11 Comments
User's avatar
Jim Dorsch's avatar

I assume your diversifying triangle weights are based on dollars and not risk. if that's the case, then risk is skewed hard toward gold and bonds. it depends on the time period, but over the lifetime of DBMF, about 50% DBMF, plus 25% each bonds and gold, would realize rough risk parity. in tests, this increases Sharpe and Sortino along with return. granted, these tests necessarily cover a brief timespan, but it is intuitive that one would want to hold more DBMF than bonds or gold, based on their standard deviations.

Sam C's avatar

Excellent post as usual. I am learning so much from your portfolio experiments, and the explanations behind them. It's a must-read for me each week.

Just one question: Substituting 50% IDMO and 50% AVDV for the AVNV portion? I realize you are trying to make a simple portfolio template, but that combo doesn't alter the simplicity much, duplicates the growth/value barbell approach used with the US stocks, and seems to produce higher yields. Thoughts?

Justin's avatar

Yeah, that's smart, and though I haven't run the numbers, my guess is that having momentum and value for the ex-US position would be a good bet for improved returns. This is getting into the territory of what I do versus what I recommend to people not so into this stuff as me (or you, I presume). I like the two competing triangles concept because I think I can explain them in two minutes, but if I start to get into further specificity, it can get to be too much.

Brian Paulson's avatar

What happens if you bring the Diversifying Triangle beyond 50%? I know that equities are the growth driver but cropping the table here makes it look like things will continue to improve with more non-equity assets. I'd like to see the table extended so that we see a peek in the SWR along with the growth falling off rapidly.

Justin's avatar

Thanks for the reply and for reading. Good suggestion. I ran another backtests and extended the table, and put it on as a postscript to the post. If you go back to the post, you can see. Noticings: 40/60 and even 25/75 don't look that bad, but then 10/90 and 0/100 start to fall apart. I wrote in the piece that 60/40 might be my sweet spot if I were really risk-adverse, but running the numbers, you can make the case that 40/60 is actually the strongest - lowest ulcer index, highest 20-year PWR and competitive with 35-year PWR. Hmmm. Thanks for the question!

Matt's avatar

Thanks for the article good info here. I know the backtest was limited by the data, but I’m thinking 1988 was a pretty favorable start date? Wonder what if you changed the start date to say 1999 how it would look, I guess I should run it 🤔 keep up the good work 👍

Justin's avatar

Point taken - yes, usual disclaimers about backtests apply, and am definitely not saying that those numbers will repeat.

I do think it will be directionally correct - increasing amounts to diversifying assets will lead to less volatility, shallower drawdowns, lower Ulcer Index and higher perpetual withdrawl rates.

Joe Q's avatar

Great article and simple analysis. Oddly enough, as I've been working on my own portfolio with Uncle Frank's knowledge, I came up to something rather similar by using 17% ZROZ as simple leverage and using the added 9% from TLT and 6% cash to spruce up the typical GR portfolio. Here's a testfolio link going back to 1969 comparing with 50/50 and 100 triangle. What is really interesting is if you go back to 1969, the 50/50 outgrows the 100% equity triangle. It pays to diversify!

https://testfol.io/?s=3JQv0W6TgzS

Justin's avatar

Yep, for sure. Thanks for sharing your version of the GR - I like Frank's framework, it does allow for lots of fun variations.

In that spirit, I took your GR Joe portfolio and swapped out some of the VUGSIM for TQQQ SIM, and then bumped up most of the other assets in the portfolio. Called it GR Joe and Justin: https://testfol.io/?s=4j1XTgY31Mq

Joe Q's avatar

That’s an impressive stack. Know of a way to backtest UPRO or TQQQ back to 1970? I added an adjustment to switch TQQQ to VUG pre 1986 which makes its all the more impressive.

That’s my main issue with stacking, lack of clear backtesting ability outside of using strips .

Justin's avatar

Yep, you can do that pretty easily. You use ticker modifiers (which you can learn more about on the help page for Testfolio), but basically, it is this:

QQQSIM?L=3&E=0.91

The L stands for leverage, and then E is the expense ratio.