta hedging), clearly this meant that large changes in price and
changes in volatility were not covered and were potentially far more onerous.
Economic capital costs: these were often not calculated as were not required by
regulators and so management did not have an overall understanding of the type and
size of risks being written in the product.
Policyholder behaviour: it was not always recognised, when the products were first
designed, that policyholder behaviour could be dramatically affected by the external
environment. This is particularly important to consider when pricing guarantees
where the policyholder has an option against the insurance company. Examples are
when policyholders lapse to take advantage of a high guaranteed surrender value or
when they do not lapse when their guarantees have value (in-the-money).
There was a lack of technical skill at the management level in the complexities within
these new products. As in all markets, once the product started to be successful,
additional companies also launched products without necessarily having the risk
processes and controls in place.
2.4
In the US, the main market was consumers that previously invested in deposit
administration style accounts and therefore had little opportunity for growth but had
very good downside protection. The possibility of equity exposure, even at a cost,
would have been something that was not available before
In Japan, premiums amounting to Billions of GBP have been written in recent years.
The product design can be such that there is an initial deduction from the single
premium of 4-5%, followed by a yearly deduction from the fund to cover both
guarantee charges and investment management fees of almost 3%. In a market where
the 10 year risk free rate is around 1.5%, the value placed by consumers on downside
protection with the possibility of equity growth can be clearly seen.
Variable annuities have started to appear in Europe. Axa has launched a regular premium
savings product in Germany with a GMIB.
In the UK, Aegon, MetLife, Hartford, AIG and Lincoln have launched VA products.
These companies all have access to US technical expertise to assist in the product design
and risk monitoring, including hedging, processes that are required. Many companies are
considering launching a VA product, including, for example, Standard Life, who have
publicly stated that they will be launching a VA product shortly.
In the US, the products have all been single premium but in Continental Europe, regular
premium savings versions are also available.
The charges for guarantees are usually quoted as a percentage of the unit fund as this then
stays fixed as a proportion of the value of the fund. This is more transparent to both the
consumer and the broker than a percentage of premium as well as being the standard
method of charging for other unit funds. It does, however, suffer from the mismatch
issue that the guarantees become more in-the-money when the funds drop in value which
is the time that the proportionate deduction becomes smaller.
Several groups selling variable annuities have set up offshore entities to aggregate
variable annuity guarantees eg MetLife in Bermuda, Tokio Marine in Isle of Man. This
can have several benefits
2.5
Media Views
The UK market is very open in its disclosure around the selling process following
tightening of regulations after a number of misselling scandals. There are a number of
different views on whether VA products will be the panacea that is hoped for. Source of
quotes: Ellen Kelleher, FT.com 15 June 2007.
Without decent underlying fund performance your income will likely fall in real
terms because of inflation, say advisers.
The items that will impact on the level of the underlying fund in addition to the
growth rate are charges for guarantees and investment management and commission.
Some advisers argue that paying additional money for a guaranteed income is not to
the investors advantage. They also argue that if the investor bought a conventional
annuity at outset they would get a much higher starting income. However supporters
of variable annuities say that investors still get an income guarantee and that if they
get half-decent fund performance their income will rise.
These mixed messages seem to be confusing both IFAs and consumers at present.
We will now move on to look at the features of variable annuity products.
Product Features
3.1
3.2
include a death benefit in the early stages of the annuity. The death benefit is generally a
return of the remaining fund balance.
The annuity benefit (either the guaranteed amount or higher) is paid until the death for the
annuitant.
3.3
10000
8000
FUND
ROP
6000
4000
2000
0
1
10
11
12
13
14
15
16
17
18
19
20
12000
10000
8000
FUND
Roll-up
6000
4000
2000
0
1
10
11
12
13
14
15
16
17
18
19
20
12000
11500
11000
10500
FUND
Ratchet
10000
9500
9000
8500
8000
date
10
11
12
13
14
15
16
17
18
19
20
(The fund guaranteed value may be adjusted downwards if there are partial withdrawals,
but this is not shown on the graph above).
3.4
3.5
The level of the payout (the annual payout as a percentage of the guaranteed amount)
is uncertain
The payout may depend on the fund value at the time that the option is taken up.
GMWB
120000
100000
80000
Guarantee fund
withdrawal
Fund
60000
40000
20000
0
1
10
11
12
13
14
15
16
17
18
19
20
21
Year
The graph above illustrates the example where the account value drops to zero, but the
withdrawal is still paid.
10
4.1
4.2
The product must address the needs of the consumers that are being targeted.
Investment in research into the needs of consumers as well as the perceived value of
the benefits offered is critical. As UK consumers often state that they would like
equity growth with no risk, this would seem to make these contracts an ideal solution.
The question remaining is the price that consumers are willing to pay and that brokers
are willing to recommend for these guarantees.
The distribution channels that are available and most suitable for the target market.
