Nick wrote (responding to Christer)
Your example of prices is moderately spectacular in that
most coefficients are rather large. At a wild guess most
or all of your covariates are dummy variables such as presence
or absence of garage and all of
your house prices are large in the currency you are using.
In practice, scaling your units by (e.g.) dividing prices
by 1000 or an even larger number is likely to be your best
route forward. In other words, tackle this problem upstream
so that the results end up smaller and format is no longer
an issue.
An alternative is to look at saved results.
I would suggest the alternative of plugging in the X values at which
you want to generate predictions as additional observations in the
data set. Run the regression over the true sample, and predict xb
over the entire sample. The result is then stored in a variable, and
you have complete control over its display format.
