A 3M FX swap is normally traded under ISDA/CSA. Hence, the credit risk of swap can be assumed to be low. However, there has been a lot of focus on deviations from covered interest rate parity, especially with FX swaps against USD. In some papers they attribute the deviation, at least in part, to heightened counterparty risk with respect to non US banks. Usually CDS premias are used to motivate the assumption that some banks have higher counterparty risk than others. Given that FX swaps are collateralised and traded under CSA how valid is this assumption? In know that CSA are not perfect and so you still have some residual risk but still would the CSA not make the counterparty risk argument in large parts invalid? I know the same goes for repo and there you saw that some banks (eg Bear Stearns) could not roll over their repos even with US treasuries as collateral. Would be interesting to hear some thoughts on this!

I think it's to do with the physical cash - the FX Fwd market was one ways banks could get hold of *physical* dollars. It doesn't matter whether the credit risk is negated or not, because these banks needed the Dollars - so supply/demand factors.Additionally, because the principals are exchanged there is significant market risk on these trades if your counterparty defaults - as an example, imagine you entered a EUR1bn 3m swap where you lend USD and borrow EUR against it, and the initial exchange of principal occured. Now imagine your counterparty goes bust: you now have a spot FX exposure of long EUR1bn (in addition to losing the interest differential on the far leg) which will cost you money to get out of in terms of bid/ask spread, but even if you have an ISDA/CSA, overnight (ie since the last collateral payment) the FX market could still have moved a percent or two causing you huge potential losses on that exposure (e.g. a 1% fall in EURUSD would cost you $10m). So there is probably some optionality priced into the basis spread given that the predominant flows were towards borrowing of USD. I've not tried tested that hypothesis, however...