Another question that should be asked is, given the distribution channels available,
what markets can be targeted?
Simple, flexible models need to be available to perform product design and pricing,
reserving, capital requirements and hedging strategy if required.
Management needs to understand the products and the risks inherent in the products.
They will also need to be clear that the risks fit in with the overall risk appetite of the
company.
The tax regime under which the product will be offered. Some products in the UK
have been written as single premium investment bonds to give full flexibility to the
features that can be included. For example, Life time GMWB would not be possible
after age 75 on a UK pension contract. Of course, flexibility needs to be weighed up
against the tax benefits of a pensions contract.
The legal, administration, actuarial and systems professionals will need to work
together to put the controls and processes in place.
Modelling requirements
A major aspect of the product design and pricing is accurately calculating the cost of the
guarantees. Rather than a deterministic profit test, a stochastic profit test should be
considered in order to capture the impact of the guarantees and options on the present
value of future profit. Ideally the model should capture policyholder decisions such as the
timing of when the guarantee is taken up where the policyholder has an option against the
company. The profitability of each surrender option may also be assessed by comparing
the deflated value of the charges for that guarantee with the deflated value of the outgo
for the guarantee. Different charges for different years to maturity could be considered, as
well as how many charging bands to have, and whether or not to offer different charges
for different funds if there is different volatility within the fund.
11
A calculation of capital requirements is also important for pricing variable annuities. This
calculation may not be straight forward and may be time consuming. In particular,
building the model to do the calculation is likely to be complex.
4.3
4.4
12
4.5
International Perspective
In the US, variable annuity sales have consistently out-sold fixed and index linked
annuities. During 2006, variable annuity sales increased by 16% while fixed annuity sales
decreased.
GMWBs are dominating the sales of variable annuities in the US.
This guarantee was first introduced in the US by Hartford Life in 2002. There have been
variations of the GMWB which has meant that focus has been on this particular
guarantee. These variations include:
Most are offering a bonus feature now, as long as withdrawals are deferred
13
Combination GMAB/GMWB
With the growing popularity of GMWBs, GMIBs have become less prevalent.
4.6
14
h) Reason not to offer: 45% are either concerned about their lack of knowledge or the
fact that the products are untested in the UK. 12% were concerned about product
complexity and 10% about mis-selling or TCF issues.
4.7
Distribution
Sales of Variable annuities in the US are less dominated by independent agents, than
sales of fixed annuities. Graphs shown the split of sales by distribution channel are shown
below:
2005 Annuity Sales
by Channel
Industry sales by channel for VA
FY 2005
Other
Direct Systems Independent
4%
Agents
7%
21%
Stock
Brokers
13%
Other
Systems
11%
Financial
Planners
5%
Career
Agents
13%
Career
Agents
20%
Financial
Planners
16%
Banks
19%
Independent
Agents
50%
Banks
21%
Financial planners, stock brokers, and direct sales are much more important for variable
annuity sales than fixed annuity sales in the United States.
15
Managing Risks
5.1
Introduction
This section discusses a method for understanding and managing the risks inherent in a
variable annuity contract. We have based this on the concept of economic capital as used
in both the UK regulatory regime and as proposed in Solvency II. We show below the
standard Solvency II risk classification matrix.
Risk Classification
Total Risk
Insurance Risk
Market Risk
Equity
Longevity
Interest
Rate
Mortality
Equity
Volatility
Interest rate
Volatility
Expenses
Liquidity
Other
Operational
Persistency
Take-up
Rates
Basis
5.2
5.3
Market Risk
The market risks should be considered separately. These will include
Equity returns
Interest rates
16
These risks are systematic and so can be hedged. The stresses for the equity and interest
rates can be derived from examining historic data or by using market agreed calibrations
such as those for Solvency II.
5.4
Mortality/Longevity
Analysis of historic mortality trends is not necessarily a useful guide to setting a stress
level for the future. The latest projections from industry bodies are usually a good source
of information. If the mortality risk with the VA products is higher than is acceptable
under the risk tolerance guidelines of the company then the risk can be transferred out via
the reinsurance market.
5.5
Policyholder behaviour
Due to the guarantees attaching to VA products, policyholders have the opportunity to
select against the insurance company. One of the most complex assumptions to set, both
for the base pricing assumptions and for the stresses to be used for the capital
requirements is that of policyholder behaviour.
Depending on the financial sophistication of the consumers and their advisors,
persistency and or take up rates can be correlated to market performance. This risk is
unhedgeable. There is also very little evidence to back up any assumption made.
5.6
Risk Mitigation
There are a number of ways of mitigating the risks in VA products. Some are discussed
below. Avoidance: Product Features. These are used to encourage behaviour from the
consumers that is in the interest of the office as well as providing features that the
consumers will value.
