Also, on Monday, William Bernstein published an article in the Wall Street Journal called, "How to Think About Risk in Retirement," in which he discusses rising equity glidepaths in retirement, and cites the rising equity glidepath article I wrote with Michael Kitces.
His column ends with:
Reverse glide path or two-bucket LMP/RP strategy? You say tuh-may-toe, I say tuh-mah-toe. Either approach will do a superb job of minimizing your risk of dying poor.I like this quote, because for me it represents the distinction between probability-based and safety-first approaches to retirement income planning. In the research article, Michael and I discussed rising equity glidepaths in the context of being a probability-based approach.
But in presentations I've done, I generally discuss the strategy as more of a safety-first approach. That is how William Bernstein eloquently describes the strategy in his column. This is what he means by the term "liability-matching portfolio." Safer assets (individual bonds or income annuities) are matched to specific retirement spending needs, and then the remaining discretionary wealth can be invested more aggressively in a "risk portfolio." That's the LMP/RP in the quote.
As you progress through retirement using a relatively conservative spending strategy, you are generally going to find, if stocks can enjoy some growth, that the percentage of remaining assets required to cover remaining spending needs is going to decline over time.
Actually, Michael and I wrote the "probability-based" article after first observing this sort of phenomenon in a "safety-first" article. (I should note that these opinions are mine, as Michael thinks the distinction between these two schools of thought is artificial and would probably not explain our research in the same way as I currently am).
We had observed in an earlier article that if you put half of your assets into an income annuity at retirement, and the other half into stocks, and then you treat the present value of remaining annuity payments as a type of fixed income asset when measuring overall household wealth, your equity allocation will generally rise throughout retirement:
Source: Kitces and Pfau, "The True Impact of Immediate Annuities on Retirement Sustainability: A Total Wealth Perspective"
This is why I'm generally confident in the idea that rising glidepaths are justifiable. It can be explained either as (1) a risk management tool to get the same or slightly improved outcomes with a lower lifetime stock allocation using a probability-based approach, or as (2) the natural outcome of someone basing asset allocation on their funded status and devoting the necessarily percentage of their assets to safer assets covering their spending needs, and the rest of their discretionary wealth to more risky assets. Tuh-may-toe or tuh-mah-toe.
Wade
ReplyDeleteIt seems to me that there is a major difference between your and Michael's approach and that of Bill Bernstein. All Bill is saying is that once you have a LMP, you can invest the rest in equities because your needs in retirement are already covered. The equities can be expected to rise, but, regardless, these are not counted on for retirement living expenses. In other words no probabilities are assigned to how the equities will perform, and in the unlikely event they go to zero, the retiree is still covered. Bill does take one small liberty with Fred's LMP in his example, he does rely on the equity piece in the event Fred lives to age 98.
As I understand your approach, you do assign probabilities to equity performance. This enables you to forecast higher expected returns from smaller portfolios that rely on some level of equity performance. Definitely not tuh-may-toe or tuh-mah-toe as I see it.
John, I agree with what you are saying. This is a fundamental difference between probability based and safety-first. The tuh-may-toe or tuh-mah-toe point is about how a rising glidepath falls out of either approach.
DeleteListening to your and Michael's presentation, and our conference conversations, both at the Academy of Financial Services conference, there is a difference. The retirement spending question is not a one-time set-and-forget exercise. Under such a perspective, the transitions into spending capability are ill defined as one ages. Such discussions ignore aging transitions since they argue beginning spending patterns, but ignore older retirees who don't fit the initial model, as well as present retirees aging into later spending.
ReplyDeleteFrom the perspective that spending is "so conservative" that the balance of equity essentially may be untouched, then logically the balance would continue to grow untouched and would eventually become so large that rising equity would make sense. However, your own data shows that, as a person ages into shorter remaining time periods, the actual equity exposure would be less relative to the longer periods (i.e., 30 years, 25 years, 20 years, etc remaining). Also, leaving money untouched on the table is not what people tend to do behavior-wise.
