In the retirement planning literature, a landmark study is Cooley, Hubbard, and Walz's 1998 paper, "Retirement Savings: Choosing a Withdrawal Rate That is Sustainable," published in the AAII Journal. The paper is often called "The Trinity Study" because the 3 authors were all professors at Trinity University. The study looked at historical data for the United States to determine the chances for success for different withdrawal rates from one's retirement savings for different time horizons and for different asset allocations. Most of the study concerned withdrawals that are not adjusted over time for inflation, but their Table 3 assumed inflation-adjusted withdrawals and is therefore the most important part of the paper. For the annual data between 1926 and 1995, they found that a portfolio consisting of 50% large-capitalization stocks and 50% high-grade long-term corporate bonds would have provided a 95% chance for success in supporting inflation-adjusted annual withdrawals starting with an initial withdrawal of 4% of the portfolio value at the time of retirement.
I think people are sometimes curious if this result may have changed if the study were repeated today, especially in light of the recent financial crisis. The simple answer is: no. Though financial markets have provided bad returns in recent years, those would have only shown up in the final years of retirement, and sustainable withdrawal rates are much more sensitive to the returns experienced at the beginning of retirement instead of the end of retirement.
To see more clearly, let's dissect that 95% chance of success number a bit more. The Trinity Study had data from 1926 to 1995. To consider retirements lasting 30 years, this means they could only consider retirement dates from 1926 to 1966. For anyone retiring after 1966, they couldn't calculate the withdrawal rate sustainable over 30 years because they didn't have the data. 1926 to 1966 represents 41 beginning retirement dates. Of those 41 dates, the 4% inflation-adjusted withdrawal rate failed 2 times, in 1965 and 1966. Thus, it's success rate was 39/41 = 95.12%, or 95% when rounded down.
Now complete annual data is available through the end of 2009, giving us now the chance to look at 55 retirement periods of 30 years, as we can now consider retirement dates between 1926 and 1980. When we want to discuss a 30 year retirement duration, we still don't know what happens for retirees after 1980, because their story has not been finished. Of the 55 periods now available, there are still only 2 failures, 1965 and 1966. There hasn't been any failures since then. Now we have a success rate of 53/55 = 96.36% or 96%.
I'm sorry for not formatting it real nicely, but here are the numbers updated with data through 2009 that you can directly compare to Table 3 in the Trinity Study, and the link for that study is provided above.
3% 4% 5% 6% 7% 8% 9% 10% 11% 12%
100% Stocks
15 Years 100 100 100 93 83 76 70 64 54 47
20 Years 100 100 91 80 71 63 54 46 40 31
25 Years 100 100 87 73 63 52 42 37 28 20
30 Years 100 96 78 62 55 42 33 29 20 11
75% Stocks
15 Years 100 100 100 96 86 74 71 57 49 41
20 Years 100 100 92 80 69 60 49 40 29 15
25 Years 100 100 85 68 57 43 35 27 12 7
30 Years 100 98 76 62 45 36 20 5 2 0
50% Stocks
15 Years 100 100 100 94 83 71 61 46 39 23
20 Years 100 100 92 80 63 46 35 22 9 6
25 Years 100 100 82 62 45 27 15 8 7 2
30 Years 100 96 69 49 20 11 2 0 0 0
25% Stocks
15 Years 100 100 100 91 76 60 46 34 26 14
20 Years 100 100 86 58 45 31 23 15 8 2
25 Years 100 95 58 35 23 13 10 8 2 2
30 Years 100 75 33 22 7 2 0 0 0 0
0% Stocks
15 Years 100 100 100 77 51 37 34 29 19 13
20 Years 100 92 58 35 29 26 20 9 5 2
25 Years 100 58 32 23 18 12 8 5 2 2
30 Years 85 35 22 11 2 0 0 0 0 0
April 2011 Update: The authors of the Trinity study published updated figures using this same data in the April 2011 Journal of Financial Planning. I provided more thoughts about this in my "Trinity Study Updates" blog entry.
It is worth noting that you can get 12 times as many data points if you use monthly data. About 52 times as many if you use weekly. Or about 250 times if you use daily.
ReplyDeleteBut you would learn almost nothing new.
The reality is there was one period of a couple of years where this would not have worked.
But it is further worth noting that using overlapping periods gives an illusion that you have lots of data, but this is false. Fifty five overlapping periods of 30 samples is less than two independent data points.
There is no way to generate statistically meaningful results from two data points.
One can report out results to four figures (95.12%) or round to two figure (95%), but with less than two independent data points both are rather overly precise.
We do not know the long term returns well enough to say they will be about 7%, but not 8% and not 6%. That is we don't even have one significant figure.
In a world were we have no guarantee of future results estimating a fixed and safe 30 years withdrawal rate is at best unwise.
The world is far too uncertain for that.
The original paper was written largely to show that the then popular notion you could withdraw 10% was unwise. People then picked it up and grabbed it as if the 4% figure and approach of fixing a withdrawal rate was an actual plan. This was not in accord with the beliefs of the authors.
Rod Cole
Dear Rod, Thank you for the comments and I due consider them a lot. I hope to write something about this one of these days.
ReplyDeleteFor now, there is another matter. I'm quite thankful that a couple nice people have provided links to this blog entry, but in the ensuing discussions about it, I've seen quotes of the first part of one of these paragraphs:
I think people are sometimes curious if this result may have changed if the study were repeated today, especially in light of the recent financial crisis. The simple answer is: no.
But leave out the VERY IMPORTANT second part of the paragraph:
Though financial markets have provided bad returns in recent years, those would have only shown up in the final years of retirement, and sustainable withdrawal rates are much more sensitive to the returns experienced at the beginning of retirement instead of the end of retirement.
Though that last sentence can qualify as a good example of a run-on sentence, it is very important and needs to be left with the first part of the paragraph. In the context of how the Trinity Study works, we don't know yet how things will turn out for recent retirees.
There is something wrong with financial planning when everyone- even Jack Bogle- uses rolling period analysis and acts as though it is valid statistically. It is not and it doesn't matter who the author's names are on the study. Example: the DOW was up 44% in 1954 and by using rolling periods you are counting that over and over again.
ReplyDeleteOne wonders what is afoot? Fred
Thanks for writing. Perhaps the best mix is to look at both Monte Carlo simulations and historical simulations, as they each have advantages and disadvantages.
ReplyDeleteYou dhould have your work proff-read prior to publishing. by an adult.
ReplyDelete