Limits on funds and switching (particularly if a level fee is levied across all funds)
Ratchets on benefits to discourage lapses when guarantees are too far out of the
money
Transfer: Reinsurance. This takes the risk off the balance sheet of the insurer but at a
price. The credit risk of the reinsurer should be considered.
This was used initially in US but then market became saturated. New entrants have
started appearing including European entrants.
Reduction: Hedging
17
Basis Risk (tracking error). This is the difference between the performance of the
hedge assets and that of the underlying funds.
Retention
The amount of risk retained will depend upon the risk capacity and the risk appetite of the
insurer.
5.7
Risk Management
Most discussions relating to the management of variable annuity risks concentrate on
managing the market risks via a hedging programme. There are, however, other risks that
need to be managed.
Lapse and mortality risk hedging programmes (usually) assume static lapse and
mortality profiles, the actual experience will differ.
Basis risk the hedge assets may not exactly match the underlying investment fund
assets. Even the replication of a liquid index fund often involves only holding 5060% of the index assets by number.
Non linearity issues for example, it is possible that fewer policyholders lapse when
markets are low and more volatile. In this situation, the portfolio would be under
hedged at the same time as the hedges cost more to put in place.
18
Hedging Techniques
6.1
19
0
0
0
0
0.9638
0.2148
0.1603
-0.524
0.7237
0.0954
0.6079
0.4756
0.3935
0.1317
-0.183
-0.566
0.3728
-0.033
-0.002
0.6663
-0.222
vega
-0.286
-0.378
-0.807
-0.312
0.1123
0.6515
0.798
0.9439
-0.486
0.1304
0.9812
-0.801
10
-0.902
-0.111
0.6851
0.5249
20
0
0
0
0
lambda
1
0
0
0
0
0.6529
-0.438
0.657
0.5402
-0.375
-0.853
0
0
0
0
-0.494
0.485
-0.536
0
0
0
0
0
0
0
0
-0.94
-0.63
-0.93
-0.89
0.793
0.911
0
0
0
0
-0.61
0.439
-0.95
0
0
0
0
0
0
0
0
-0.51
0.911
-0.75
-0.79
0.89
-0.81
0
0
0
0
-0.89
0.308
-0.93
0
0
0
0
10
0
0
0
0
-0.05
-0.31
-0.99
-0.73
-0.35
0.611
0
0
0
0
0.159
-0.25
0.468
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
tracking error
Combined LSLE stress
#101 #102 #103 #104 #0
#1
#2
#3
policyholder behaviour
20 #1
#2
#3
#4
0.98
0.021 0.8129 0.9855 0.5859 0.6447 0.6592 0.9473 0.6221 0.5444 0.0686 0.271 0.8183 0.9078 0.3587 0.079 0.453 0.674 0.277 0.754 0.357 0.49 0.055 0.466 0.098 0.328 0.573 0.353 0.678 0.048
-0.002 -0.144 -0.068 -0.228 -0.058 -0.061 -0.177 -0.069 -0.232 -0.003 -0.127 -0.068 -0.452 -0.16 -0.03 -0.14 -0.13 -0.01 -0.11 -0.04 -0.14 -0.03 -0.07
-0 -0.07 -0.09 -0.16 -0.33 -0.02
0.0188 0.6692 0.9172 0.3574 0.5864 0.5983 0.7702 0.5528 0.3125 0.0651 0.144 0.7507 0.4557 0.1987 0.051 0.316 0.543 0.267 0.648 0.317 0.354 0.029 0.398 0.094 0.254 0.482 0.194 0.35 0.028
0.99 1.055 0.896 0.456 0.709 0.678 0.116 0.174 0.941 0.901 0.897 0.538
ECAP
Div
With div
0.14
1.1677 1.0773 1.1035 1.0895 0.2736 0.3929 1.075 1.0786 0.8179 0.009 0.4625 0.6033 0.9777 0.6475 1.036
Position / Limit
4.2793 0.6872 0.0347 0.1374 6.3634 5.1416 1.3964 0.3011 2.7517 340.36 0.5317 1.9775 1.7005 0.2298 0.838 9.844 0.508 0.292 1.353 2.387 0.878 0.184 3.252
-2.02 1.5012 -0.325 2.2507 -3.048 0.2459 -1.193 -1.662 -0.149 0.868 1.382 0.503 -0.31 1.213 -1.09 -0.62 0.125 0.379 2.616 -0.99 -1.41 -0.21 -0.55 1.067
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
-0.463
0
0
0
0
0 0.5012
0.1043
0
0
0
0
0 -0.721
-0.588
0
0
0
0
0 0.807
0.6222
0
0
0
0
0 0.4814
-0.116
0
0
0
0
0 -0.167
-0.326
0
0
0
0
0 0.333
-0.986
0
0
0
0
0
0
0.313
0
0
0
0
0
0
0.7902
0
0
0
0
0
0
-0.697
0
0
0
0
0
0
-0.374 0.7428 0.1305 0.3692 0.7154 -0.452 0.1294
0.3606 -0.914 -0.597 -0.277 -0.298 0.5697 0.7886
-0.87 0.4465 -0.535 0.9419 0.5541 0.8391 0.5316
0.5227
0.0357
-0.594
0.8397
20
Position limit
Aggregate position
total
4.9968 -0.74 0.0383 0.1497 -1.741
-0.965
0
0
0
0.1827
0
0
0
0.6377
0
0
0
-0.716
0
0
0
-0.978
0 -0.511 -0.251
-0.021
0 -0.688 -0.286
-0.949
0 0.8907 0.9622
-0.097
0 -0.106 -0.818
-0.782
0 -0.929 -0.764
-0.862
0 0.6207 -0.032
0
0 -0.929 -0.337
0
0 -0.31 -0.593
0
0 -0.728 0.7451
0
0 0.4496 -0.865
0.6028 -0.045 -0.089 -0.205