Now - should a person begin to rely on their portfolio balance for a larger share of their income, then that exposure to equity should logically be reduced to reduce sequence risk exposure. However, as we've talked about Wade, there are a couple of approaches to measuring exposure to sequence risk (the scary term for safety first/LMP). One is through simply evaluating the percentage of failing sequences (commonly referred to as POF) of a Monte Carlo simulation as Mitchell, Blanchett and I demonstrated in our JFP paper (Nov 2011). As
POF rises, so does the withdrawal rate - but POF is a clearer signal than the withdrawal rate that something should be adjusted (i.e., spending).
This said, another method is simply to look at Standard Deviations. standard deviation 68 95 99 rule http://en.wikipedia.org/wiki/68%E2%80%9395%E2%80%9399.7_rule . One could pre-calculate using standard deviations to see how far a portfolio would need to decline to arrive at 1, 2 or 3 STDs from the present value and also get a sense of how often such a decline may occur (e.g., 3 STD covers 99.7% of the time periods - arguably a very conservative portfolio value if spending were to be based on that). So continuing the example, using 3 STD portfolio value (one may use a 2 STD calculation as well) would represent the LMP component of the portfolio value. The difference between the # STD portfolio value and the present portfolio value would represent the "discretionary" spending component of the portfolio. The entire portfolio is used for both components of the spending plan. However, there's no need to separate assets into different parts, because a properly diversified indexed portfolio is unlikely to go to zero value (requires ALL holdings in the indexes to go to zero simultaneously).
This method separates spending from the portfolio into the likelihood of discretionary spending reductions and unlikely spending reductions (LMP component using 2 or 3 STD portfolio value). It also more closely aligns with client spending behaviors and is more dynamic in measuring and monitoring transitions EACH YEAR as a person ages.
The above link for my name takes the reader to a post-AFS summary that goes into this further and compares SWR to the Dynamic Updating method. LMP models are simply a subset of a more complete Dynamic Updating model.
An interesting post Wade. The real difference is that tuh-may-toe is static and tuh-mah-toe is dynamic.
PS. Here's a great book on how portfolio design done properly adjusts sequence risk exposure http://www.amazon.com/Reducing-Risk-Black-Swans-Volatility/dp/0615992978/ref=sr_1_2?s=books&ie=UTF8&qid=1417727590&sr=1-2&keywords=black+swans ... the concepts here is where my idea about the degree of exposure to risk using STD comes from. But really, STD is simply another way of determining POF exposure, albeit without the use of Monte Carlo to get the answer.
ReplyDeleteLarry, Thanks.
ReplyDeleteI was at a conference today and saw a presentation about JP Morgan's whitepaper, "Breaking the 4% Rule"
A few things clicked for me. And I think this is what you are getting at. Also Joe Tomlinson has been making the point.
The JP Morgan paper calls for a declining equity glidepath with age. But their paper is very much a dynamic spending strategy type paper that will keep upping the withdrawal rate with age to try and get the portfolio down toward zero at death.
Whereas, as you know, and as you keeping emphasizing, Michael and I are assuming the constant inflation-adjusted spending, which leads to a very wide distribution for terminal wealth, and very frequently results in a declining withdrawal rate with age, as well as an increased funded status.
With our assumption, I think the rising glidepath makes sense.
But with a dynamic spending assumption aimed at not leaving so much terminal wealth, I can very much see how the rising glidepath idea can go out the window.
Have we talked about before how there is a certain time frame around maybe about 12 years, where if the horizon is less than that then all bonds maximizes success, but if the time horizon is longer, than all stocks maximizes success?
Basically, I'm starting to see better why rising glidepaths may not fit in well with dynamic spending strategies.
I'm glad to see some light is shinning on the differences Wade. I saw a summary of their paper. However, an ever shortening time frame "near death" has a risk, as you well understand, of being outlived. Thus, this is the purpose of our paper (link via my name above) JFP Dec 2012. There is a tendency in static withdrawals based on fixed periods to develop an exponential growth (RMD table has this too).
DeleteThe adjustment to remove the exponential nature is to multiply the derived withdrawal rate by (1 - 1/n) where n is the distribution period used to derive the raw withdrawal rate. This adjustment helps preserve the portfolio balance for the contingency of outliving that period of n years. Another adjustment may also be to use slow, ever increasing percentiles of the period life tables, i.e., slightly longer periods that people may not outlive as they get into their late 70's, early 80's and beyond. This would be based on their health - healthy continue to live longer relative to their peers.
