Professional Documents
Culture Documents
The Concept of IBNR
The Concept of IBNR
S. K. Dattagupta avers that reliable data is of essence to the reinsurers if they have to be successful in
loss-estimation and in pricing the contracts suitably.
most cases, losses increase from one The longest diagonal ie (250000,
year. 550000, 650000, 850000, 860000)
evaluation to the next. A completed
triangle (with hypothetical figure) will includes all payments made in during
d) Column '1' represents the year after
look as under. the year 2005as well as past calendar
the accident occurred etc.
years.
■
30
-
-----------------------i•i
The next step is to measure the increase / decrease be estimated. For example, the 2005 -12
(Ratio) months evaluation of R 600000 (Table 1)
Months of development is multiplied by the 12 month to ultimate
Loss period
12 (a) 24 (b) 36 (c) 48 (d) 60 (e) loss development factor of 3.207 to yield
2001 2.200 * 1.181 ** 1.307 *** 1.011 -------- an ultimate estimated loss of R 1924200.
2002 1.750 1.228 1.116 7. Link ratio.
2003 1.420 1.338 Year 12 to 24
2004 1.818 months
(Table 4)
(Ratio)
Months of development
Loss period 8. Homogeneous data
12 to 24-(a) 24 to 36-(b) 36 to 48-(c) 48 to 60 -(d) 60 to ult -(e)
It is necessary to have separate reviews for
average 2.200 1.181 1.307 1.011
loss development data and IBNR
industry 2.000 1.200 1.100 1.011 1.110
selected 2.200 1.181 1.100 1.011 1.110 calculations for different insurance
cumulative 3.207* 1.457 ** 1.234*** 1.122**** 1.110 ***** portfolios. This is because the liability and
loss development patterns differ widely
(Table 3)
despite the fact that in short term period
* Ratio obtained by (a) x (b) x (c) x (c) x (d) an average that excludes the high or low there is possibility of combination of
x (e) points may be used. Selected factors are portfolios and variability in combined
** Ratio obtained by (b)x (c) x (d) x (e) usually a combination of the averages and portfolios are not material. Due weight
*** Ratio obtained by (c) x (d) x (e) industry. Because of the consistency in the age should be given on the volume of
**** Ratio obtained by (d) x (e) data in the above example, the selected data while considering homogeneity.
***** Ratio obtained by industry factor of ratios equals the average for all periods Breakup of data should be pursued only
60 to ultimate since unique data beyond except 60 months to ultimate for which where it is feasible and subdivision is
irda journal May 2012
60 months is not available. the industry factor is chosen. Once the useful. As an eminent expert puts it "If we
development between periods is try to slice the cake more than one way at
The above table is a case of straight
estimated, then the total development is a time, we shall be left with a useless
averages. For volatile data, weighted
calculated. The ultimate losses can then collection of crumbs".
averages / two or three years average or
■
31
thinking cap
casualty business they are to be carried be used but the inflation rate inherent in • Incurred development
over number of years for which the size of claims may be quite different i.e. price of
consumer goods, cost of settling claims, • Paid development
IBNR may become large.
and special indices of inflation like cost of
• Frequency & severity
12. Claim adjustment philosophy repairing vehicles which needs to be
It is necessary to have a close monitoring separately considered. • Projection & completion.
of the reserve for effecting changes either
at the year end or during the year for 14. Pre-reserving ratios / indicators. In the Incurred Development Method,
change in loss development exposure or The following important ratios need to be known incurred losses - payments plus
changes in the reserve required than what worked out before creation of reserves; outstanding reserves are developed to
was made in earlier periods. The reserve estimate ultimate losses using incurred
• Paid loss to incurred loss - for
should be tuned to the actual changes loss development factors. The method can
ascertaining the consistency of the
from the expected. be effectively applied when historical data
development of paid and reported
from a particular exposure is used to
losses and gives a warning of case
13. Inflation calculate loss development factor, since,
reserves inadequacies.
irda journal May 2012
■
32
Paid Development Method uses historical information may be left uncovered within from the fact that important information
loss development factors of paid losses for the aggregate incurred or paid amounts. contained in current loss data may not
which there is no effect on estimates even The method may be of limited value be unused .
if reserving procedures are altered. The because the data could lack stable
16. Methods to assess IBNR -
disadvantage of this method is that, for patterns unless the volume is large.
