In this post, just so we have some numbers to discuss, I present some backtesting results for safe withdrawal rates with various combinations and glidepaths of bonds and bills.
1) JST dataset (https://www.macrohistory.net
) of annual stock, bond and bill returns (for the US, data are from 1872-2016). Bonds prior to 1929 are around 10 year maturity, From 1929 onwards, bonds have a maturity of about 20 years (see.the document data from “The Rate of Return on Everything, 1870–2015”, downloaded from https://www.macrohistory.net/database/
and references therein). Bills were of 3 month maturity, although I’ve assumed that the interest rate was available across the whole year (implicitly assuming that a one year bond was available at the same yield as the 3 month variety – this would have been a good enough approximation in most years).
2) Portfolio consisting of 50% stocks and 50% fixed income with withdrawals and rebalancing taking place on an annual basis. Results for a 30 year retirement period are presented.
3) Three runs were undertaken with constant allocations to fixed income (bonds only, used as a baseline, bills only, and a 50/50 mix of bonds and bills).
4) Three glide paths were tested (see following figure), Glide00 where a linear transition from bonds to bills start immediately after retirement, Glide10, where the transition started after 10 years, and Glide20 where the transition started after 20 years.
In the following table, the percentiles of the difference between the SWR for the test case (i.e. bills, 50/50 mix etc.) and the SWR for bonds for each of the rolling retirements are shown. So, what do the numbers mean? For example, for Bills only, at the 1st percentile, the number -2.17 indicates that in 1% of historical cases, the SWR for bills was over 2 percentage points lower than the SWR for bonds. The final column in the table (P>Bonds) indicates the percentage of rolling retirement periods where the SWR for the portfolio under test was better than the SWR using bonds. For example, the SWR for bills was better than that for bonds in 68% of retirement periods.
Code: Select all
1 10 25 50 75 90 99 P>Bonds
Bills -2.17 -0.51 -0.21 0.27 0.47 0.57 0.77 68
50/50 -1.08 -0.25 -0.10 0.13 0.24 0.29 0.39 69
Glide00 -0.45 -0.19 -0.08 0.07 0.13 0.20 0.32 63
Glide10 -0.21 -0.10 -0.05 0.02 0.06 0.11 0.22 58
Glide20 -0.06 -0.02 -0.01 0.00 0.02 0.03 0.09 51
A few general comments on the table:
1) Although historically, in the USA, bills have tended to produce a higher SWR than bonds, when bills did badly they did very badly indeed (although this tended to happen where the SWR was high, see some graphs I posted in viewtopic.php?t=376989
2) As might be expected, reducing the fraction of bills to 50% of the fixed income component brings the results closer to that of bonds, but about the same proportion of cases (i.e. 69%) remained better than using bonds alone.
3) Again, as expected, starting the transition from bonds to bills later gave SWR values closer to those with bonds. The glidepaths reduced the number of retirement periods where the SWR was higher than using bonds alone.
It is important to note that I have used rebalancing across the whole portfolio here – if the intention is to maintain the fixed income portfolio independently of the stock component then these results will not apply.
To my mind, the performance of the glide paths is not sufficiently attractive to warrant any additional complexity and I will stick to my (roughly) 50/50 mix of bonds and cash (with most of the cash in what, in the UK, are confusing known as fixed-rate bond accounts and, I think, cash deposits in the US).
Results for different countries were quite different (e.g. for Japan since 1950, the equivalent P>bonds values were all below 50%). So there is no reason to believe the historical outcomes presented here will continue into the future.