SEC Filing | American International Group, Inc.

View:

CORRESP

September 10, 2008

Mr. Jim B. Rosenberg
Senior Assistant Chief Accountant
United States
Securities and Exchange Commission
Division of Corporation Finance
100 F Street, NE
Mail Stop 6010
Washington, D. C. 20549


Re:		American International Group, Inc. Form 10-K for the Year Ended December 31, 2007 Form 10-Q for the Quarter Ended June 30, 2008 File No. 1-8787

Dear Mr. Rosenberg:

We are in receipt of your letter dated September 5, 2008 and thank you for your comments concerning American International Group, Inc.’s (AIG) captioned filings. We are pleased to respond to your questions and to have the opportunity to work with the Staff of the Commission to further enhance the overall presentation and disclosure in our filings.

As more fully described below, we intend to expand certain of our disclosures in our Form 10-Q for the quarter ended September 30, 2008. Our responses and those disclosures are provided below. If further clarification or information is required, please let us know at your earliest convenience.

AIG acknowledges that the adequacy and accuracy of the disclosure in AIG’s filings is the responsibility of AIG, that Staff comments or changes to disclosure in response to Staff comments do not foreclose the Commission from taking any action with respect to the filings and that Staff comments may not be asserted by AIG as a defense in any proceeding initiated by the Commission or any person under the Federal securities laws of the United States.

We have repeated your questions below to facilitate your review.

Form 10-Q for the Fiscal Quarter Ended June 30, 2008

Note 3-Fair Value Measurements, page 13

Please refer to your tabular disclosure on page 17. Please explain the basis for the counterparty netting adjustments shown in the table. Please tell us whether any portion of the counterparty netting adjustments are related to your super senior credit default swap portfolio, and explain how the adjustments are consistent with requirements of SFAS 157 or other applicable literature.

			AIG Response:

			The counterparty netting adjustments in the table on page 17 of the June 30, 2008 Form 10-Q were made in accordance with the provisions of paragraph 10 of FIN 39, as interpreted by FSP FIN 39-1. FIN 39 sets forth guidelines for when it is appropriate to present certain financial assets and liabilities on a net basis on the balance sheet. There are special provisions for derivatives that are subject to master netting agreements, whereby derivative balances with the same counterparty are to be reported net on the balance sheet, as long as the following criteria have been met:

	1)		Each of two parties owes the other determinable amounts

	2)		The reporting party has the right to set off the amount owed with the amount owed by the other party

	3)		The right of setoff is enforceable at law

			Standard ISDA Master Netting Agreements used by AIG and throughout the over-the-counter derivative industry meet these criteria. As described more fully in Note 1 to AIG’s June 30, 2008 financial statements, FSP FIN 39-1 provides an interpretation of FIN 39 that requires cash collateral posted/received by AIG with respect to master netting agreements to also be netted against the derivative balances with the same counterparty when applying the netting provisions of paragraph 10.

			AIGFP’s super senior credit default swaps are subject to such Master Netting Agreements and thus the counterparty netting adjustments in the tabular disclosure on page 17 include adjustments to set off fair value amounts recognized and amounts recognized for the cash collateral posted for super senior credit default swaps.

			The cash collateral posted and other derivative positions with a particular counterparty, including the super senior credit default swaps, are incorporated into the determination of the effect of AIG’s own credit risk on the fair value of the derivative liability.

Management’s Discussion and Analysis of Financial Condition and Results of Operations

Critical Accounting Estimates

AIGFP’s Super Senior Credit Default Swap Portfolio, page 49

You disclose that the unrealized market valuation losses on your super senior credit default swap portfolio represents the amount you would need to pay a willing, able and knowledgeable third-party to assume the obligations. Please revise your disclosure to discuss how you determined the principal or most

			advantageous market in accordance with SFAS 157, and disclose who you believe the market participants would be.
			AIG Response:

			In response to the Staff’s comment, AIG intends to include additional disclosure in its Form 10-Q for the quarterly period ended September 30, 2008 as follows (please refer to Form 10-Q for the Quarter Ended June 30, 2008, page 15 and page 42):

The principal market was determined to be the market in which super senior credit default swaps of this type and size would be transacted, or have been transacted, with the greatest volume or level of activity. AIG has determined that the principal market participants, where available, therefore, would consist of other large financial institutions who participate in sophisticated over-the-counter derivatives markets.

