Kitces on Simulation

In a recent post on Monte Carlo simulation on Michael Kitces site (Monte Carlo Investment Assumptions In Your Retirement Planning Projections) he describes some of the counter-intuitive effects of adding more and more asset classes to simulation (from a planning perspective one can inadvertently understate risk too much) and the importance of understanding asset class correlation. That was fine and interesting and important and useful and maybe even a little obvious if one has worked with simulators.  That was not what caught my eye, though. It was a comment later in the article on another topic:

But unfortunately, there’s no Monte Carlo software that can actually show that, what I like to call regime-based retirement projections. Where we’re in a low return regime for a decade and then we normalize. It’s possible mathematically to do it, the software just doesn’t do it now. Which means, the only alternative is to haircut long-term returns, which is what we actually do in practice. We reduce long-term returns by about 1% or 100 basis points, recognizing some of the risk of the low return environment.

Well...maybe not in his software.

Mine does it, though. It was one of the first things I added to the first round of enhancements.  While doing an exact real world representation of this is tricky, if one is willing to make some simplification trade-offs it is a relatively trivial thing to add.  And add I did because I, like a lot of people, think that there is at least some non zero risk of horrible returns for some number of years.  The impact is obvious: fail rates go up. The impact on planning, though (what exactly is one supposed to do if a projected fail rate goes from 14% to 27%, say), is less obvious.  My opinion is that the planning change is mainly to be more conservative in spending programs and maybe to consider adding to an asset allocation some non-correlative (hopefully non-correlative, anyway) evidence-based (of course) risk premia (which ones?) other than credit and equity risk.  The other question is how big of an impact is there?

While my approach (suppression of overall portfolio returns for x years by y%) is overly simplified (by quite a bit), blunt, reductive, and eminently debatable it does the basic job of adding risk for a period or "regime." Here was a quick attempt at running some numbers. This is for a standard assemblage of assumptions like 

$1M endowment

50/50 alloc.

fee and tax effects

no SS/annuity/pension

age 60-95 fixed longevity

3% constant spend

No spend trends, inflections, shocks, variance

4k sims (short)

other

Then applied twice: once for 10 years of return suppression with different suppress amounts and a second time with a fixed suppress amt (2%) for different "regime" durations. 

1.)  10 years duration, vary suppression rate

rate      fail%   fail dur.

.00       .146     6

.01       .197     7

.02       .276     7

.03       .361     8                     

2.)  .02 suppression, vary years duration

years    fail%   fail dur.

0          .146     6*

5          .212     7

10        .276     7*

15        .325     8

20        .376     8

So, there you go...