My light bulb moment was when Michael explained in his AFS presentation that a low withdrawal rate at the beginning of retirement, that is basically adjusted so it remained low throughout retirement means that the portfolio balance is essentially not fully used for spending. Thus that portfolio growth continues to grow resulting in even greater unused balances - which logically, and demonstrated by your results, would lead to greater equity allocation since larger and larger portions of the portfolio is not dedicated to spending (SWR restricts spending).
Nothing wrong intrinsically with either approach - however, retirees and practitioners should more deeply understand the dynamics of either. Our model simply simulates what an annual review would look like throughout future older ages and shines light on issues that should be understood and managed early on, and throughout retirement.
Yes, withdrawal rates can go up with age each year, because the retiree's time remaining decreases slightly each year (not a year for year exchange as you well understand). This also translates into recognizing that all the principal one used to have, is not necessary to preserve (unless bequest motive is present) because one also needs less money to last for a shorter period of time. The interplay between period and balance may be evaluated through a Dynamic Updating approach.
Always a great conversation Wade!
I think prior link to JFP Dec 2012 paper returns an error here it is again ...
Deletehttp://www.onefpa.org/journal/Pages/Transition%20Through%20Old%20Age%20in%20a%20Dynamic%20Retirement%20Distribution%20Model.aspx
Larry,
DeleteIf I were going to put together a table which reports on estimated withdrawal rates to be used, according to various methodologies out there, and I wanted to include an entry for the Frank et al. work, which of your articles has what you view to be the most up-to-date methodology that best reflects your current views?
Thanks, Wade
Wade, this blog post summarizes the use standard deviation as a way to measure necessary versus discretionary (floor and ceiling) expenses as well as comparing SWR to a Dynamic Updating model:
Deletehttp://blog.betterfinancialeducation.com/multi-media/how-income-may-compare-between-dynamic-and-safe-approaches/
This post goes into detail describing how the probability of failure (percentage of simulations that fail reaching their respective distribution period) may design pre-established decision rules:
http://blog.betterfinancialeducation.com/sustainable-retirement/how-to-use-dynamic-updating-to-determine-a-prudent-retirement-income-based-on-age/
Finally, this post discusses how all the papers relate to each other ... they build on each other progressively looking at going from static distribution periods, to age based, to 3D, to superannuation, to comparing SPIAs to managed portfolio expected cash flows:
http://blog.betterfinancialeducation.com/sustainable-retirement/how-to-use-dynamic-updating-to-determine-a-prudent-retirement-income-based-on-age/
Let me know where you have questions. Thanks, Larry
I'm writing from the UK where everyone but me says Tuh-mah-toe! As I've learned more about dynamic withdrawals, I think that there may be a connection to the type of utility analysis that Bill Sharpe and others have done where they show that if a person has a risk aversion coefficient of X, their optimal stock allocation is Y%. I believe what may be happening with variable consumption (dynamic withdrawals) is that the flexibility to vary the withdrawals gives the individual the flexibility to stay closer to their optimal manner of managing wealth (combination of withdrawals and asset allocation) rather than having the portfolio and optimal allocation drift as may happen with fixed withdrawals. This may have been what Paul Samuelson was getting at in 1969 with some math that goes over my head. It would be interesting to see some down-to-earth analysis using dynamic withdrawals (RMDs or something similar) and seeing which glide paths work best.
ReplyDeleteHi Joe, I've come to realize that there really is no specific glide path that would work for everyone. We found in our research data, that the distribution period along with the desired POF is what drives the withdrawal rate primarily. Then, there is an allocation that provides a maximum withdrawal rate given the other two factors (recall we looked at this through our research approach which we called 3D ... (Figure 2 in this blog post are selected time slices through the 3 dimensional data:
Deletehttp://blog.betterfinancialeducation.com/sustainable-retirement/how-to-use-dynamic-updating-to-determine-a-prudent-retirement-income-based-on-age/ ).
However, this said, there is a spectrum of glide paths that exist between Wade and Michael's rising one (that is based on keeping withdrawal rates on the very conservative side of the spectrum - and thus, portfolio balances grow "unused" towards retirement income until such time withdrawals may be reset and increased) and what we see in our data which is a declining equity allocation (based on continually calculating and evaluating spending and adjusting it based on the period life table expected longevity (50th percentile), or some more conservative percentile adjustment that a cohort may be continually unlikely to outlive (link to the superannuation research paper found here: http://www.betterfinancialeducation.com/page-2 ).
