Feb. 15 (Bloomberg) -- It doesn't seem to matter who's running American International Group Inc. Three years after an accounting scandal cost Hank Greenberg his job as head of the world's largest insurance company, AIG still can't get its numbers right.

In a regulatory filing Monday, AIG said the value of its portfolio of ``super senior'' credit-default swaps fell by $4.88 billion in October and November, four times more than previously disclosed. Its explanation for this revision was a model of obfuscation, and something about it doesn't ring true.

Not once in its seven-paragraph narrative discussion did AIG say how or why the gross losses on these derivatives had soared after Sept. 30. Rather, in almost-unintelligible jargon, it told how in November it developed a new ability to value some large offsetting gains -- and how it then decided this month to exclude most of them because the data weren't reliable.

That was news to investors, who sent AIG's shares plunging.

In a Dec. 7 filing, AIG had estimated its year-to-date loss on these swaps as of Nov. 30 was no more than about $1.5 billion, the bulk of which it said had occurred after September. What it didn't say then was that this number included $4.36 billion of offsets, $3.63 billion of which it has now disavowed; nor did it disclose the size of the gross losses.

Fortunately, AIG's disclosure Monday included a chart showing the hard numbers, the footnotes to which tell a lot.

The short of it: AIG was using a rickety valuation model for these swaps in September. When the company got better data in October, its gross losses grew significantly. And when it got even better data in November, the gross losses skyrocketed.

Plunging Values

AIG sold these swaps to investors who used them as protection against losses on various collateralized-debt obligations, or CDOs, which have become notoriously illiquid and difficult to value.

Much of the post-September losses undoubtedly resulted from plunging CDO markets. Still, based on what the footnotes show, it's hard to believe AIG got it correct back in November when it first reported that its year-to-date loss on these swaps at Sept. 30 was just $352 million.

AIG says it stands by this figure. Yet it won't say what the number would have been using its revised valuation approach.

``It would be an interesting exercise,'' AIG spokesman Chris Winans says. ``There's no reason to do that, because there's no indication of an error.''

Generic Brand

Actually, there may be. In calculating the $352 million gross loss at Sept. 30, AIG said it used inputs consisting of ``generic credit spreads on asset-backed securities provided by a third party'' that it didn't name. Generic, in this instance, seems to mean not very good, because AIG felt compelled to augment this approach the next month with more reliable, market- based, data inputs.

AIG showed an $899 million year-to-date gross loss on the swaps at Oct. 31. The footnote to that number said the company started with the generic data, and then adjusted it ``using inputs derived by management from observed changes in the relevant ABX indices.'' Investors use these indexes to bet on subprime-bond defaults.

Then November came, and AIG abandoned the generic data. This time, it reached out to real human beings who monitor cash prices on these sorts of things for a living. Specifically, AIG said it used ``cash bond prices provided by the managers of the underlying CDO collateral pools, or, where not provided by the managers, prices derived from a price matrix based on cash bond prices that were provided.''

Using this information, AIG showed a year-to-date gross loss on the swaps portfolio of $5.96 billion at Nov. 30.

No Evidence

How do we know the change in approach made a difference? In a footnote, AIG said the gross loss at Nov. 30 would have been just $2.55 billion, or 57 percent less, had the company used the same types of inputs that it used to calculate the Oct. 31 figure.

AIG didn't disclose what the Sept. 30 or Oct. 31 gross-loss figures would have been using the November approach. Perhaps it's true that the results would have been the same. However, AIG has provided investors no evidence to believe that's the case.

Instead, we're left with a picture of a company that changed its valuation model twice in two months in search of better data. Then, when the losses came in huge at Nov. 30, AIG acquired a newfound capability to measure some large offsetting gains, the size of which it didn't initially disclose.

AIG's outside auditor, PricewaterhouseCoopers LLP, put a stop to that. That's why the company's year-end results will not include the offsets. The accounting firm also concluded that, at Dec. 31, AIG had a ``material weakness'' in its internal control over financial reporting related to the fair values for these swaps.

AIG said it hadn't yet determined the portfolio's value as of Dec. 31. It's a sure thing the losses have grown. And when AIG reports its year-end results, scheduled for this month, investors will need clear explanations of how and why.

The fact they didn't get them this time can only fuel suspicions that AIG has something to hide.

(Jonathan Weil is a Bloomberg News columnist. The opinions expressed are his own.)

To contact the writer of this column: Jonathan Weil in Boulder, Colorado, at jweil6@bloomberg.net

Last Updated: February 15, 2008 00:44 EST