In December 2000 Scott Burns (click here for his website) published a retirement withdrawal strategy that he referred to as the "Omega Strategy". Click here for a direct link to this topic in Scott's website. The fundamental principle of the strategy is to buffer the equity portion of your portfolio with fixed income instruments and 'withdraw from your fixed income assets' during off market years and withdraw from your (more volatile) equity assets during good market years. He also suggested using either Treasury Inflation Protected Securities (TIPS) or I-Series Savings Bonds (whose returns are also indexed to inflation rates) as the asset class to use a 'the buffer'.
So there are two changes in 'traditional strategies' that Scott Burns suggested:
Scott was not specific about how this strategy would be implemented and (to my knowledge) no one has ever done any back-testing on this approach. This was intended to be the subject of this chapter. The truth is that I ended up (partly through my own misinterpretations) simulating a strategy that is fundamentally different than what Scott was trying to describe, although at a high level it matches his description as I stated it above. But it isn't the same thing and shouldn't be interpreted that way.
My interest in this withdrawal strategy is driven by the fact that the methodology discussed in the 'Simple and Efficient' chapter, while being simple and efficient, is also quite brutal in its treatment of withdrawals in bad market years. A single year income drop of 40% from the target income is not unrealistic using this approach.
To back test my alternations to the Omega strategy I had to decide what securities to use (I-Bonds, TIPS, or both) and exactly what does it mean to 'sell bonds in down markets'. If the market drops 20% in year n (certainly a down market) and then rises 25% in year n+1, do you then sell equities in year n+1 even though equity value has gone two years with no net growth? My withdrawal policy is listed below (there are probably better 'policies', but this seemed reasonable).
I used strictly I-Bonds as they have liquidation parameters that are easier to deal with as well as some attractive tax characteristics. This may not be directly applicable for all people because of lack time to accumulate the proper amount of IBonds ($30K/year/person limit or those having most/all of their assets in tax deferred accounts), but it should be implementable (spell check insists that this is not a word - well, it should be) by a fair number of future retirees, assuming that the government continues to offer I-Bonds at current real return rates. In my backtesting I used 3.4% for the fixed return portion of IBonds.
My withdrawal rules/portfolio management rules were as follows:
The general principles suggested by Scott Burns could be implemented in other ways, but these seemed reasonable to me.
As a starting point I used an initial withdrawal rate of 4.4% as suggested in the Jarrett Study and my work in The Methodology chapter. I started with an initial $1,000,000 portfolio and I chose to use the same 20%, 40%, 60%, and 80% equity ratios as I've used in previous analyses. In Scott's terms of 'portfolio immunization' this would be roughly 20, 15, 10, and 5 years of portfolio immunization.
I used the same basic simulation approach (54 different 30 year retirement scenarios) as described in The Methodology chapter and compared it to the Omega strategy (as I defined it). I compared the following resulting data points from the two simulations.
Now for the data:
This data makes it pretty clear that this altered Omega plan is clearly superior to the withdrawal/portfolio strategy and asset classes used the The Methodology chapter. It is particularly encouraging as it would indicate that (historically) withdrawal rates of 4.4% are sustainable beyond 30 years. But it leaves open the question of whether this pretty dramatic improvement in results is due to the change in withdrawal methodology or the efficiency of I-Bonds over Intermediate Government bonds (which made up the majority of the 'bond class' asset in previous analysis). I haven't done a complete analysis of this question yet, but the work that I've done so far would indicate that somewhere around 2/3's of the improvement came from the use of I-Bonds, with the rest of the improvement coming from the withdrawal rules changes in the Omega strategy. But this needs more work and I've decided to publish this as is.
For those who care the major differences between what I did and what Scott was suggesting are listed below.
In this chapter I investigated a strategy where you divide your portfolio into two pieces (one equities and one fixed income) and I defined withdrawal rules that said either withdraw from equities or fixed income, but never rebalance between these two pieces of your portfolio. The analytical data presented indicated a significant performance improvement over an annually rebalanced portfolio. But it used I-Bonds (real yield 3.4%) as the fixed income instrument which were not included in the analysis with which OmegaNot was compared. This naturally leads to the question "where did the improvement come from - an alternative withdrawal methodology or the use of IBonds?". I have repeated the OmegaNot analysis using the same asset classes (Intermediate Government Bonds for Fixed Income) in all cases and compared the OmegaNot withdrawal strategy with a 'normal' withdrawal strategy involving rebalancing to the target asset allocation at the end of each year. The results of this analysis also led me to an investigation into the volatility implications, which I'll also discuss.
The methodology simulated for the annual rebalancing case (I will refer to this as the 'base case') is:
The differences between previously published analyses and this one are (click here for documentation of the previously used methodology).