(deterministic methods)
early period of development the factors
Projection & Completion Method is There are number of methods to assess
required are large. Any larger claim
suitable for estimating losses for the IBNR. In order to understand the methods
payment may have a significant effect on
current policy period. Here, the estimates one must understand the loss
ultimate loss estimates.
of ultimate losses for earlier historical development factors underlying those
The Frequency & severity Method is useful policy periods are adjusted to current year methods. Prominent amongst them are;-
where experience analysis is needed for cost levels by considering changes in
• Incurred Chain Ladder
an acquisition or divestiture. By looking exposures and average loss costs. The
separately at severity and frequency estimated ultimate loss for recent earlier • Paid Chain Ladder
elements, probable adjustments to periods are adjusted to current years cost
estimated ultimate losses should be levels by trend factors eg consumer price • Incurred Bornheutter- Ferguson
quantified and then the two can be index or more specific index. The method
• Paid Bornheutter- Ferguson
recombined to arrive at the estimated has the advantage of not being influenced
ultimate losses. However the by unusual small/large claims or by • Exposure based methods
disadvantage of the system is that in such changes in claim handling and reserve
separate analysis, some important philosophies but suffers disadvantage • Expected loss ratio
Age 12
Ultimate loss development factor
24 36 48 60 72 84
I projection method and can be done using
irda journal May 2012
■
33
thinking cap
applied to a triangle of loss ratio data market data could be used as a basis for
rather than cumulative payments. an estimate of the eventual loss and
It is assumed that for hence the outstanding claims.
Advantage of Chain ladder system.
• Easily understandable by all
each origin year, both
Advantage of expected loss ratio
• Considers aspect of changing the number and method.
development average amount of • Can be used with no previous / internal
• Cent percent credibility to actual claims relating to each loss history
experience given. • Useful for new or volatile line of
development year are
business
Disadvantage of Chain ladder system. constant proportions
• Unaffected by settlement speed or case
• Too responsive to data resulting in of the totals from that reserve adequacy.
years to develop a report and affected
origin year.
by volatile data either big loss or no Disadvantage of expected loss ratio
loss. method.
• Affected by settlement speed • Doesn't apply to changing
• May not be reliable for most recent environment
accident year. to ascertain the development factor or • Ignores useful information.
grossing up factor. A grossing up factor
The inflation adjusted chain ladder gives the proportion of the ultimate claim Bornhuetter Ferguson method.
system require adjustment in the figures amount that has been paid so far .The The system takes into account loss ratios
in the triangle to allow for the effects of projected ultimate claim can be calculated (relative to the premium) plus chain
inflation. Generally it is assumed that by multiplying together for each accident ladder system in calculating reserves. The
inflation is at the same annual rate for all year the projected figures for the average concepts behind the method are;
claims within a particular calendar year of claim amounts and claim numbers. A
• Claims developed in relation to a given
payment in case of past inflation. In case reserve is the calculated by subtracting all
origin year will follow that experienced
of future inflation, an assumed rate of payments to date in respect of claims
for other origin years in future.
future inflation is needed and it is also relating to data of claims. There is no
necessary to convert to non cumulative unique way of defining the method and • The past development for a given
data rather than cumulative totals before assumption. However it is assumed that origin year does not provide a better
adjusting these for future inflation as is for each origin year, both the number and clue to future claims than the general
used in past inflation . average amount of claims relating to each loss ratio.
d e v e l o p m e n t y e a r a re co n s t a n t
Average cost per claim method. The future claim development can be
proportions of the totals from that origin
The method requires development tables expressed as Premium x Estimated loss
year.
