“It’s Still an Evolving Body of Research”

An earlier blog entry described the background of my recent correspondence with Michael Kitces on safe withdrawal rates (SWRs). Set forth below is the text of an e-mail that Michael sent to me on August 24 .

Rob,

I don’t mean to understate the impact, but I think the essence of your disagreement with the Old School SWR research really boils down to nothing more than the implicit assumption of the original SWR research that “by definition” the worst case scenario that has ever occurred in history IS the worst case scenario. Obviously, you believe differently – with what I believe is a reasonable basis for doing so – particularly to the extent that valuations since the first SWR research have left the range of the original SWR research history (i.e., the valuations at the start of the decade were beyond anything measured in any mid-1990’s historical SWR research).

However, I don’t entirely agree with your analogy example. It is true that the original SWR research looked at ALL scenarios, regardless of valuation. But I don’t believe that, in and of itself, is necessarily a flaw. If you look at every case scenario in history – all the good ones, and all the bad ones – and pick the absolute worst, by definition you have already included all of the worst valuation periods in history. This is supported not only theoretically, but in practice – the worst SWR in history, using the Old School approach, correctly identified that the worst case actual historical scenario really WAS 1966-1995, and as WE know it turned out this way BECAUSE that was in fact the secular valuation peak of that era. By your drunk driver example below, the comparable analogy of the old school research is to say that because there was at least one instance ever where a drunk driver had an accident, the conclusion is to NEVER drive AT ALL, because it might turn out that you’re drunk and that an accident may result. Likewise, the old school SWR research indicated that because anything more than 4% might have failed once in history, you should never take anything above 4%, and therefore you will never fail under any historical scenario. By definition, that already incorporates EVERY bad valuation scenario (because the research looked at all of them).

The “flaw” of the research, in essence, is the implicit assumption that a withdrawal that survives EVERY historical scenario (REGARDLESS of valuation, good or bad) must be safe in any future scenario. It assumes that nothing in the future could ever be (materially) worse than the historical record. Ironically, Bengen did his first research on this in 1993, and it only took 7 years for us to produce a dramatically worse valuation environment, that ultimately probably WILL demonstrate a “new low” in required safe withdrawal rates.

But for better or for worse, the fault of the research is simply the assumption that “any withdrawal rate which survives every historical scenario will survive every future one.” Now of course, I don’t believe any prudent practitioner quite delivers the advice that way – there’s always an *asterisk footnote that indicates the future might really produce a scenario worse than any historical scenario – but the progress of John Walter Russell’s research is to provide a better framework about HOW those failures might happen and WHEN they might happen. At the end of the day, Russell theorizes that there was only a 1-in-3 chance of actually surviving the 1966 retirement at 4% [Editor’s Note — I believe that Michael meant for this to be a reference to retirements beginning in January 2000; a 4 percent withdrawal had a 2-in-3 chance of surviving for retirements beginning in the mid-1960s, according to Russell’s research]. That’s ostensibly because his models deliver a wider range of potential results than “just” what happened in history. Realistically, either he’s right, OR his model is wrong.

Obviously, I’ll grant that I think there’s a lot to the approach he utilizes, and I don’t mean to claim that I think his model is invalid. But the reality is that his model is a THEORY of what other alternatives might have happened, and likewise what might happen in the future. The 4% SWR still has never ACTUALLY failed any of the historical scenarios we can produce using the real data, which by virtue of being what really happened automatically accounts for all the economic complexities that Russell’s models MIGHT be failing to fully incorporate (e.g., to what extent are the correlations, cross correlations, serial correlations, cross-serial correlations, etc. REALLY fully modeled, and to what extent do simplifying modeling assumptions create a model that deviates from reality?). 

Again, I agree with the approach that Russell is exploring, PARTICULARLY to the extent of valuations that are larger than anything our historical SWR research has ever been able to observe in full. And I think it’s a valid criticism to at least question whether the 4% SWR approach might be pushing it a little too tightly at simply any valuation point that happens to coincide with a similar historical high valuation point (which we still see at today’s valuation levels). But I still feel compelled to show some respect to research implying a safe withdrawal rate that really WOULD have genuinely survived every historical scenario our markets have ever actually followed.

But that’s why it’s still an evolving body of research. 🙂

On a sidenote, I’m moderating a panel session at the NAPFA Northeast Regional conference in Hershey, PA, this November – and the topic is entirely about safe withdrawal rates, and my panelists will be Bill Bengen and Jon Guyton (arguably THE two leading researchers on SWRs from the financial planning community). One of my goals for the session is to invite both of these individuals – who are both researchers and planning practitioners who work with clients – about how their views may have changed over the past 8 years of market history and in their work with clients. I’m very curious to see how the session goes! 

With warm regards,
Michael