I would like to see that study saying most projects come in under budget. I'm guessing it doesn't apply to the large and complex projects we're talking about here, which are probably harder to estimate. And, the fact that it's in California may matter, too. According to Musk yesterday, without a citation:

    I don’t think we should do the high-speed 
    rail thing because it’s currently slated 
    to be roughly $70 billion but if one ratio 
    is the cost at approval time versus the 
    cost at completion time of most large 
    projects I think it’s probably going to be 
    close to $100 billion.

http://www.forbes.com/sites/hannahelliott/2013/08/12/latest-...

And, according to this:

http://californiawatch.org/dailyreport/bullet-train-bidder-h...

...the lowest (and winning) bidder for early work on the California HSR has a history of cost overruns, so things aren't starting out so hot.

Your last paragraph seems to say that we don't want accurate estimates because then we might not do the projects. That's the point of knowing the costs. We don't want to do projects that are too expensive.

Large projects, like Hyperloop or HSR, are probably the ones particularly likely to be "outliers" (if indeed projects with overruns are outliers) and big projects are the ones where that damage costs the most, so it's a double whammy.

Meanwhile, if we're now accepting that lowballing is a good idea, as your last paragraph says, then I don't understand why you seem to be defending the original article as Musk must just be lowballing and he expects overruns. It will take a lot of overruns to catch up with the ostensible California HSR budget.

I will say that, with SpaceX, Musk and the government have fixed-price contracts, so SpaceX eats cost overruns. They also have the lowest space cargo delivery cost for their class. This gives me some evidence that Musk could deliver more cheaply and on budget than the linked industry observer expects.