for both total claim amounts and claim ratio (1-1/ f). As the final estimate of the
numbers. Normally cumulative figures are Expected loss ratio method. ultimate loss is based on observed data
irda journal May 2012
used. A third development table of The ratio of incurred claims to earned and an initial estimate ignoring the
average claim amount is formed by premiums over a defined period is called observations, this method can be viewed
dividing the figures in the corresponding the loss ratio. The ratio is based on trends as using a Bayesian approach. Using the
cells of the first two tables. The next step is of past data, underwriter's views or method each year's revised ultimate loss
■
34
can be expressed as a combination of the Advantage of Bornhuetter - Ferguson Disadvantage of Bornhuetter -
chain ultimate loss (CL) and the method. Ferguson method
independent loss ratio ultimate loss (LR). • Over-reaction is avoided as the method • Reliance on expected loss ratio
Using a credibility factor (Z) it is expressed doesn't apply development factors to • Requirement on development factors.
an unusual claim occurrence.
as BF revised ultimate loss = Zx CL + (1-Z) x
• Suitability in new or volatile line of A triangular representation of cumulative
LR, where Z= A/CL and A is the actual
business. incurred claims of a particular insurance
claims developed at that time. Thus BF portfolio in BF method is given in Table 7
• Good adjustments between loss
revised ultimate loss can be finally development and expected loss ratio and that of hypothetical IBNR calculation
expressed as A+ (1-1/f) x LR. methods. are given in Table 8 and Table 9
(Table 7)
Hypothetical loss development data
Incurred case losses (including Allocated Claim Expense)
Net Earned Year First Second Third Fourth Fifth Sixth
Premiums Report Report Report Report Report Report
1000000 1999 500000 650000 700000 720000 721440 720718
1.300 1.077 1.028 1.002 .999
1500000 2000 460000 621000 677511 707321 706613
1.350 1.091 1.044 .999
1600000 2001 595320 762010 803921 820000
1.280 1.055 1.020
1800000 2002 881407 918426 957000
1.402 1.042
1850000 2003 596381 853421
1.431
2000000 2004 920000
(Table 8)
Three year data Loss Dev factor Accident Expected Expected Indicated Loss Adjusted
Dev R Year osses losses IBNR Method Loss method
period Indicated To R IBNR (6x7) IBNR IBNR R
beginning ending Factor** (~)
(2/1) ultimate
1st *2073108 *2523857 1.217 ^1.327 2004 1900000 .246 467400 300840 257587
to 2nd
2nd 2301436 2438432 1.059 1.090 .2003 1757500 .083 145872 76807 80825
to 3rd
3rd 2181431 2247321 1..030 1.030 .2002 1710000 ..030 51300 28710 27967
to 4th
irda journal May 2012
■
35
thinking cap
Explanation
selected than a higher ratio is required on incurred losses shown in triangle diagram
On the above exhibits a purely
conservative basis. (R 920000 x .327; R 853421 x .090).A
hypothetical IBNR reserve computation is
refinement result shown in column 10 is
displayed. Table 7 portrays the actual case Column 7 sets forth the conversion of the sometime used to smooth out the data is
loss developments along with the loss development factors in column 4 to a to average the case incurred for the same
individual loss development factors by basis appropriate for use with expected report, say for two years; adjust that
interval. In the left column, the earned losses. The factor of 246 for accident year average case incurred figure to reflect the
premiums net of commission for the 2004 indicates that 75.4 % of the total difference between the average exposure
financial year associated with each losses for that accident year have been for two years studied and the current year;
accident year are displayed. This is the reported as of the first report (31 March times the IBNR factor eg;
basic tool with which IBNR reserve is 2004) and thus that 24.6 % of the losses
developed.