All data is stochastic (even the IBM of 20 years ago is different than the IBM of today, etc) so updating data as it comes fact and re-evaluating the new situation is part and parcel to the Dynamic Updating model. The allocation answer is, in truth, a spectrum of allocations depending on how the distribution model is designed. And there does exist flexibility in switching between them constrained by the retiree's situation.
Hi Larry, It's hard for me to translate between the way you think about this and the way I do, because I'm thinking in terms of maximizing the utility of lifetime consumption and I believe your annual updates are using more traditional financial planning measures—probability of failure, bequest amounts. In general, I think the dynamic programming approach I like (which will also involve dynamic withdrawals) tends to generate higher initial withdrawal rates and stock allocations than traditional financial planning models (e.g. , 4% rule). So my way would not likely call for a rising equity glide path. Perhaps the dynamic withdrawals give more flexibility to be aggressive in the early years.
DeleteIndeed Joe, the beginning withdrawal rates do tend to be higher than the 4% Rule (30 years), unless the joint time remaining from the period life tables (Annuity or Soc Sec) generate a time frame longer than 30 to 35 years, in which case the rate is less. Most of the time I find the allocations start at about 40% equity and reduce very slowly from there to about 15 to 20% in late 80/early 90's.
DeleteUtility is a slippery definition for me because that is very sensitive to how utility is defined. I find Lifetime Expected Cashflow, between median (50th percentile), poor (75th percentile) or good (25th percentile) markets (simulations) provides a better sense of a potential range of outcomes dependent on the market sequences. Yes, the withdrawals are dynamic in the sense that not all the portfolio is at risk of requiring spending reductions (unless all the portfolio were invested in a single stock, e.g., Enron).
I discuss the perception of invested portfolios being all in or all out as far as supporting spending through a floor and ceiling discussion midway in this blog http://blog.betterfinancialeducation.com/multi-media/how-income-may-compare-between-dynamic-and-safe-approaches/ ... basically, one could use standard deviation to evaluate a "floor" (aka SWR) withdrawal dollar amount, and a "ceiling" dollar amount. The spending floor may not need adjustment at all if 2 to 3 STD values are used (unlikely events - but possible). The spending amounts between the floor and ceiling are discretionary - by definition, adjustable.
I agree, using spending above the floor up to a ceiling amount tends to result in a declining equity glide path simply because the portfolio balance is not allowed to balloon into a bajilion dollars as Michael says tends to happen under the SWR (you recall his presentation at AFS). That excess funding would naturally allow for higher equity later since it is unused for spending needs. In a dynamic model, that excess balance is used for discretionary spending ... thus requiring less equity exposure to keep volatility drag muted.
As you may perceive, and have said, leaving more on the table, not used for spending, implies spending less relative to a dynamic spending model (dynamic in as far as the discretionary spending component may change - NOT all of the spending; there still remains a necessary expense buffer if calculated properly).
The dynamic model thus tends to give more flexibility, to either/both allocation and spending, not only in the early years, but throughout the lifetime of the retiree - relative to the SWR model.
Great discussion Joe, and happy to correspond with you again! Look forward to seeing you again at AFS in 2015!
Great to see the WSJ piece on HECM reverse mortgages! (I've asked you about this on one of your older blog entries.)
ReplyDeleteWade - wonder how the chart would change if the AUM of the Stock (Equity Portion) were to have a 1% fee-drag and taxed as NQ assets as compared to the SPIA with no fee-drag and FIBO (First In/Blend Out) tax treatment?
ReplyDeleteWade -- thanks for sharing your work. Is there an optimal glidepath leading up to retirement age for a rising glidepath strategy? Is a linear or precipitous decrease in equity allocation preferable when approaching the departure point for a rising glidepath strategy?
ReplyDeleteHi, it's a good question and one I haven't specifically attempted to answer precisely. But I do think the correct answer will be closer to the precipitous decrease side in the last 5 or 10 years before retirement, rather than being linear the whole time.
Delete