Well, as somebody who is currently coding a calculation service that will handle a 500 billions portfolio, I can tell you this: the specs require the data to be interpolated + extrapolated left and right. And they use that to estimate risks and take huge decisions. And when they don't have enough data to interpolate from, they create a synthetic instrument from other instruments that have enough data, and use that to estimate risks and take decisions.
The only thing that makes the whole jenga tower stable is the fact it's all interdependent. E.G: to diminish your exposure, you take a swap with some other entity, and if everybody do that enough, the risk is averaged on the entire economy.
So from the inside, it's worse than it looks from the outside.
And mind you, the client I do that for is one of the most serious in the business. They are extremely risk adverse, and they have been acting nothing but honestly, at least at the level I can observe.
They are so by-the-book we have once found an error in $FAMOUS_DATA_PROVIDER calculations, because we thought we made a mistake, and we looked for it until we realized the data we compared ourselves to was wrong.
So if they manage half a trillion by using excel FORECAST.LINEAR and Matlab interp, imagine what is done by JP Morgan.