You have provided extensive disclosure regarding the valuation of CDS relating to multi-sector CDOs, but appear to have provided only limited disclosures regarding the valuation of CDS relating to corporate debt/CLOs. Please revise to provide disclosure of the contractual terms, methodology, inputs and assumptions regarding the valuation of your CDS relating to corporate debt/CLOs.

AIG Response:

In response to the Staff’s comment, AIG intends to include additional disclosure in its Form 10-Q for the quarterly period ended September 30, 2008 as follows (please refer to Form 10-Q for the Quarter Ended June 30, 2008, page 15 and page 49):

The valuation of credit default swaps written on portfolios of investment-grade corporate debt and collateralized loan obligations (CLOs) is less complex than the valuation of super senior multi-sector CDO credit default swaps and the valuation inputs are more transparent and readily available.

In the case of credit default swaps written on portfolios of investment-grade corporate debt, AIGFP estimates the fair value of its obligations by comparing the contractual premium of each contract to the current market levels of the senior tranches of comparable credit indices, the iTraxx index for European corporate issuances and the CDX index for U.S. corporate issuances. These indices are considered to be reasonable proxies for the referenced portfolios.

AIGFP estimates the fair value of its obligations resulting from credit default swaps written on CLOs to be equivalent to the par value less the current market value of the referenced obligation. Accordingly, the value is

determined by obtaining third-party quotes on the underlying super senior tranches referenced under the credit default swap contract.

	4.		You disclose that your modified version of the Binomial Expansion Technique (BET) model to value your credit default swap portfolio written on super senior tranches of CDOs of asset-backed securities. Please clarify how the output of the modified BET model translates into the amount that you would need to pay to a willing, able and knowledgeable third-party to assume the obligations.

			AIG Response:

			The underlying assumption of the valuation methodology is that, in order to be willing to assume the obligations under a credit default swap, a market participant would require payment of the full difference between the cash price of the underlying tranches of the referenced securities portfolio and the notional amount specified in the credit default swap.

			The BET model uses assumptions with respect to market prices, weighted average lives and recovery rates for each underlying security of each CDO, the diversity score of each CDO and discount rates. In the absence of observable market data for transfers of such liabilities, AIG believes that the BET model provides a reasonable basis for estimating the fair value of AIGFP’s credit default swap portfolio.

	5.		Please revise to disclose how the BET model calculates the probabilistic measure of expected loss for your credit default swaps. Explain each significant step in the valuation process.

			AIG Response:

			In response to the Staff’s comment, AIG intends to include additional disclosure in its Form 10-Q for the quarterly period ended September 30, 2008 as follows (please refer to Form 10-Q for the Quarter Ended June 30, 2008, page 49):

AIGFP has adapted the BET model inputs so that the model utilizes the price for the securities underlying the CDO as an input and converts the price to a credit spread using parameters consistent with those used for valuation. This credit spread is used to determine implied probabilities of default and expected losses on the credit default swap. For each individual super senior tranche of the CDO in the portfolio, the application of the modified BET model to determine the probabilistic measure of expected loss involves the following steps:

	1)		Calculation of the cash flow pattern that matches the weighted average life for each underlying security of the CDO;

	2)		Calculation of an implied credit spread for each security from the price and conversion of the spread into the probability of default;

	3)		Generation of expected losses for the security using the probability of default and recovery rate;

	4)		Aggregation of the cash flows for all securities to create a cash flow profile of the entire collateral pool within the CDO;

	5)		Division of the collateral pool into a number of independent identical securities (idealized securities) based on the CDO’s diversity score;

	6)		Simulation of the default behavior of the idealized securities using a Monte Carlo simulation and aggregation of the results to derive the effect of the expected losses on the cash flow pattern of the super senior tranche taking into account the cash flow diversion mechanism of the CDO;

	7)		Discounting of the expected cash flows to estimate the value of the super senior tranche of the CDO; and

	8)		Adjusting of the model value for the super senior multi-sector CDO credit default swap for the effect of the risk of non-performance by AIG using the credit spreads of AIG available in the marketplace.

	6.		Please refer to prior comment two of our letter dated June 25, 2008. Please revise your disclosure to provide a summary that quantifies the CDO pool pricing assumptions for each period presented, and explains any significant changes between periods.

			AIG Response:

			In response to the Staff’s comment, AIG intends to include additional disclosure in its Form 10-Q for the quarterly period ended September 30, 2008 as follows (please refer to Form 10-Q for the Quarter Ended June 30, 2008, page 49):

			The most significant assumption used in the BET model is the pricing of the individual securities within the CDO collateral pools. The following table summarizes the weighted average price at June 30, 2008 and September 30, 2008, and the percentage of the total CDO collateral pools at September 30, 2008, by ABS category.