~ (R 920000+ R 596381) x R 2000000 x .327
----------------------------- ------------------------------ = R 257587
Table 8 sets forth the actual IBNR reserve 2 R 1850000 + R 2000000
computation based on the data in Table --------------------------------
7. To produce average loss development 2
factors for each interval (one report to the (R 918426+ R 853421) x R 1850000 x .090
next) the latest three years data were ----------------------------- ------------------------------ = R 80825
2 R 1850000 + R 1800000
used. Columns (1 &2) are the basic data
--------------------------------
from the loss development triangle in 2
Table7.* For example, the sum of the first
(R 957000+ R 803921) x R 1800000 x .030
reports for accident years 2001, 2002 and
----------------------------- ------------------------------ = R 27967
2003 is R 2073108 while R 2523857 is the 2 R 1600000 + R 1800000
sum of incurred losses as of the second --------------------------------
2
report for those same three accident
years. The division of these figures
are yet to emerge. The factor of. 246 is In the above example each method
produces the average loss development
determined as follows; produced a different IBNR reserve. The
factor of 1.217 shown in column 3.
first method (IBNR reserve as a function of
Column 4 is simply the upward
accumulation of column 3 loss IBNR factor = 1.000-1.000/ loss expected losses) could be brought more
development factor to project a given development factor to ultimate into line with the other two methods if it
accident year to its estimated ultimate was felt that a lower expected loss ratio
result. ^The factor from the third report to =1.000-1.000/ 1.327 = .246 could be justified. If the ultimate loss ratio
ultimate is obtained by multiplying 1x to premiums earned net of commission
1.030 while the factor1.090 second to The IBNR reserves are set forth in column 8 was completely predictable, all the three
ultimate is the product of 1x 1.030 x 1.059. and are the product of IBNR factors in
methods would produce the same reserve
column 7 and the expected losses in
If loss ratios were that predictable, the
The expected losses for each accident year column 6.Thus R 467400 is the IBNR
determination of IBNR reserves would be
are set forth in column 6.In the example reserve assigned to accident year 2004
a trivial matter. One would simply subtract
they are obtained by applying an while R 145872 is allocated to accident
the incurred losses to date by accident
expected loss ratio of 95 % (after year 2003 and R 51300 for accident year
year from the expected ultimate incurred,
adjustment to eliminate estimated 2002 resulting in a total reserve of
R 664572 at 31 March 2004. the remainder being the needed IBNR
overhead costs). The selection of an
estimated loss ratio is affected by the reserve. It is necessary to workout loss
irda journal May 2012
stability of the data. If a company proves The technique for developing the IBNR d e v e l o p m e n t re v i e w s a n d I B N R
that it can consistently produce an reserve as a function of case incurred calculations independently for type of
ultimate loss ratio of 50 % it can apply the losses. Column 9 Table 8 involves applying business which is different. Thus OD,
same in IBNR calculation. If it is proved the loss development factor, column 4, Health, Fire, Marine and Miscellaneous are
that an expected loss ratio cannot be Table 8, less unity to the proper case calculated separately. As stated earlier
■
36
that good number of adjustments is Loss development factor for the period-- 1 quarters. Here one can rely on empirical
required to be made for internal and (unity) x Indicated IBNR formula or judgments. Where during
external factors like inflation, reinsurance ----------------------------------------------- monitoring there is deterioration, this
adjustments, claim adjustment Ultimate loss development factor for the should be scrutinized to ascertain where it
philosophy, product mix, processing lags, period-- 1 (unity) Or exists and adjustment made on
no-fault programmes etc before arriving 1.217-- 1 x 467400 judgmental basis.
at final reserves estimate
------------------- --- = 310170
Note
1.327-1
Interim reserving techniques 1. Increase in current year's exposure -
Between the annual calculations of an A summation of similar calculations over R 365000 (R 1705000 - R 1340000).
IBNR reserve structure it is necessary to all accident years produces the total Increase in total exposure - R 2574501
periodically review the reserve from two expected developments for the calendar (R 4995050 - R 2420549).
angles. At the close of each accounting year.
