AIG 2013 Annual Report Download - page 201

the losses. For example, for a long-tail class of business for which only 10 percent of the losses are expected to be
reported at the end of the accident year, the expected loss ratio would be applied to the 90 percent of the losses still
unreported. The actual reported losses at the end of the accident year would be added to determine the total
ultimate loss estimate for the accident year. Subtracting the reported paid losses and loss expenses would result in
the indicated loss reserve. In the example above, the expected loss ratio of 70 percent would be multiplied by
90 percent. The result of 63 percent would be applied to the earned premium of $10 million resulting in an estimated
unreported loss of $6.3 million. Actual reported losses would be added to arrive at the total ultimate losses. If the
reported losses were $1 million, the ultimate loss estimate under the ‘‘Bornhuetter Ferguson’’ method would be
$7.3 million versus the $7 million amount under the expected loss ratio method described above. Thus, the
‘‘Bornhuetter Ferguson’’ method gives partial credibility to the actual loss experience to date for the class of
business. Loss development methods generally give full credibility to the reported loss experience to date. In the
example above, loss development methods would typically indicate an ultimate loss estimate of $10 million, as the
reported losses of $1 million would be estimated to reflect only 10 percent of the ultimate losses.
A key advantage of loss development methods is that they respond quickly to any actual changes in loss costs for
the class of business. Therefore, if loss experience is unexpectedly deteriorating or improving, the loss development
method gives full credibility to the changing experience. Expected loss ratio methods would be slower to respond to
the change, as they would continue to give more weight to the expected loss ratio, until enough evidence emerged to
modify the expected loss ratio to reflect the changing loss experience. On the other hand, loss development methods
have the disadvantage of overreacting to changes in reported losses if the loss experience is not credible. For
example, the presence or absence of large losses at the early stages of loss development could cause the loss
development method to overreact to the favorable or unfavorable experience by assuming it will continue at later
stages of development. In these instances, expected loss ratio methods such as ‘‘Bornhuetter Ferguson’’ have the
advantage of recognizing large losses without extrapolating unusual large loss activity onto the unreported portion of
the losses for the accident year.
Frequency/severity methods generally rely on the determination of an ultimate number of claims and an
average severity for each claim for each accident year. Multiplying the estimated ultimate number of claims for
each accident year by the expected average severity of each claim produces the estimated ultimate loss for the
accident year. Frequency/severity methods generally require a sufficient volume of claims in order for the average
severity to be predictable. Average severity for subsequent accident years is generally determined by applying an
estimated annual loss cost trend to the estimated average claim severity from prior accident years. In certain cases,
a structural approach may also be used to predict the ultimate loss cost. Frequency/severity methods have the
advantage that ultimate claim counts can generally be estimated more quickly and accurately than can ultimate
losses. Thus, if the average claim severity can be accurately estimated, these methods can more quickly respond to
changes in loss experience than other methods. However, for average severity to be predictable, the class of
business must consist of homogeneous types of claims for which loss severity trends from one year to the next are
reasonably consistent. Generally these methods work best for high frequency, low severity classes of business such
as personal auto.
Structural drivers analytics seek to explain the underlying drivers of frequency/severity. A structural drivers
analysis of frequency/severity is particularly useful for understanding the key drivers of uncertainty in the ultimate loss
cost. For example, for the excess workers’ compensation class of business, we have attempted to corroborate our
judgment by considering the impact on severity of the future propensity for deterioration of an injured worker’s
medical condition, the impact of price inflation on the various categories of medical expense and cost of living
adjustments on indemnity benefits, the impact of injured worker mortality and claim specific settlement and loss
mitigation strategies, etc., using the following:
• Claim by claim reviews to determine the stability and likelihood of settling an injured worker’s indemnity and
medical benefits — the claim file review was facilitated by third party specialists experienced in workers’
compensation claims;
• Analysis of the potential for future deterioration in medical condition unlikely to be picked up by a claim file review
and associated with potentially costly medical procedures (i.e., increases in both utilization and intensity of medical
care) over the course of the injured worker’s lifetime;
• Analysis of the cost of medical price inflation for each category of medical spend (services and devices) and for
cost of living adjustments in line with statutory requirements;
..................................................................................................................................................................................................................................
AIG 2013 Form 10-K 183
ITEM 7 / CRITICAL ACCOUNTING ESTIMATES
..................................................................................................................................................................................