Percentage of

Weighted

Total CDO

Average Price

Collateral Pools

Weighted Average

September 30,

ABS Category

Price June 30, 2008

2008

Inner CDOs

XX%

CMBS

XX%

Prime

XX%

Alt-A

XX%

Subprime

XX%

Other

XX%

Total

XX%

100.00

In addition to the table, AIG will include disclosure in its Form 10-Q for the quarterly period ended September 30, 2008 for any significant changes between periods.

You disclose that the most significant assumption in your CDS valuation is the pricing of securities within the CDO collateral pools, which you state are “obtained in most cases from the CDO collateral managers, to the extent available.” Please expand your disclosure to include the following information:

	•		Describe the relative significance and effect on the valuation of each pricing methodology used. (A)

	•		Describe more specifically the third party pricing services and in-house valuation models used by collateral managers. (B)

	•		Describe more specifically your analysis and testing that verifies the accuracy and completeness of pricing information provided by collateral managers. (C)

	•		Quantify the impact of the time lag in prices provided by the collateral managers. Describe how and to what extent this time lag affects your fair value estimate. (D)

	•		Describe more specifically the methods and assumptions used to determine pricing for individual securities when the collateral manager has been unable to provide this information. (E)

	•		Explain the relative significance of the Monte Carlo simulation process to the overall valuation process. If the Monte Carlo simulation process has a material effect on the valuation, describe the process in more detail. (F)

			AIG Response:

			For the convenience of the Staff, we have labeled the subparts of comment 7 and respond as follows.

Relative significance and effect on the valuation of each pricing methodology used

For the quarter ended June 30, 2008, CDO collateral managers provided market prices for approximately 75 percent of the underlying securities. When a price for an individual security is not provided by a CDO collateral manager, AIGFP derives the price through a pricing matrix using prices from CDO collateral managers for similar securities.

Third party pricing services and in-house valuation models used by collateral managers

Based on due diligence, AIGFP determined that approximately 95 percent of the CDO collateral managers who provide prices use dealer prices for all or part of the underlying securities, in some cases supplemented by third party pricing services, primarily IDC. AIG has evaluated IDC’s pricing model in connection with its use of the IDC pricing service for a significant portion of its investments in similar securities. AIGFP’s due diligence allowed it to understand how the CDO collateral managers were sourcing the market prices of the underlying securities.

Analysis and testing that verifies the accuracy and completeness of pricing information provided by collateral managers

AIGFP reviews general trends in the pricing data for consistency with market observations. The pricing information obtained from the CDO collateral managers is also indirectly validated by comparing the BET model results to the prices collected from third parties for the super senior tranches of the CDOs and to the prices inferred from collateral calls received from AIGFP’s counterparties.

Time lag

The time lag is approximately one month. AIGFP rolls forward these prices to the end of the quarter using data provided by a third-party pricing service. Therefore, the inputs to the model reflect the market information available as of the end of the quarter. Further, the output of the valuation model is compared to current market information to ensure that it reflects the current market conditions and thus the quarter-end pricing reflects fair value as of that date.

Methods and assumptions used to determine pricing for individual securities when the collateral manager has been unable to provide this information

When a price for an individual security is not provided by a CDO collateral manager, AIGFP derives the price through a pricing matrix uses prices from CDO collateral managers for similar securities.

Monte Carlo simulation

The Monte Carlo simulation allows AIGFP to take into account a simplified version of the diversion mechanism of each CDO. The Monte Carlo simulation is used to determine whether an idealized security defaults in a given simulation scenario and, if it does, the security’s implied random default time and expected loss. Cash flow streams for the idealized assets are then created using the computed constant prepayment rate and the simulated default time. These cash flows are passed through to the liability tranches using the cash flow diversion algorithm, which uses all significant provisions including balances, tranche spreads, over-collateralization tests, and over-collateralization event of default triggers to allocate the available cash flows and compute resulting losses. The losses are aggregated across all idealized assets in the idealized portfolio and averaged across the Monte Carlo scenarios to arrive at portfolio expected losses.

AIG intends to modify and expand its previous disclosure with respect to the above in its Form 10-Q for the quarterly period ended September 30, 2008 as follows (please refer to Form 10-Q for the Quarter Ended June 30, 2008, page 49):

AIGFP’s Super Senior Credit Default Swap Portfolio

AIGFP values its credit default swaps written on the most senior risk layers (super senior) of designated pools of debt securities or loans using internal valuation models, third-party prices and market indices. The specific valuation methodologies vary based on the nature of the referenced obligations and availability of market prices.

