Yesterday I briefly talked about Basu & Drew’s research on glidepaths for retirement accounts. Arnott et al’s “The Glidepath Problem…and Potential Solutions” comes to similar conclusions (that glidepaths aren’t a good idea) but is different enough from Basu & Drew to be interesting.
Whereas Basu & Drew used bootstrapping, Arnott et al do historical backtesting. I think they make the same category of mistake that Basu & Drew do — of discounting rare poor returns under the assumption that a rational investor is willing to take their chances that they won’t be in that one-in-a-hundred scenario.
Arnott et al don’t even bother to show the 1-percentile case; which I think is a mistake.
They do have an interesting insight: if you look at the total dollar weighting of your portfolio across your entire life, with a glidepath you will be heavily bond-centric; possibly more than you expected to be. As a simple example:
- You are 80% equities when you are young and your portfolio is $100,000. So you have $80,000 in equities and $20,000 in bonds.
- You are 20% equities when you are old and your portfolio is $1,000,000. So you have $200,000 in equities and $800,000 in bonds.
- Overall, between the two scenarios, you have a 50/50 asset allocation when viewed traditionally.
- But if you dollar weight your portfolio, between the two scenarios, you are actually at a 25/75 asset allocation. $280,000 in equities and $820,000 in bonds.
Misspecification of Risk and Return
If they left things there it would just be rehash of Basu & Drew with a slightly different data set. But they do something that is pretty interesting: they argue that most people are using the wrong definitions of risk & return.
For most investors, the balance shouldn’t matter nearly as much as the life-long spending power that our defined contribution (DC) portfolio can sustain.
This is something that I’ve slowly come around to being a big believer in. There’s a reason why people talk about Social Security or their pension in terms of annual income: because that’s what is most important.
Arnott et al argue that just looking at final portfolio values in meaningless because when you convert that portfolio to income you suddenly start caring about interest rates (for instance, because you cash in part of your portfolio for an annuity). When you look at the two things in conjunction — portfolio values and interest rates something interesting emerges.
It is possible for your portfolio to go down but still end up with greater effective wealth — if interest rates go up enough. Similarly, your portfolio could grow, but if interest rates go down enough, you’ll still feel a reduction in effective wealth.
(I think that something like this has happened right now. Portfolios have grown but interest rates are so low that annuities or people relying bond interest payments feel anxious about their standard of living.)
With that perspective, Arnott et al ignore portfolio volatility and instead focus on the volatility of real spending power the portfolio could buy when you retire — which adds interest rates into the equation.
They call this their Real Annuity Framework.
When you add interest rates into the equation, it is no longer clear that simply shifting into more bonds is actually that safe. Arnott et al argue that there are better approaches than a glidepath to manage the real risks.
Instead of lowering risk by shifting to bonds, especially when bond yields are plumbing near-record lows, and exposing our clients to some very dangerous risks if they revert to historical norms, we can rein in the risks that matter (duration risk and beta) without sacrificing return.
I think the current environment shows some of this in action. Would you have felt confident having 80% of your portfolio in bonds over the past week? Or would you be one of the people saying “Why are bonds dropping, they are supposed to be safe”?
I really like their approach of marrying up portfolio values with current interest rates to put the focus back on income in retirement. Unfortunately, getting historical annuity prices seems to be problematic — I’m not aware of any free sources.
