The infamous transaction at the heart of the trial of former Goldman executive Fabrice Tourre almost never happened because Paulson & Co., the hedge fund that successfully wagered that the housing market would crash, almost backed out.
Ultimately, Paulson would make more than $1 billion in profit by buying protection on the super-senior tranche of a security called Abacus 2007. But according to an email from Tourre that was shown at his trial in federal court this week, Paulson was starting to get "cold feet" in the weeks before the deal closed in May 2007.
The market for subprime residential mortgage backed securities had been rallying in the spring of 2007, seeming to recover from a sharp decline in the autumn and winter. Paulson had been placing bets against those securities, a strategy that meant he would lose money if their prices continued to climb.
Paulson's bet was already costly and risky. It required his hedge fund to make payments to sellers of protection on the bonds, which meant that the fund had to continue to spend money to keep its position in place. If, as some thought in the summer of 2007, the worst of the subprime deterioration was over, Paulson would wind up with expensive short positions that were worthless.
(Read more: The love song of Fabulous Fabrice Tourre.)
Goldman Sachs was having trouble completing the deal because its internal rules would not allow it to deal directly with ACA Management, the company that was acting as an agent in the transaction and looking to invest in it "synthetically" by selling protection on its safest tranche. Goldman required counterparties in synthetic deals to post collateral to back the obligations, something ACA would not do. ACA preferred to make deals backed by only its guarantee.
This meant that the Goldman traders working on the deal had to find what they call an "intermediary" that would stand between Goldman and ACA, posting collateral to Goldman and accepting ACA's guarantee. Goldman hoped Dutch bank ABN Amro would play the role. But putting this arrangement in place was taking longer than anticipated, and ABN Amro appeared hesitant to agree to it.
Tourre suggested to his desk that Goldman just sell protection directly to Paulson without waiting for the other part of the transaction to be put in place. As an alternative, he suggested that Goldman waive its collateral rule and deal directly with ACA. Either move would have required Goldman to take on a lot more risk.
"Would like to take down their super senior risk tomorrow in order to avoid losing the order," Tourre said in an email to the heads of his unit at Goldman.
(Read more: Bove on Goldman Sachs' top 5 scandals.)
As it turned out, of course, this wasn't necessary. Paulson's feet never got cold enough to walk away. ABN Amro agreed to its role as intermediary. ACA sold protection on the top tranche of Abacus.
The email was put into evidence by Tourre's lawyers. It suggests that Tourre didn't really believe that Abacus was "designed to fail," as he was advocating that Goldman go long.
Goldman actually did wind up with a small long position in Abacus. The protection Paulson bought and ACA sold did not match exactly. The gap was filled by Goldman itself. In the weeks after the deal closed, Goldman tried to sell this part of the deal to outside investors but found no takers at the asking price. According to one of Tourre's bosses who testified at the trial last week, Goldman ended up losing between $80 million and $90 million because it retained this risk.
—By CNBC's John Carney. Follow me on Twitter @Carney