AIGFP uses a modified version of the Binomial Expansion Technique (BET) model to value its credit default swap portfolio written on super senior tranches of CDOs of asset-backed securities (ABS), including maturity-shortening puts that allow the holders of the securities issued by certain CDOs to treat the securities as short-term eligible 2a-7 investments under the Investment Company Act of 1940 (2a-7 Puts). The BET model uses default probabilities derived from credit spreads implied from market prices for the individual securities included in the underlying collateral pools securing the CDOs, as well as diversity scores, weighted average lives, recovery rates and discount rates. The determination of some of these inputs requires the use of judgment and estimates, particularly in the absence of market observable data.

Prices for the individual securities held by a CDO are obtained in most cases from the CDO collateral managers, to the extent available. For the quarter ended September 30, 2008, CDO collateral managers provided market prices for approximately XX percent of the underlying securities. When a price for an individual security is not provided by a CDO

collateral manager, AIGFP derives the price through a pricing matrix using prices from CDO collateral managers for similar securities. Substantially all of the CDO collateral managers who provided prices used dealer prices for all or part of the underlying securities, in some cases supplemented by third party pricing services.

AIGFP employs a Monte Carlo simulation to assist in quantifying the effect on the valuation of the CDOs of the unique aspects of the CDOs’ structure such as triggers that divert cash flows to the most senior part of the capital structure. The Monte Carlo simulation is used to determine whether an underlying security defaults in a given simulation scenario and, if it does, the security’s implied random default time and expected loss. This information is used to project cash flow streams and to determine the portfolio expected losses.

	8.		You state that AIG considers all available information in addition to the results of the BET model and that for each individual super senior bond, you compare the BET-driven valuation to the highest third party prices and use the lower value if they do not approximate each other. Please expand your disclosure to describe more specifically the nature and sources of these third party prices and provide a breakdown of your fair valuation that quantifies the third party and BET-driven components for each period presented.

			AIG Response:

			In response to the Staff’s comment, AIG intends to include additional disclosure in its Form 10-Q for the quarterly period ended September 30, 2008 as follows (please refer to Form 10-Q for the Quarter Ended June 30, 2008, page 49):

In the determination of fair value, AIGFP also considers collateral calls and the price estimates for the super senior CDO securities provide by third parties, including counterparties to these transactions.

The following table presents the notional amount and fair value of the multi-sector super senior credit default swap portfolio using AIGFP’s fair value methodology at September 30, 2008.


		Fair Value
		Loss at
	Notional	September 30,
(in millions)	Amount	2008

BET model	$XX	$XX
Third party price	XX	XX
Average of BET model and third party price	XX	XX
Other	XX	XX

Total	$XX	$XX

	9.		You disclose that other inputs to the BET model include diversity scores, weighted average lives, recovery rates and discount rates. Please disclose how each of these inputs are obtained, along with any key underlying assumptions.

			AIG Response:

			In response to the Staff’s comment, AIG intends to include additional disclosure in its Form 10-Q for the quarterly period ended September 30, 2008 as follows (please refer to Form 10-Q for the Quarter Ended June 30, 2008, page 49):

Diversity scores (which reflect default correlations between the underlying securities of a CDO) are obtained from CDO trustees or implied from default correlations. Weighted average lives of the underlying securities are obtained, when available, from external subscription services such as Bloomberg and Intex and if not available, AIGFP utilizes an estimate reflecting known weighted average lives. Collateral recovery rates are obtained from the multi-sector CDO recovery data of a major rating agency. AIGFP utilizes a LIBOR-based interest rate curve to derive its discount rates.

	10.		We acknowledge the new disclosure in your second quarter Form 10-Q. However, we believe that investors would benefit from an expanded description and quantification of the key assumptions underlying your roll rate analysis. Please provide this information, much of which appears on pages 10-14 of your August 8, 2008 response letter.

			AIG Response:

			In response to the Staff’s comment, AIG intends to include additional disclosure in its Form 10-Q for the quarterly period ended September 30, 2008 as follows (please refer to Form 10-Q for the Quarter Ended June 30, 2008, page 122):

Sub-prime RMBS comprise the majority of collateral securities within the multi-sector CDOs. Given adverse real estate market conditions, sub-prime mortgage losses comprise the largest percentage of AIG’s pre-tax credit impairment losses in scenarios A and B.