0.8581 -0.621 -0.061 0.4784
0.3117 -0.355 0.5429 -0.712
0
0
0
0
-0.494
-0.591
0.7545
-0.957
0.9247
0.9937
0.33
0.9819
-0.727
-0.894
-0.393
-0.451
0.3839
Hedge portfolio
fund #101 -0.965
fund #102 0.183
fund #103 0.638
fund #104 -0.716
index future -0.978
index future -0.021
index future -0.949
index future -0.097
index future -0.782
index future -0.862
swap
-0.287
swap
0.754
swap
-0.212
swap
-0.205
put option 0.826
put option 0.162
put option -0.852
0.9448
-0.374
-0.861
0.8611
10
conts.
term
20
The effect of changes in the risk drivers on the various greeks can be calculated and
graphed:
For example changes in equity values have the following effect
-0.8
-25%
-20%
-15%
-10%
-5%
0%
5%
10%
15%
20%
25%
-0.9
-1
-1.1
-1.2
GMAB
GMIB
GMDB
GMWB
-1.3
-1.4
-1.5
-1.6
-1.7
-1.8
-5%
-3%
-0.8
-1%
1%
3%
5%
7%
9%
11%
13%
15%
-0.9
-1
-1.1
-1.2
GMAB
GMIB
GMDB
GMWB
-1.3
-1.4
-1.5
-1.6
21
As well as considering the greeks, the cost of rebalancing the portfolio must also be
considered. For example the cost of trading in the derivatives necessary to hedge the
portfolio may be more expensive than the benefit of the hedging. In practice a trader will
take a view on how the volatilities in the market will move and therefore whether a given
derivative is over or under priced, the cost of changing the hedge and the advantage of the
hedge. It is common to have tolerance levels around the greeks so that traders have a
range that they need to hedge to. This gives them freedom to reduce the amount if trades
that they do and limit dealing costs.
A company may hedge a portfolio using much shorter instruments than the duration of
its liabilities. In this case, it will be exposed to some unhedged financial risks.
There are some limitations with delta hedging. The formula for delta is based on a linear
relationship between the price of the option and the price of the underlying asset.
However, the relationship is non-linear. Therefore the hedge doesnt hold for large
movements in the price of the underlying asset. The hedge could also move out if it was
left un-reset for long periods of time. Similar issues occur for rho hedging, however there
is the additional complication that exposure to the risk free rate generally comes from
assets whose prices are non-linear with respect to rho, for example bonds and swaps. The
swap market is generally the most liquid market for rho hedging, but using swaps also
introduces basis risk. Therefore other greeks such as gamma and vega are often used as
well. Additionally an expert trader may make decisions based on his/her views of the
market.
Greeks relating to other variables such as policyholder behaviour may also be calculated.
For example, the change in price with respect to a change in lapse rates. However, these
are more for information and are not usually hedged, except for maybe the cases stated
below.
In addition to hedging financial risk, there has also been some interest in hedging
longevity risk. Many life insurance companies choose to hedge this risk internally by
using their protection portfolio to offset risk in their annuity portfolio, or through
reinsurance. However, in the UK there are investment banks looking to offer mortality
bonds similar to US style products. This is a mechanism for owners of larger annuity
portfolios to effectively sell off their mortality risk. The holders of these bonds make
gains or losses depending on how the portfolio performs versus a national mortality
index. Such bonds have not been sold widely but there are investment banks willing to
create a market in them if a company wanted to crystalise its mortality risk, and a willing
buyer could be found.
6.2
Policyholder options
If the policyholder has an option which he/she could exercise against the company at an
uncertain point in the future, the company may choose to hedge this with American style
options. American options may be exercised at regular intervals up until maturity. Other
product features such as knock-out options (where the contract is surrendered if the total
accumulated investment return passes a specified level) may lead to the use of
Bermudan options, (half way between the US and UK)!
