A combination of wasteful architecture astronomy and legitimate need to divvy up absolutely mammoth line-of-business applications among teams with hundreds of members operating for years with wildly varying skill levels, often on different subsystems from different physical locations, with billions of dollars on the line. You can't let the least senior programmer in the Melbourne office bring down the bank if he screws up with something, so instead you make it virtually impossible for him to touch anything than the single DAO which he is assigned to, and you can conclusively prove that changing that only affects the operation of the one report for the admin screen of a tier 3 analyst in the risk management group for trans-Pacific shipping insurance policies sold to US customers.

The tradeoff is, historically, that you're going to have to hire a team of seven developers, three business analysts, and a project manager to do what is, honestly speaking, two decent engineers worth of work if they were working in e.g. Rails. This is a worthwhile tradeoff for many enterprises, as they care about risk much, much, much more than the salary bill.

(I spent several years in the Big Freaking Enterprise Java Web Applications salt mines. These days I generally work in Rails, and vastly prefer it for aesthetic and productivity reasons, but I'm at least intellectually capable of appreciating the advantages that the Java stack is sold as bringing to users. You can certainly ship enterprise apps in Rails, too, but "the enterprise" has a process which works for shipping Java apps and applying the same development methodology to e.g. a Rails app would result in a highly effective gatling gun for shooting oneself in the foot.)

I was briefly exposed to this doctrine at school, so I hope dropping the following buzzwords help: POJO, Java Persistent API, Java Enterprise Beans, Gang of Four Design Patterns