The roll rate analysis, as mentioned above, consists of the projection of credit losses by projecting mortgage defaults and applying loss severities to these defaults. Mortgage defaults are estimated by applying segmented roll rate frequencies to existing delinquent mortgages and by using loss timing curves to forecast future defaults from currently performing mortgages.

The roll rate default frequency assumptions for sub-prime mortgages by vintage used in the scenario A roll rate analysis are as follows:


	Pre-
Segment	2005			2005			2006			2007
30+ days delinquent		60	%		70	%		80	%		80	%
60+ days delinquent		70	%		80	%		80	%		80	%
90+ days delinquent + borrower bankruptcies		70	%		80	%		90	%		90	%
Foreclosed/REO mortgages		100	%		100	%		100	%		100	%

The sub-prime mortgage loss severity assumptions by vintage used in the scenario A roll rate analysis are as follows


Pre 2H 2004	2H 2004			1H 2005			2H 2005			2006/2007
50%		50	%		50	%		55	%		60	%

	11.		You state that the roll rate and loss severity assumptions were based on several sources, including the research of a well known investment bank and rating agency roll rate models and assumptions. Please include in your revised disclosure a more detailed explanation of the specific factors and sources that you considered in selecting each key assumption. For each of your key assumptions, quantify the impact of changes in key assumptions underlying the estimates at March 31, 2008 and June 30, 2008 (i.e. both scenarios).

			AIG Response:

			In response to the Staff’s comment, AIG intends to include additional disclosure in its Form 10-Q for the quarterly period ended September 30, 2008 as follows (please refer to Form 10-Q for the Quarter Ended June 30, 2008, page 122):

AIG based its roll rate analysis and assumptions on information from three major sources. AIG utilized actual mortgage delinquencies, defaults and losses disclosed by an industry standard source of mortgage information. AIG also incorporated projections of future mortgage collateral behavior based on roll rate transition matrices and loss severities of two different investment banks. Third, AIG incorporated roll rate and loss severity estimates from the three major rating agencies. Key assumptions include roll rate default frequency, loss severities based upon forecasts of declining home values, and loss timing curves used to forecast future defaults from currently performing mortgages.

In addition, AIG intends to include the following additional disclosure:

At March 31, 2008, AIG’s credit-based analyses estimated potential pre-tax realized credit losses at approximately $1.2 billion to $2.4 billion. The estimate of $2.4 billion was derived using the roll rate stress test described above. The increase in the estimated potential realized credit loss illustrated by Scenarios A and B was the result of both enhancements to

the model and changes in the assumptions used. The model was enhanced by inclusion of prime RMBS into the portfolio of securities subjected to the roll rate analysis, which increased the estimate by approximately $200 million, and the introduction of analytics to capture the potential effects of the cash flow waterfall, which increased the estimate by approximately $1.0 billion. Changes in assumptions included revisions to the roll rate percentages and loss severities on subprime and Alt-A mortgages in view of deteriorating real estate market conditions, as well as a higher stress to other ABS collateral and the use of current inner CDO ratings in the rating-based static percentage, which in aggregate increased the estimate by approximately $1.4 billion. The potential realized credit loss illustrated by Scenario B is the result of applying different, more highly stressed assumptions to the roll rate analysis model than those used in Scenario A. Scenario B illustrates the effect of a 20 percent increase (but not in excess of 100 percent) in all roll rate default frequency assumptions and in all loss severity assumptions across all mortgage collateral (for example, 60 percent increased to 72 percent).

	12.		You provide sensitivity disclosure for your SFAS 157 compliant fair valuation but not for your estimate of the potential realized credit losses for AIGFP’s super senior multi-sector credit default swap portfolio. Please discuss and quantify the sensitivity of this estimate to reasonably likely changes in key assumptions.

			AIG Response:

			In response to the Staff’s comment, AIG intends to include additional disclosure in its Form 10-Q for the quarterly period ended September 30, 2008 as follows (please refer to Form 10-Q for the Quarter Ended June 30, 2008, page 122):

Scenario B illustrates the effect of a 20 percent increase (but not in excess of 100 percent) in all roll rate default frequency assumptions and in all loss severity assumptions across all mortgage collateral (for example, 60 percent increased to 72 percent).

Very truly yours,

Kathleen E. Shannon
Senior Vice President, Secretary & Deputy General Counsel


cc:		Frank Wyman, Staff Accountant Carl Tartar, Accounting Branch Chief (Securities and Exchange Commission) David Herzog