An American option is more difficult to price than an European option because it has
many possible exercise dates. The American option will be worth at least as much as the
22
European option and maybe more, depending on how likely early exercise is. An
American option pricing model may be calibrated using the prices of European options,
with the same strike price, etc. The price of European options may be calculated using the
standard Black Scholes formula.
Options on surrender have been studied and various mathematical models suggested.
These models are appropriate if the surrender value is guaranteed, (and could possibly be
extended for a guaranteed death benefit).
For example, Grosen and Jorgensen (1997)
Grosen and Jorgensen analyzed the early exercise options embedded in unit-linked
insurance contracts in their 1997 paper and extended this to with profit contracts in their
2000 paper.
Albizzali and Geman (1994)
This model is a deterministic lapse function where the decision of whether or not to lapse
depends on a stochastic interest rate.
Consiglio and De Giovanni (2007)
A super-replicating model which uses a stochastic model to replicate the payout
process. This models the payoffs of the hedge at any point in time.
The value of a claim can be expressed as:
C(t) = max{ Lt, It}
= max { L0erGt It, 0} + It
Where :
C(t) = the claim value at time t
L(t) = The initial premium accumulated at a guaranteed rate
L(0) = The initial premium
I(t) = The fund value at time t
This can then be evaluated stochastically at each future time period. Therefore a company
may be able to understand and model the optionality in its portfolio better and purchase
derivatives to hedge some of the policyholder behaviour risk that may be present in its
portfolio.
23
As In-the-moneyness
increases, surrender
rates fall, as the value of
the guarantee becomes
significant to more
policyholders
M 2
ul
ti
pli 1
n
er
de
r
R
0
at
0.
e
4
0.
7
1.
0
-
1.
3
1.
6
1.
9
2.
2
Does the evidence back up the correlation with financial markets assumed above?
Interestingly, a study performed by a European group showed little correlation of lapses
with the value of guarantees or market movements. The biggest correlation was with
market sentiment about the company itself. It could be argued that the recent run-onthe-bank that was seen with Northern Rock was an example of this.
The fully market correlated policyholder behaviour could be considered as one where all
policies are owned by a financially sophisticated purchaser such as a hedge fund. In this
case, calculations could be performed each day to see if the policy was in the money and
worth surrendering. This could be seen as the worst case scenario for policyholder
behaviour.
In practice, many aspects of policyholder behaviour may be determined by external
events. A policyholder may stop paying premiums if she becomes unemployed. Policies
may be surrendered to finance school fees, a round-the-world cruise or a family wedding.
Policy surrenders may be driven more by the commission churning behaviour of
intermediaries than by the policyholders themselves. Dice-rolling models of persistency,
based on historic lapse experience, can be a good way to capture all these kind of
behaviour. This used to form the basis of deterministic profit testing for variable
annuities.
There may, however, be some policyholders who ruthlessly exercise their options to
maximise the value of their policy. We call this rational behaviour. Rationality may be
close to maximising guarantee costs to the insurer, which is why insurers should worry
about it. It is important to consider the effect of rationality when profit testing even if
24
only as a sensitivity to the dice-rolling model. Where local law permits it, there may be a
risk that third parties, such as hedge funds, acquire policies in bulk and start exercising
large numbers of options rationally.
Rational behaviour is not easy to define. A perfect hindsight definition would mean
policyholders lapse when guarantees mature out of the money, and persist when
guarantees are in the money by more than the guarantee charges. But of course, nobody,
not even rational policyholders, knows at the date of lapse whether an option will later
come into the money or not. A workable definition of rationality must take account of
relevant information without allowing policyholders a crystal ball unavailable to the rest
of the market.
Consider first guaranteed minimum accumulation benefit (GMAB). Suppose this is paid
for by regular proportional charges to the fund. If the fund has performed badly, a rational
policyholder persists, because the charges are a proportion of a reduced fund, while the
guarantees are valuable. On the other hand, if the fund performs very well, the value of
the guarantee charges has increased at the same time as the guarantee has moved out of
the money. We can imagine a strategy of persisting when the fund value is below some
decision point level and lapsing the first time the fund crosses above the decision point.
How might rational policyholders choose the decision point? If the decision point is set
very low, only slightly above the current fund level, then a decision point crossing is
likely. Effectively, that means a high lapse rate both for charges and for guarantees. The
charges are not worth much, but then neither are the guarantees.
If the decision point is set very high, then it is unlikely ever to be touched. It is likely that
the policy is held to maturity, that the charges are paid in full and that all policyholders
benefit from the guarantee provided it matures in the money. With this strategy, a
traditional profit test with a low lapse assumption, accurately captures the products
profitability.
