Last year, a piece by Michael Kitces and Wade Pfau made the claim that mechanically increasing the equity allocation during retirement — which they term a “rising glidepath” — could reduce the likelihood that a retiree outlives his or her assets and could decrease the magnitude of shortfall when capital market returns disappoint. Specifically, the paper stated:
“We find, surprisingly, that rising equity glidepaths in retirement … have the potential to actually reduce both the probability of failure and the magnitude of failure for client portfolios.”
I initially suspected the basic claims might not hold up under analytic scrutiny and recently got the chance to dig into their data tables to find out. In my analysis, I focus on the success rates and 5th percentile shortfall results they report in Tables 2, 4 and 6 in their paper. These three tables contain results for 4 percent withdrawal rates under three different sets of capital market assumptions.
Interpreting Differences in Success Rates
When you run any form of retirement-planning simulation analysis, a crucial aspect of interpreting differences in success rates — which are simply the proportion of cases where the retiree didn’t run out of money — is determining whether the differences are truly material or just noise. For example, if you run such an analysis and find that one strategy had a success rate of 80 percent while another had a success rate of 81 percent, you likely shouldn’t conclude that the second strategy is better because the results are basically identical. In fact, you might get a completely different result if you ran the same simulation with slight changes in your capital market assumptions. In other words, you should conclude that your results aren’t robust since they are highly dependent on your capital market assumptions.
If you like baseball analogies, you wouldn’t claim that a player with a 2014 batting average of .300 is a definitively better hitter than another player with a batting average of .299 because the difference is so small. It could be that the first player hit a couple of “seeing eye” singles (i.e., he got lucky) that explain the slight difference in the batting averages.
In the table below, I use data taken directly from the Kitces and Pfau paper to report the average success rates for the rising and declining equity glidepaths and the difference in these success rates.
I’ll summarize the main takeaway from the first table: There’s nothing here demonstrating robust superiority of the rising glidepath approach compared with the declining glidepath approach.
In one case (Capital Market Assumptions II), the rising glidepath is markedly inferior to the declining glidepath. In the other two cases, the rising glidepath success rate is basically identical to the declining glidepath result in one (Capital Market Assumptions I), and the rising glidepath rate is modestly better in the other (Capital Market Assumptions III). In looking at the aggregate of the three sets of results, it’s impossible to conclude the rising glidepath produces reliably better success rates than the declining glidepath. The data simply does not support that conclusion. Instead, it shows that the “superiority” of one approach relative to the other is highly sensitive to the capital market assumptions used. (I also did the same calculations looking only at cases in which the equity allocation was never more than 70 percent or less than 20 percent, and the conclusions were basically the same.)
Evaluating Magnitude of Failure Results
Even though the success rates don’t reveal material differences between the rising and declining glidepath approaches, it could be that the rising glidepath is better when comparing magnitude of failure results. At first glance, that indeed appears to be the case. However, a deeper analysis of these results shows that the rising glidepath isn’t superior here, either.
In this table, the more negative the dollar figure associated with a 5th percentile outcome, the worse the result is for the retiree since a negative number indicates that spending exceeded the value of the portfolio (i.e., the retiree ran out of money before the end of the analysis). The numbers effectively are a measure of how badly the plan failed when capital market returns were much lower than expected (or when bad results occurred early in retirement). In all three cases, t-stats exceed +2.0 by a significant margin, which indicates that the rising glidepath appears to be superior to the declining glidepath approach.
Instead, though, these results seem to reflect the fact that the rising glidepath simulations tend to start with much lower equity allocations than the declining glidepath simulations. This fact alone — and not whether the simulation implemented a rising or declining glidepath approach — seems to account for the vast majority of the differences we see above.
For each set of capital market assumptions, Kitces and Pfau ran 55 simulations focusing on rising glidepath approaches and 55 simulations focusing on declining glidepath approaches. Here’s the key point: The average starting equity allocation for the 55 rising glidepath approaches was 30 percent while the corresponding number for the 55 declining glidepath approaches was a 70 percent allocation. Holding all else equal, it’s well known that higher equity allocations in retirement tend to fail more spectacularly than lower equity allocations since equity returns are much more volatile than fixed income returns. So, to determine whether the rising glidepath truly performs better than the declining glidepath, any analysis has to control for differences in starting equity allocation before any conclusions can be drawn.
To do that, I set up a relatively straightforward regression analysis. I regressed each of the 110 5th percentile outcomes (remember, there were 55 simulations for both rising and declining glidepath simulations for each set of capital market assumptions) on the starting equity allocation associated with a particular simulation and a “dummy” variable that takes on the value of “1” when the simulation was associated with a rising glidepath approach and the value of “0” when associated with a declining glidepath approach. If the improvement in 5th percentile results is directly attributable to the rising glidepath approach, the coefficient on the dummy variable should be positive. Why? Because a positive coefficient would indicate that the 5th percentile results were better for the rising glidepath approach even after controlling for starting equity allocation. The next table reports the dummy variable coefficients for each of these three regressions.
In two of the three cases, the point estimate for the coefficient was negative, indicating the results were worse for the rising glidepath approach after controlling for starting equity allocation. The third case was positive but statistically insignificant. As we saw with the success rate analysis, nothing in these results indicates the rising glidepath approach is superior to the declining glidepath approach. The 5th percentile results seem to be driven more by whether the starting equity allocation was high or low and not whether a rising or declining glidepath approach was implemented. Said differently, my analysis indicates that a rising glidepath approach that starts with a low equity allocation that increases over time is basically the same as a declining glidepath approach that starts with a slightly higher equity allocation that decreases over time.
I have a lot of respect for the contributions both Kitces and Pfau have made (and continue to make) to advisory practice, but the rising glidepath claims simply don’t hold up under analytical scrutiny. The success rate results are on average close to identical when comparing the rising and declining glidepath approaches across the three different sets of capital market assumptions, and the 5th percentile results seem to be driven more by starting equity allocation than whether the equity allocation was rising or declining over time.