Exponential growth will ultimately trump all other linear factors, even if the exponent is small. And that's also what makes small differences in that exponent so important (hence why computer scientists go through great pains to reduce complexity of matrix operations from O(2^2.51) to O(2^2.36)). Even if the investment only earned 1% per year, he can just sock it away in a 100-year trust fund:
>>> 1000 * (1.01 ** 100)
2704.8138294215287
>>> 523 * (1.005 ** 100) * .45
387.54342961962465
BAM - 85% "marginal" tax rate.Plus, he has deliberately framed his scenario to take the maximum hit from taxes at every step. Corporate tax eats into his dividends? OK - invest in a growth stock instead of an income stock, and pay capital gains taxes when realizing the profits instead. Estate taxes steal most of the money from his heirs? How about he write a check to the university on behalf of his grandkids -- Uncle Sam won't see a dime of that.
A less comprehensible explanation is available from the IRS here: http://www.irs.gov/businesses/small/article/0,,id=108139,00....
But I think you probably mean mikeryan. I don't think he was talking about a marginal dollar. His point was that the first $12k ($13k in 2010) gifted are taxed differently than dollars beyond that threshold.
I suppose the real question is the relative importance of the behavior at the margin versus the behavior for the first dollar. For the ultra-rich, $13k is obviously irrelevant, but for someone near the $250k annual income threshold like Mankiw, I don't think it is. The present value of $12k per year given every year for 30 years is substantially larger than the same amount invested and then hit by estate tax at year 30.
Not a good way to rid yourself of millions, obviously.