Now suppose the decision point is set at an intermediate level, with a significant chance
that the fund level crosses the decision point. To understand such a strategy, we need to
look separately at the value of the charges and of the guarantees. To value the charges, we
need to allow for the probability that the decision point is crossed. However, if we turned
the probability of hitting the decision point into an annual lapse rate, this would overstate
the value of the charges. The error happens because the high lapses happen in the high
market scenarios where the charges are most valuable. The positive correlation between
charges and lapse rates, which acts to reduce the value of the charges, is overlooked by a
deterministic profit test. Equivalently, to replicate the true value of charges with a
deterministic profit test requires us to increase the lapse assumption by a margin to reflect
the correlation with the charge amounts.
When it comes to the guarantees, the opposite situation applies. A deterministic lapse
assumption overstates the value of the guarantees. That is because the high lapse
scenarios are those where the market has performed well, where the guarantees are least
valuable. We could allow for this correlation using a stochastic profit test, or implicitly by
taking a margin off the assumed lapse rate to reflect the negative correlation between
lapse rates and guarantees. As it happens, in the special case of a flat decision point and
25
under the Black-Scholes assumptions, there are analytical pricing solutions for such up
and out put options.
Overall, the decision point strategy has created a situation where the implied lapse rate for
valuing the charges is higher than the implied rate for valuing the charges. This is worse
than any single lapse assumption in a deterministic model. A deterministic profit test,
even with extreme behaviour assumptions, is simply incapable of producing outcomes as
bad as rational policyholders can force in a stochastic world.
Rational strategies are more difficult to formulate for more complex options. Consider for
example, a guaranteed minimum withdrawal benefit, under which a policyholder can
withdraw a guaranteed minimum amount per year, even if the fund assets are exhausted.
If the fund performs disastrously, and assets fall to zero, then the policyholder should
start withdrawing immediately, and should continue to do so, as any delay involves a loss
of interest on the sums withdrawn. On the other hand, if the fund performs very well
indeed, then an optimal strategy may be to surrender the policy and take the unit fund,
because the value of the future guarantee charges exceeds the value of the guarantees.
Finally, at intermediate fund levels, the best policy is to wait, in order to maximise the
time value of the embedded option. So optimal behaviour divides into three regions: an
upper lapse region, a middle wait region and a lower withdraw region. The
boundaries depend not only on the level of the fund but also on the number of guaranteed
payments still to be withdrawn. The optimisation is a complex computational task.
There are various approaches to reflecting partial rationality in profit testing models. One
common approach is to formulate dynamic lapse rules, where the rate of lapse is
expressed as a function of market levels. Although this approach is probably the most
popular, it is difficult to determine how far a particular dynamic lapse rule lies on the path
to rationality. Another informative approach is to assume that a certain proportion of
policyholders are rational, with the remainder following a dice-rolling approach. The
possibility of 100% rationality then emerges as a limiting case.
26
Economic Capital
When investigating the adequacy of a hedging program and a suitable level of capital to
hold for a portfolio of variable annuity business, a combined stress approach may be
used.
Definition: economic capital may be defined as the amount of capital required to maintain
the positive Net Asset Value of the fund with X%, typically 99.5%, confidence over one
year. Methods may vary between companies depending companies depending on which
future fees, capital and hedging costs are included.
The methodology used in the UK for ICA purposes is to calculate individual stresses at a
1 in 200 year likelihood in the underlying risk drivers. The level of these stresses used
for a variable annuity capital study may be as follows (largely based on the Solvency II
Qualitative Impact Study III stresses):
Equity
-32%
Fixed interest
+/- 200bp
Equity volatility
Lapses
*(+/- 50)/100
Expenses
Tracking error
8% of unit fund
27
Economic capital may be calculated with and without hedging and ultimately should be
calculated using a full economic or realistic balance sheet. Items to be included may be as
follows:
UL Assets
1,000,000,000
PV Future fees
160,000,000
Total
1,160,000,000
Unit Reserves
1,000,000,000
PV guarantees
41,000,000
PV management expenses
90,000,000
4,500,000
Cost of Hedging
4,500,000
8,000,000
Total
1,148,000,000
Excess assets
12,000,000
Also, the guarantee charges and outgo may be considered by themselves. This forms a
subset of the total economic capital, but is useful in seeing the impact of the shocks on
this part of the balance sheet in isolation. In particular, this enables an insurer to
investigate the adequacy of their hedging program under extreme market conditions.
In a recent study of a variable annuity product we found that after developing a hedging
strategy that matched the liabilities in delta and rho, there was still a substantial capital
requirement under and the interest rate down stress. This was because, for the product
concerned the present value of the guaranteed annuity payments was calculated using a
market yield curve. Therefore not only was the unit fund value able to bring the guarantee
in and out of the money, but also movements in interest rates, causing the present value of
the guaranteed amounts to increase. This movement was not linear with respect to interest
rates. (As mentioned previously hedging programs are designed to cope with small
movements in the underlying risk drivers).
