The other day I responded to an email from an
academic/professor that I had received on a question I had sent to him where I
had asked about portfolio optimization.
In writing my response to his response, which included a paragraph on
why I was doing what I was doing, I realized it was a concise summary of my
ret-fin journey to date. Since I have
been down this road for about three or four years now I think it's reasonable to
pause, step to the side, take a breath, and see where I am. This is consistent
with the purpose of the blog which is to publicly record my journey by which
means I intend to consolidate my own understanding of what I am doing and maybe
(or maybe not) help at least one person out there that might or might not be on
the same parallel path.
GMM here refers to Geometric Mean Maximization [Estrada
2010] or sometimes: the Kelly Criterion, growth optimal portfolio, maximum
expected log, etc. It has a long and sometimes controversial history in
economic circles. MV refers to
Mean-variance optimization (MPT/Markowitz). MC stands for Monte
Carlo .
------------
"I got interested in life-cycle finance a
few years ago when going through the door of a too-early retirement made the
concept of portfolio outcomes go from abstract to a little too real. I started
by building a 5-asset-class MV optimizer -- but that's single period stuff with
no spending constraint and anyway expected return estimates seem to go stale
pretty fast. Then I built both a custom MC simulator and an amateur
backward induction engine using dynamic programming (with help) because they
had multiple-period perspective and consumption constraints. My
conclusion after all that, plus a lot of reading and modeling, was that: a)
risk allocation does matter a bit within certain ranges[1] (but
spending control is better), b) portfolios seem like they can tolerate high
risk allocations under certain conditions related to plan year and portfolio
size, and c) that satisfying the consumption constraint with some kind of
partial-to-full floor (whether using pooled risk or something else) is probably
a better idea than not. This is where GMM gets interesting. While
the volatility profile and time horizons of GMM may be ill suited to retirees
in general, once a "floor" is in place, the excess non-committed
portion of a portfolio, to the extent it is not a reserve for contingent
expenses, becomes a very long term proposition and pretty tolerant of risk and
even leverage which means it probably is a good fit for what I know of
GMM-style thinking. The excess portfolio turns, in fact, into a long-term
option on upside potential and it likely accrues more to legacy goals than
consumption. This approach, if it works, also moots a bit, I think, the
Samuelson critique of GMM. I'd like to simulate it out some day if I can make
it understandable in my own layperson terms."
------------
[1] I didn't say but my current un-proved opinion is that
the smaller the portfolio and the earlier in the plan, the more likely that the
range is something like 40-70% committed to risk assets depending on the level
of ever-changing risk-aversion. The later in the plan and the larger the portfolio
the more likely that one is perhaps at 70% or even beyond, up to and including the
possibility of 100% or even leveraged risk portfolios.
No comments:
Post a Comment