As threatened, I performed some simulations with a spreadsheet. I used BEGBX (American Century international bond fund) and VBMFX, with price and dividend data from Yahoo going back to April 92. From the monthly returns for each, I subtract the risk-free rate of return, represented by one-month treasuries, which I think is the proper thing to do. I posted the spreadsheet at

http://j.mp/eyFFLp
The results are interesting, I suppose.

Whether rebalancing monthly or yearly (in the latter case each April 1), moving the bond portfolio from 100% to 99% domestic

increases standard deviation. Even though domestic and foreign are only correlated 0.48, the latter are so much more volatile that this overrides the diversification benefit of even a tiny bit of foreign bond.

It happens that over the April 1992-December 2010 period domestic beat foreign by a cumulative 75.26% vs. 68.43% (both figures relative to the risk-free return rate). So as one shifts to foreign, one's return goes down. I put less weight on that, as a predictor of future patterns, than the results just mentioned regarding s.d.

The spreadsheet includes a scatter plot of BEGBX - r against VBMFX - r, where r is the risk-free rate of return. BEGBX emerges as far more volatile, yet broadly correlated with VBMFX.

So, no super-strong argument for foreign bonds here.

On the other hand, I did the same thing for stocks, and the empirical argument for foreign stocks didn't seem that much stronger. It's also in the spreadsheet.

Here, I used VTSMX and VGTSX, starting in July 1996. The statistical relationship between domestic and foreign is quite different. The correlation is 0.86 instead of 0.52, meaning much less potential for diversification benefit. S.d. is minimized at 82% domestic, 18% foreign. But the reduction in s.d. compared to 100% domestic is tiny: 4.89% instead of 4.91%.

Even more so than with bonds, domestic beats foreign over the period with data--64.31% vs. 36.66% cumulative, net of the risk-free return.

So what surprises me about this is that I didn't find a particularly strong case for foreign stocks *or* foreign bonds. Maybe my data periods are still too short?