First baby step in hacking out a resampling technique

I'm only half way through R. Michaud's 2008 book Efficient Asset Management: A Practical Guide to Stock Portfolio Optimization and Asset Allocation and I have printed but not yet read Bernd Sherer's paper "Portfolio Resampling: Review and Critique" 2002.  Before I go any further, to make sure I grasp what they are talking about, I thought I'd try my hand at what I gather to be the first step: showing that the efficient frontier is more subjective than it looks or at least prone to estimation error.


I don't know if I followed the "approved" methods but I at least followed my layman's common sense which I'm not sure is enough but will work to illustrate what I was thinking about.  To do this thing I took some SPY and AGG data from 2012 to today and did the following: 1) extracted the portfolio means and standard deviations for a vector of weights, 2) used asset(i)'s mean and sd to recast asset(i)'s data within the expected distribution for each i, all two of them, and then 3) repeated step 1.  While I don't know yet if this is the sanctioned way it at least looked pretty cool when charted:

Blue was the original frontier and the grey are all the frontiers in alternative universes based on the method I chose.  It is enough, I guess, to convince me that projecting expectations into the future based on one small read of the past data could be prone to quite a bit of estimation error. It's also why I have limited myself for now to using mean-variance to look backwards where things are more certain. Next step? I keep reading.