1) I did not include a 5% allocation to Treasury Bills
2) I did not take into account any returns that could be achieved on your annual withdrawal during the year
3) I have only analyzed a 50/50 equities/fixed income split
The OmegaNot withdrawal rules were:
I am presenting the following data:
Fixed inflation adjusted $44,000 withdrawals from an initial $1M portfolio
*1969 failed in the final year of withdrawals
WOW!!!!!! - the data certainly makes OmegaNot look like a real winner. In fact you can actually raise your withdrawal target from 4.4% to 4.9% without encountering a portfolio failure. This is a greater than 10% increase in income for simply changing your withdrawal strategy. Despite the fact that this is my data and my withdrawal strategy (although it originated as a misinterpretation of Scott Burns' Omega strategy), this doesn't pass my personal sniff test and looks suspiciously like a "free lunch" for these reasons:
As I have stated on the Morningstar Website (click here) in Investing During Retirement I find the the results of the simulations that I have done regarding the volatility of "MVI" (Market Variable Income - a term used there to describe the 'Simple and Efficient' withdrawal methodology) most disconcerting. However, as I stated in Efficient Frontiers in Withdrawal (click here) I spent a non-trivial amount of time on this topic and came to the conclusion that MVI-like withdrawal strategies (click here), given the work that I had done to date, are relatively efficient. It is also fair to observe that including international equity asset classes, REIT's, and a variety of alternative fixed income instruments such as IIS might well improve the volatility results significantly (the same could be said for the Trinity and Jarrett studies).
So I'm going to use "MVI" (as I understand it) to evaluate the volatility impacts of OmegaNot. For both the base case and OmegaNot the withdrawal amount will be 4.4% of your actual portfolio value, rather than 4.4% of your initial portfolio value (adjusted for inflation). As before I'll be expressing all results in inflation adjusted terms.
Volatility is a reasonably difficult issue to express succinctly. What we have is 54 different 30 year retirement simulations. Each 30 year simulation has its own unique set of results which are:
1) Inflation adjusted end portfolio value
2) The 30 different inflation adjusted incomes that you achieved
And we have 54 different sets of the above 31 different results (1674 numbers). For each 30 year retirement simulation I will calculate the following.
1) Median ending portfolio value (half more/half less)
2) Median inflation adjusted income across all 30 years
3) The 75th percentile minimum inflation adjusted income across all 30 years (75% were better)
4) The 95th percentile minimum inflation adjusted income across all 30 years (95% were better)
5) The minimum inflation adjusted income across all 30 years
This has reduced 31 numbers to 5, but we still have 54 sets of these 5 numbers (54 median ending portfolio values, 54 minimum inflation adjusted income values, etc). So I will simply "do it again" to these 5 sets of 54 numbers. I'll find the median value of all 54 different median ending portfolio values, I'll find the 75th percentile minimum of all the median ending portfolio values, .... I'll find the 95th percentile minimum of all the 54 different 75th percentile minimum inflation adjusted incomes .. It is easy to automate but tends to make your mind think in ways that it will probably object to. Mine does and it always takes me a minute or two to get it back after I leave this stuff for a while. "The 95th percentile minimum of 75th percentile minimum single year incomes" is not a concept that fits nicely in my mind for sure. Fortunately in this case a larger number is always a better number. So you don't necessarily have to have a complete grasp of this stuff to get a feel for how the numbers compare. I was using this analytical technique as a way to compare volatilities. Normal volatility measurements (e.g., mean and standard deviatio) always leave you with the question "which is better at a 95% certainty level - a mean income of $40,000 with a standard deviation of $5,000 or a mean income of $38,500 and a standard deviation of $4,000?"
I'll be presenting this data in a table and each entry in the table will have the form BC#/ON# where BC# will be the base case result for that entry and ON# will be the OmegaNot result for that entry. For example an entry of '34,713/36,123' is a base case result of 34,713 and an OmegaNot result of 36,123. The table across the top represents the statistics of the individual 30 year retirement scenarios. For example if you were interested in investigating the minimum single year income achieved you would read across the top of the table to the 'Minimum Single Year Income' entry. Remember that there are 54 of these numbers. Then you go down the column and find the median, 75th percentile, 95th percentile, and absolute minimum from these 54 different numbers. Now for the data. It is all in units of $1,000.
Base Case Results/OmegaNot Results - 4.4% 'MVI' withdrawals from a starting $1M portfolio
These numbers also show OmegaNot to (in almost all cases) be 'better' than the Base Case, although I find this data to be somewhat less compelling (more like a free snack rather than a free lunch). Even though these simulations did not ever allow your bond portfolio to go to zero, I still find this aspect of OmegaNot troublesome. I tend to agree with William Bernstein in that rebalancing between asset classes of dramatically differing mean returns is not particularly efficient. This leads me to the conclusion that the improvement of OmegaNot is mostly due to not doing annual rebalancing between equities and bonds. And there are certainly ways saner than OmegaNot to achieve this. Some alternatives are:
1) Rebalancing every 'n' years where 'n' is greater than 1
2) Rebalancing when your portfolio is more than x% out of balance. You might chose x=15% where you would only rebalance if (for the 50/50 target allocation case) your equities rose above 65% or fell below 35% (and this still leaves you with the option of either rebalancing to the original target or to the upper/lower limit).
3) Rebalance only when your equities are out of balance high
4) Never rebalance directly between equities and fixed income instruments, simply draw from the asset class(es) that is out of balance high
5) Chose a maximum annual rebalancing amount (in percent of your portfolio)
6) Rebalance half way between your current allocation and the target (or 2/3's or 1/3 or .4287465 or whatever).
7) Follow OmegaNot but set a lower limit to your bond allocation
The message that I am personally taking from all this is that I will continue to rebalance between various equity asset classes, but I'm going to be less likely than before to annually rebalance between equities and fixed income assets. My inclination is toward option #4 (and never say never).