The solution to this problem was to add swaptions to the derivative portfolio used for the
hedging. This meant that the optionality surrounding interest rate movements was
captured in the assets as well as in the liabilities. The swaptions substantially reduced the
amount of capital that was required in the 1 in 200 year stresses.
28
The risk drivers that have the most influence on the capital results vary through the term
of the contract, especially for contracts that have a long accumulation phase. This is due
to the strength of the unit linked VIF early on in the contract compared to the guarantee
costs and charges. An example set of capital results for a contract in force for one year are
as follows:
Equity
Interest rate
675,000
IR volatility
490,000
Equity vole
180,000
Lapses
1,800,000
Longevity
25,000
Expenses
170,000
Tracking error
370,000
560,000
Operational risk
Total
4,270,000
Diversification
1,990,000
Aggregate
2,280,000
In this example, there is a large diversification effect shown here due to negative
correlation between the different types of policyholder behaviour. Dynamic lapses are
triggered when the contract is out of the money and dynamic annuity take-up is triggered
when the product is in the money.
Tracking error has been calculated by regression analysis of past fund performance
compared to the tracker indices that will be used in the hedging.
29
Economic capital for the same block of business after 20 years may give results such as:
Equity
200,000
Interest rate
70,000
IR volatility
5,000
Equity vol
185,000
Lapses
50,000
Longevity
145,000
Expenses
8,000
Tracking error
370,000
695,000
Operational risk
Total
1,728,000
Diversification benefit
783,000
Economic Capital
945,000
These results are after twenty years of policy decrements and so are based on a lower
number of policies in force and a lower opening fund balance. However, the split of the
contribution to the capital of the various risk drivers can be examined and compared with
the figures for business one year in force.
In this business the result for the equity stress is far more dominant as the policy is closer
to annuitisation, therefore there is less time to recover from a fall in equity values. In this
example delta and rho hedging have been used but due to non-linearity in the hedging
there is still an equity capital requirement. Lapses are far less significant as the remaining
accumulation phase is much shorter so the impact of lapses on future free income is much
less than the one year case. The decision over whether or not to defer annuity take-up is
more dominant after twenty years. If annuities are taken up as soon as the guarantee is in
the money then the fund has time to recover and the cost to the company of the guarantee
biting is higher.
30
-3
-2
-1
0
-1
-2
-3
We can also graph a probability density function for the risk drivers. In the example
below a bivariate normal distribution has been used, giving a characteristic bell-shaped
surface.
-3
-2
-1
0
-1
-2
-3
31
3
2
1
0
-1
-2
-3
-3
-2
-1
We seek the drivers which are both painful (net assets negative) and likely (the cone
connects points of equal likelihood). Therefore we combine the two graphs to look for the
point where the line of insolvency is nearest to the sphere at the likelihood we are
examining. For European economic capital this is usually taken at 99.5%, giving 1 in 100
year events. A simplified version of the two graphs can be drawn as follows:
The above shows a two dimensional case where the LSLE is the point on the likely locus
that is closest to the line of ruin events.
A lapse increase may be bad, but a lapse increase at the same time as a rise in interest
rates may be very bad for the company. This could trigger a huge amount of lapses, and
32
increase operational and expense risk. An example set of economic capital and LSLE
stresses may look like this:
Stress
LSLE
Index
32%
-24.30%
Equity vol
10%
7.07%
Risk Free
2%
0.79%
Dividend
1%
0.22%
Track
5%
-1.42%
Behaviour
100%
33.47%
This shows that in a combined stress for this example, the Index and the policyholder
behaviour stresses are the most onerous for the company. This knowledge can be used to
focus further testing and stressing of the model.
Several iterations of the LSLE can be performed. This will eventually (typically after 4
interations) converge at the point that is most onerous and likely for the life insurer. The
capital requirements under the LSLE stress can then be compared with the requirements
under a traditional correlation matrix approach. The difference is due to the non-linearity
which is ignored in the correlation matrix approach but is implicitly allowed for in the
LSLE approach.
If all risks were fully correlated, non-linearity could be allowed for by applying all risks
simultaneously at the required aggregate level of confidence (e.g. 99.5%ile). For the
company, as the risks are not fully correlated, allowance should be made for
diversification.
A common approach to allow for non-linearity where diversification is present is to
reduce the level of confidence of each of the individual stress tests such that the sum of
the individual capital requirements equals the diversified capital amount (the medium
bang approach). The capital requirement allowing for non-linearity is then obtained by
applying all of the stress tests simultaneously at this level of confidence. This implicitly
makes allowance for the level of diversification around the base stresses rather than
around the actual stressed scenario of interest.
This approach would lead to an over estimate of non-linearity if there is greater
diversification around the stressed scenario. In the base stress tests, the capital
requirements due to changes in asset prices and asset volatilities can be due to the
hedging. However, in a "stressed" scenario, (e.g. where there have been fewer lapses
than expected) the fund would not be well hedged and therefore exposure to equity prices
and volatilities would be increased. Greater credit for diversification could be expected
because there are a greater number of active risks.
