Based on the trinity study and cfiresim you should be very safe at 4%. Saying that it only has a 52% success rate shows something is wrong. Also at some point having 50 years or 250 years should have a minimal change to the SWR. I will stick to cfiresim until a better model comes out.
Some food for thought - in MMM's post
http://www.mrmoneymustache.com/2012/05/29/how-much-do-i-need-for-retirement/ he shows the 4.04% SAFEMAX value in Figure 2.1. Actually, that figure is derived from William Bengen's work in 1994 (the author, Wade Pfau, extended the results to 1981 using data up to 2010), not the Trinity study. The Trinity study occurred four years after Bengen's seminal work. The only difference between their work is the bond indices they chose. With Bengen, he showed 100% success at 4.15% WR; the Trinity study showed a 95% success rate at 4% WR (not 100%!!!).
If you go to the webpage MMM got that figure from, it redirects to
http://retirementresearcher.com/the-trinity-study-and-portfolio-success-rates/. Looking at Table 2.1, you can see that at 4% withdrawal rates you have 100% success at 30 years, but only 96% at 35 years and 85% at 40 years.
I tried a really simple experiment in cFIREsim and obtained comparable results. I started with 2015-2045 for a 30 year retirement, with $100 portfolio, $4 spending (4%) with inflation adjust , 50/50 stocks/bonds, no fees, and everything else set to zero. That matches the study parameters. I then ran simulations for 30, 35, 40, ...., 70, and 75 year retirements. I stopped at 75 years for two reasons: most people won't be retired that long and also the data starts to be statistically insignificant because at 75 years only 70 cycles are ran, 80 years gives 65 cycles, etc.
Well, the results were so bad for 50/50 stock/bond allocation I also ran 70/30. The results are shown in the table below. As you can see the success rate keeps dropping and it is not clear to me that there is an asymptotic limit. I'm not sure where the assumption that if the money lasts 30 years it will last forever came from but I consider it suspect. If someone can post some hard calculations or a research paper that shows this I would appreciate it.
Years Retired | 50/50 Success | 70/30 Success |
30 | 90.4% | 93.9% |
35 | 82.7% | 90.9% |
40 | 67.6% | 84.8% |
45 | 65.0% | 80.0% |
50 | 63.2% | 81.1% |
55 | 60.0% | 82.2% |
60 | 60.0% | 82.4% |
65 | 55.0% | 80.0% |
70 | 53.3% | 78.7% |
75 | 50.0% | 77.1% |
Also, I ran the MMM values in cFIREsim that I used in the spreadsheet. 90/10 allocation, 4% spend, 0.1% fees, inflation adjust, 46 year retirement. I got an 83.84% success rate with cFIREsim. Higher than the simple spreadsheet but still not 100%.
So, the safety factors MMM mentions in article listed above are not safety factors; they are an
absolute requirement.