For that reason, creating a system like SecDB is a high priority these days.
In my experience there are all sorts of reasons why different desks WANT to price things differently. When 2 internal desks cross markets, they trade with each other instead of the broad market. Its advantageous for both internal desks because if the order is crossed internally, they don't have to print on an exchange, they save transaction costs, and they can be incentivized to give internal desks better prices. Its important for them to be able to shift risk from desk to desk so they can properly attribute PnL to the right agent.
edit: jrockway, just read further down that you said you work at a bank, where I'm assuming you had that experience. Ultimately the question is whether it is or should be the case that security XYZ should be considered to have the same price firm wide.
To be honest, Buffett's interest in Goldman and his friendship with Obama could have had some play in the government's interaction with them.
The real issue here, is that all the models and ratings on these securities did not correlate the failure of similar bonds with similar underlying collateral. When running the models, they had something like 10% cross correlation between different bonds with extremely similar collateral according to someone who used to work at Fitch. In reality it should have been something like 50 to 75%.
How do you guys do real-time P&L and risk calculations? Given the frequency and volume of the orders you guys pump to the market, I imagine that there must be a latency between when you receive ack/fills from market vs. real time market price of your positions. Given that HFT makes or breaks on pennies, cancel/modifying your orders in sync with real-time market info must be critical.
How do you guys solve this problem, both software-wise and telecom-wise? Via colo, but that solves the issue with only latency to trading servers. What about streaming and constructing your own consolidated quote book (because there's latency on CQS)?
For instance, a liquid stock such as MSFT is easy to value; just look at the last traded price. Chances are, if you are liquidate your entire holding of MSFT (even if you are a big fund or bank), it'll fall roughly in the range of couple of cents of last traded price.
A fairly illiquid stock such as a penny stock or a stock with relative low shares float; if you were to close your position as a bank, you are probably unloading/covering so many shares that it'll affect the price of that stock significantly. So you need to model that into your P/L model.
Things get even messier with a derivative, because let's say you are trying to sell an option position. Whoever your counterparty you sold your option contracts too, probably has some counter-strategy in which they might hedge their option transaction with an underlying equity or with another option spread at a different price whose counterparty might choose to hedge with an underlying equity position. All of which would in term affect the underlying equity pricing and how the options call/put parity is re-adjusted and then in terms, the pricing of the option contracts you just traded. So you need to model that.
So in your derivative pricing model, you might have to consider underlying pricing/volatility/liquidity, options pricing/time-decay/liquidity, futures pricing, currency exchange rates (for an ADR security) and how all of them all interact and influence one another.
And even with the math aside, constructing and consolidating all exchange/quote systems that trade all of these products is a system integration nightmare.
One more thing I should mention - this isn't a purely IT-driven story. Goldman doesn't simply have better IT than Citi or the Morgans. Citi has had many mergers and acquisitions (e.g., Vikram Pandit came onboard when Citi bought his hedge fund), whereas Goldman is mostly homegrown. Even if Goldman's geniuses ran Citi, they would still never be able to do what Goldman did.
During the financial crisis, GS knew their positions and their risks. They could also calculate the side effects of proposed trades as quickly as their computers could calculate it. This meant the people at the top could actively plan what to do next during the day.
In contrast, MS and JPM can only get information like this a few hours after the end of the day, and supposedly Citi just can't calculate such things without massive effort.