Some of it comes from paper documents and PDFs. Bloomberg has a team of people who organize and cross-reference such documents.

If you're wondering why the original documents are not more directly machine readable, consider that some of them are published by companies due to regulatory obligations, and those companies would much prefer not to publish the information at all. So making the documents easier to consume is where Bloomberg comes in--because the originators don't want to.

The main part is not purely interface but the aggregation of data - integration with thousands of various systems, licencing agreements with thousands of proprietary data sources, automated and manual data cleanup. This isn't scalable and simply requires thousands of people and lots of work, because obtaining ocean tanker data doesn't really help you get integration for e.g. Nigerian mining industry data.

Also, error scrubbing.

There's lots of financial data available on yahoo, for example, but it's just lousy quality.

Can you give an example? Not that I don't believe you.

Ok, look at the Thai Baht exchange rate:

http://finance.yahoo.com/quote/THB=X/?p=THB=X

You see the graph on the right, it hovers around 30 USD/THB (that's 30 THB per USD, in somewhat confusing FX convention).

Now click on 5Y, then click on 10Y, and compare. The 10Y graphs probably includes some wrong blip around 700, which is why the 10Y graph scale is from 0 to 750, and you just see a basically flat line displayed at the bottom.

(The 10Y PHP graph is fine, while the 10Y INR graph has a similar issue).

So, if you do any trading strategy where you're trying to figure out whether your current value is relatively high or low by comparing it to the 10Y max and min, you'll most likely be off.

Or consider equity, for example. To properly deal with shares, you first need the trading calendar for every exchange (when is the exchange open and you can make a trade), and the settlement calendar (when will the shares actually be settled). US settlement is typically T+3, Europe T+2, UK Gilts T+1, but all according to certain calendars, naturally. These calendars vary by country, and even exchange and even function (in other words, there might be days when shares at a certain exchange trade, but don't settle, and vice versa).

Bloomberg has all these details - good luck finding it on Yahoo or anywhere else, really.

Thanks for the example. What you just did is one of the hardest things to communicate about data. A casual user would never notice this. I find "data curation" doesn't always get the importance it deserves.

Oh, man, tell me about it!

To get from raw data to data you can use can be a hard, tedious and difficult slog. I agree that it's often undervalued. On the other hand, with Bloomberg you certainly pay for it! :-)

And don't get me started with corporate actions (stock splits, mergers), dividends (some jurisdictions announce gross divs, Australia announces net divs; Korea announces AFTER the ex-date; some shares pay a dividend in a ccy different from the ccy the shares are denominated in, etc.), unscheduled exchange closures (due to terrorism, typhoons, etc.), stamp duty and taxes, 1 week holidays for Chinese New Year that screw up the rate curve (because 1d and 1w LIBOR cover the same period), quanto futures, etc.

And time zones. And daylight savings. Oh my oh my. And Asian trading sessions with a lunch break in the middle.

Haha, I loved the HK trading day with a 2h lunch break!

2.5 hours in the morning (10:00 to 12:30), 1.5 hours in the afternoon (14:30 to 16:00), who needs more than 4 hours trading a day?

Shame they extended it in 2011.

Yes, data cleanup and maintenance is why proprietary data vendors still survive.

As others have mentioned there is a large amount of expert knowledge to collecting and organizing extremely disparate data sets. There are also longstanding business relationships, lots across a lot of markets, and a reputation for great data security, that has many pricing sources and exchanges offering data almost exclusively through the terminal, and some folks putting their own data into the system. The risk with mishandling(making unavailable or available at unreliable/inconsistent times, for example) some of that data is just so high that 30+ years of reputation in some spaces is worth a (maybe perceived) premium.

Some of it is fed in directly by the providers. For example Barclays and JPMorgan interest rate swap rates.

There is a network effect: all the customers the dealers want to reach are on Bloomberg, so they have to publish on Bloomberg. If someone else comes along and says "publish on my channel too", that's extra work, so probably won't get done.