Busy week at the blog!
On Monday, I reported on a Scott Burns column about a March 2012 Journal of Financial Planning article I co-authored with Michael Finke and Duncan Williams. Our essential argument was that when a retiree properly considers their available resources outside of their financial portfolio (such as Social Security, fixed income annuities (SPIAs), pensions, etc.) as well as their flexibility to reduce spending in their later years, it may make sense to choose a retirement income strategy with a much higher failure rate than is commonly considered in safe withdrawal rate studies. That study was based on Monte Carlo simulations calibrated to the historical data.
On Wednesday, I updated my explanations about a paper I published in the August 2011 Journal of Financial Planning in which I argued that historical data does not provide a proper basis for determining forward-looking safe withdrawal rates. Future returns and sustainable withdrawal rates are connected to the current market environment, not historical averages. Some great comments followed that post, including a question about why we used historical averages in the March 2012 article, given what I reported in August 2011. As well, one commenter made a good point which ultimately means that the out-of-sample predictions connected to low recent dividend yields may not be reliable.
About the issue of why we used historical averages in March 2012, my basic view is that there are many changes I would like to consider for retirement income studies, but not all the changes should be made at the same time. Rather, I think it is more helpful to change assumptions one at a time so that it is easier to see the impacts of each of those assumptions. We should certainly redo the March 2012 with alternative returns assumptions at some point though.
Today I would like to make some initial efforts toward this end. In doing so, I harken back to my January 2012 Journal of Financial Planning article, "Capital Market Expectations, Asset Allocation, and Safe Withdrawal Rates." The whole point of that article was to provide a framework for seeing how any types of assumptions about returns and asset classes choices impact the results for safe withdrawal rates. I think this article provides a framework for re-considering the August 2011 article framework too, though admittedly it is not super-straightforward about how to do this since it is very hard to know about proper assumptions for the means, standard deviations, and correlations for each asset class.
Nonetheless, in moving toward today's post, Joe Tomlinson's new column at Advisor Perspectives this week is also excellent (it compares GLWBs and deferred income annuities). While I'm not talking now about his results, I do want to adapt his general data assumptions methodology. He is thinking about things in the right way, and I want to show what happens when you apply his approach to the framework from my January 2012 article.
Here are three tables of data assumptions which will provide the general summary of the impacts from changing assumptions. Again, though I am not following Joe exactly, I am adapting his ideas here. First, Table 1A shows the historical data parameters which I use when I say I am calibrating things to the historical data. This is real data after adjusting for inflation. Then, Table 1B keeps the same historical equity premium over bonds, but it adjusts all of the returns downward by 1.52%. This is to get the real bond return to more closely match current real bond yields. Current yields are the best predictor for longer-term bond returns, even if not held to maturity. Finally, in Table 1C, I maintain the lower bond returns and also adjust the stock return downward as well to assume a lower equity premium than historically witnessed. The equity premium in 1C is only 4% instead of 6.18%. I don't try to mess with standard deviations or correlations.
Next, it is just a matter of seeing what happens when using these alternative sets of assumptions. The next table is the same Table from my January article showing results for the historical data parameters seen in Table 1A. As I explained in a recent blog entry, I think this table provides a lot of nice advantages for thinking about safe withdrawal rates in the classical safe withdrawal rate framework introduced by William Bengen and the Trinity study. [Though, the March 2012 article points in a whole different direction for thinking about safe withdrawal rates]. Here is the table with advice calibrated to the historical data:
You can study it, but to keep the discussion manageable I will focus on one specific row from the table to describe differences for the varying assumptions. Consider a 30-year retirement duration for a retiree willing to accept a 10% chance for failure. That is, the retiree accepts that with constant inflation-adjusted withdrawals, there is a 10% chance that wealth will run out before the end of 30 years. Also, there is no bequest motive or no worry about the magnitude of failure [my Advisor Perspectives column scheduled for publication on April 17 will delve further into these issues]. The table shows a recommended withdrawal rate of 4.3% and an optimal asset allocation of 45% stocks and 55% bonds. No cash. Another purpose of that article was to show that asset allocation does not matter all that much, as a stock allocation anywhere between 28% and 69% supports as nearly as high of a withdrawal rate as the optimal asset allocation.
Next, let's move to Table 1B, in which returns drop downward to get future real bond returns of 1% and the same historical equity premium. This figure compares the efficient frontiers for the historical data (in red) and this revised scenario (in black) for the 30-year retirement and 10% failure rate example:
And we need a table to accompany that figure to show about the asset allocation. This Table 2.3B is produced using the modified return assumptions in Table 1B:
Feel free to study and compare the tables. Again, to keep things manageable, I focus only on the one row. With a 30-year retirement and 10% failure, the sustainable withdrawal rate has dropped from 4.3% down to 3.6%. Actually, I don't think asset allocation should change when all returns drop the same amount. The small changes are an artifact of 10,000 simulations not being enough to get the perfectly exact results. There is still some fuzziness due to random variation, but as it was my rather souped-up computer had to run for 72 hours last August just to get the background info guiding this analysis with 10,000 simulations.
Finally, let's consider about Table 1C which also reduces the equity premium. Here is a figure again showing the efficient frontiers for the historical data (in red) and this revised scenario (in black) for the 30-year retirement and 10% failure rate example:
And here is a corresponding Table 2.3C produced using the modified return assumptions in Table 1C:
With a 30-year retirement and 10% failure, the sustainable withdrawal rate has dropped from 4.3% down to 3.2%. That equity premium reduction cut another 0.4% off the 3.6% withdrawal rate in Table 2.3B. It also leads to a lower optimal stock allocation, as the lower equity premium penalizes the more volatile stocks and makes them less attractive. The optimal stock allocation dropped from 45% to 29%, with the range of nearly optimal stock allocations falling to between 18% and 48%.
And so, assumptions about future returns do matter a lot! But the problem is: we don't know what the future will bring. It pays to be cautious and flexible. Don't rely completely on the idea of 4% being a safe withdrawal rate, because you only get one whack at the cat.