Scott Adams, the creator of Dilbert, draws on his past experience with budget projections to give us valuable insight into what they guys at the CBO are really doing. His two rules of budget forecasting are:
1. You must assume that trends will continue
2. Trends never continue
While this isn't as true for demographic trends that involve the aging of a large portion of the population, point number two is generally correct. He also has some valuable insight about certain components of government spending.
The budget estimates for defense spending are obviously complete nonsense too. I can't imagine that the guy who handles that part of the forecast for the CBO includes, for example, an assumption that we'll invade at least two smaller countries per decade. I think there would be a lot of pressure on that guy to remove those assumptions, no matter how right he is.
In order to be nonpartisan the CBO has to follow some pretty explicit rules, so they can only forecast what bills say and not what it likely to happen. When it comes to partisan forecasting the bias is much more obvious, as even just forecasting a higher cost for a project can get a person kicked off the team, as Lawrence Lindsey realized in 2002.