Stochastic Retirement Start Age - Blanchett on Retirement Starts

I feel partially vindicated.  David Blanchett has a 2018 paper (Blanchett, David M. 2018. “The Impact of Retirement Age Uncertainty on Retirement Outcomes.” Journal of Financial Planning 31 (9): 36–45) that finally broaches a point I made as recently as 2017 -- but predates my blog at least back to 2015 when I was trying to post stuff on LinkedIn -- that retirement start age is underappreciated as a random variable. Here are some excerpts from Blanchett:

• Research suggests people tend to retire earlier than expected. Retiring early can have a significant (negative) impact on a retiree’s likelihood of achieving retirement success.

• A nonlinear relation exists between actual and expected retirement age, where individuals targeting retirement before age 61 tend to retire later than expected, and those targeting retirement at an age after 61 generally retire approximately a half-year early for each additional year targeted past age 61

• Incorporating retirement age uncertainty into a financial plan can have a significant impact on required retirement savings levels. 

Me? I tend to focus on a world where one is already retired which makes a paper like this a wee bit moot.  But if one were to still be accumulating then it is more interesting.  The conclusion here in Blanchett seems banal but that's only after having seen it when it then seems obvious. But it is less than boring if you really think about it carefully.  It means one probably needs to save at a much faster rate when saving than one might otherwise save in order to accommodate the uncertain start and in order to keep retirement lifestyle expectations more or less constant at some random start age.  Or, I guess, by some kind of obvious inference, one might have to accept a random lifestyle level on retirement start, if one were to keep the savings rate constant after one realizes that the start date might be random.

Glad to see the paper though. It continues to build up what I want to call the stages of retirement finance modelling.  To simplify, the stages might look like this where stage 1 is Bengen except using historical data rather than constant returns, stage 2 predates stage 1 if we are honest about Yaari's contribution, stage 3 is what seems like the current state of most Monte Carlo simulation, and Stage 5 is what the Blanchett paper seems like it is stepping into. With your indulgence, I'll visualize it something like this:

The conclusions from this type of stage 5 incremental movement are not hyper momentous as the bullets above should make clear.  The net effect is that it is nothing other than adding uncertainty to the length of the horizon and the start date.  The implications are, short of winning lotteries, also not hyper-momentous: (1) save more faster, and (2) spend less in retirement, at least initially anyway, in order to hedge the longer, uncertain horizon.  There is a story about buying lifetime income in there somewhere but not in this post.

In 2017, I visualized the start-and-end randomness like this below.  The left distribution (random retirement start age due to layoffs, personal choice, health problems, care-giving demands, etc) I actually pulled from somewhere credible but can't find my source. The right is a mortality PDF for a 60 year old using conservative SOA data. This illustrates the Blanchett proposition well I think.




You can see here the traditional "30-year horizon from age 65" as well as some "other" horizon that is obviously a policy/planning choice since it depends on how you want to think about it.  Mode to mode? 5th percentile to 95th? Mean to mean? 95th to 95th? You decide.  Maybe the least we can say is that 80 years, with a start age of, say, 30 (30?!?) is an upper/early bound for a planning horizon. Maybe.


-------------------------------

Don't forget spending is a random process as well. And not just random. Random + chaotic. Just adds to the fun.