33
Due to the nature of variable annuity contacts, in particular the guarantees involved and
the interaction between market risk and lapse risk, the result using risk geographies can
be quite different to the result using a medium bang approach.
The difference between the two approaches has been analysed for a GMAB product and
gives the following results:
lapse
credit
track
vol
interest
equity
2
1
0
no div
correlation
mkt first
In this example, the capital requirement is higher using the LSLE approach than using the
undiversified combined stress approach.
34
Cost of Hedging
An initial hedge will be set up by the company. The company will choose how many of
the Greeks to hedge to. After the hedge has been set up it will diverge from its initial
position due to non-linearity as mentioned previously. Therefore regular rebalancing of
the hedge is required. Companies need to consider :
the allowance to make in their pricing for cost of capital and cost of hedging.
We can develop a model for this by thinking about the hedging process in the future.
Delta ()
max
-max
The hedging variable, say delta, will take an initial value. In this case delta = 0 at the start
of the process. Delta will move as the markets move, so will move about the perfectly
hedged position. However, the hedged will not be continuously re-balanced. The discrete
times of re-balancing are shown by the vertical lines in the diagram. The red horizontal
lines represent the limits for the hedge. Delta may go outside the limits but will this will
only be corrected at the time of re-balancing. Delta may correct itself and return within
the position limits without re-balancing.
If delta is outside the position limit at the time of re-balancing then the assets will be rebalanced to bring delta back inside the limit. It is optimal to not rebalance back to zero.
This is because transaction costs are cheaper if you only re-balance to the edge of the
limit. At the edge of the limit delta may move back inside the limit or out again with
equal likelihood. Therefore it may move closer to zero by itself without the extra
transaction costs of moving from the edge of the limit to zero. If you re-balance to zero
you will definitely pay these transaction costs. If you re-balance to zero you may pay
these transaction cost or you may pay less. Therefore it is more cost effective to not rebalance back to zero.
35
If we assume that the hedge position of the portfolio is a random walk, we can derive a
formula for it and call it Rt. The formula develops by considering the next period
transaction costs. This depends on how close to delta = 0 we are at the current period. If
we are close to delta = o then we expect that the next-period transaction costs will be less
than if we are close to the position limits. This is illustrated in the graph below:
If initial is close to
zero, then it is unlikely
to turn into a breach
within a single period.
-1
-0.5
0.5
Likewise, the next period capital costs increase the closer delta is to the position limit,
illustrated as follows:
Capital cost
-1
-0.5
0.5
36
The model assumes that Rt, is a Markov process and that we are required to investigate the
steady state distribution of Rt. This is approximated by assuming that the stationary
distribution has point masses of m at r = +/- lambda and a uniform distribution (with total
mass 1-2m) on (- lambda, lambda). Therefore the weighted average cost of capital and
cost of transactions can be estimated. The effect of different position limits on the cost of
transactions and the cost of capital can then be demonstrated. This may be graphed as
follows:
{ {
) }}
Rt + h = min max Rt + N 0, h ,
stress *
Capital =
Cost of hedge =
BOS *
2
4
COC * stress *
Cost of capital =
=
The position limit is:
BOS
3
*
COC * stress 4
37
10
Acknowledgements
The authors would like to thank Rob Green and Andrew Smith for their contributions to
this paper. Also Sarah Lowe and Pat Capobianco for assistance with the editing. Views
and errors remain, of course, our own.
38
11
Bibliography
Pricing the Option to surrender in Incomplete markets, Andra Consiglio and Domenico
De Giovanni, Draft March 2007, University of Palmero
Results of the Survey on Variable Annuity Hedging programs for Life Insurance
Companies, Charles Gilbert, K Ravindran, Robert Reitano, January 2007, Society of
Actuaries.
39
Appendix: US GAAP
In the US market, the accounting for the Variable Annuity differs by guarantee. The
GMDB and direct written GMIB fall under FAS 97. Clarification of how they should be
treated is covered in SOP-03-01. In essence this means that it is an EGP approach as per
FAS 97, but that the expected profits are calculated on a stochastic basis to determine the
profit stream.
GMABs, GMWBs and reinsured GMIBs are not considered to have any "insurance"
content and as such are treated as embedded derivatives and valued at fair value under
FAS 133. As there have been various interpretations as to what "fair value" really means,
this has now been clarified under FAS 157 which comes in force as per 31.12.2007.
The main points in the FAS 157 definition of "fair value include the need to add a
(potentially significant) risk margin. Other issues to note are the need to adjust for the
issuing company's own credit rating, the requirement to use market specific
assumptions on elements such as expenses and mortality (not company specific
assumptions), and the need to calibrate to market prices and market observable inputs
wherever they are available.
40