2. Increase in total IBNR R 513049
period it must be determined if changes in
If it is desired to tie IBNR to expected losses (R 2420549 - R 1907500)
the amount of reserve are necessitated by
changing exposures. Secondly the reserve instead of actual losses, then 3. It has been assumed that there was no
established at the prior year end must be determination of expected development change in IBNR factors and that
continuously monitored to see if the loss will be determined as under; expected losses were direct function of
developments observed are what were premium earned
Expected development of next calendar
contemplated when the reserve was
year= expected losses for a given accident 4. There was upward disproportionate
established. To the extent that the actual
year x (IBNR factor of current year - IBNR trend in premium growth
development is different from the
factor of previous year). Based on data at 5. Increase in total IBNR ie R 513049 is
expected and credible, the reserve
Table 8, the 2004 accident year greater than current year's exposure of
structure should be 'fine - tuned'. In order
development factor will be; R365000 although sounds exceptional
to determine the amount of development
expected in a year following the is due to increase in current year's
1900000x (.246- .083) = 309700.
establishment of reserve the following exposure by R 365000 as against the
formula is adopted taking data from Table total increase of exposure by
After determination of the expected
B; Thus expected development in 2005 on R 2574501.
developments for the year, the next step is
the accident year 2004 would be derived
to allocate the expected developments to 17. IBNR & IFRS.
as follows
International Financial Reporting
(Table 9) Standards (IFRS) doesn't consider future
Hypothetical IBNR calculation liability. IFRS will bring a complete change
As on 31 March 2003 As on 31 March 2004 in calculation of reserves. In IFRS all
Year Expected IBNR IBNR Expected IBNR IBNR prospective cash flows including
losses-R factor Reserve-R losses R factor Re serve
R projected claims will be recognized when
~
1999 600000 -.033 -20000 --- ---- the insurance contract is written which
indicates that there will be immediate
2000 800050 .478 382500 800050 -.033 -26401
recognition of the contract's likely profit or
2001 1150000 .543 625000 1150000 .478 549700
loss considering all acquisition costs,
2002 1340000 .686 920000 1340000 .543 727620
irda journal May 2012
■
37
thinking cap
techniques will have to be supplemented provides input for financial analysis and
by probabilistic or stochastic analysis calculation of economic capital. The
using simulations to forecast projected a number of Guidance Note/Actuarial Standard of
claims. Practice in UK, USA and Australia are
projections are to be
summarized as under;
The second change is the discounting of initiated that would
the reserves by general insurers. This will • "Uncertainty for a particular point
play a dominant part
remove the buffer of investment income estimate would normally be quantified
on the reserves and may render the in the best estimation by providing a range of values around
general insurer's Balance Sheet more of assets and the point estimate together with an
volatile. In this situation the margin of indication of the likelihood that the
liabilities and thus the
error in the estimation of reserves will be true values lie above, below or simply
further reduced and this further solvency of the outside the range." Provision of a range
emphasizes the need for improved company. of outcomes without a specific point of
reserving techniques to be used by estimate could be open to
general insurers in India. misrepresentation. - UK.
■
38
estimate of the standard error of the parameterization issue as well as Observation:-
mean. reduces error involved in curve fitting. • Actuarial Estimate is less than Chain
Ladder Estimate.
• Bayesian evaluation - expands the Reasonable range concept.
analytical approach by treating the Range arises from uncertainty associated • Faster claim settlement in the previous
parameters of the fitted model as a with. The range of reasonable estimates is two years may be plausible reasons for
further set of random variables. a range of estimates that would be actuarial low estimate.
produced by alternative sets of
• Parametric - to overcome over- assumptions that the actuary judges to be • There is a need to investigate further
parameterization caused in chain reasonable considering all information and create lesser liability due to
ladder method a parametric curve be reviewed by the actuary. The concept obvious reasons.
fitted to loss development pattern. considers the reserve range for location of
actuarial estimate. The same is
• Non- Parametric - addresses the over-
diagrammatically represented.
50%
Probability 75%
Expected Value
Mode
Median
Liability Estimate
■
39