Permanent Portfolio vs. All-Weather - an in-depth analysis

[FF9000]

Both portfolios have a similar goal and approach. I've assessed the portfolios against each other over several periods, using a stochastic bootstrapped model, which I explain below.

Harry Browne's Permanent Portfolio:

• 25% VTSMX (Total Stock Market)
• 25% VUSTX (Long-term government bonds)
• 25% VUSXX (T-bills)
• 25% GLD (Gold)

Ray Dalio's All-Weather Portfolio:

• 30% VTSMX (Total Stock Market)
• 7.5% PCRIX (Commodities)
• 40% VUSTX (Long-term government bonds)
• 15% VFITX (T-bills)
• 7.5% GLD (Gold)

Modeling approach
Next, I also ran a bootstrapped stochastic model. For a given test period (e.g., 1972-2015 or 1980-1989, etc.), I ran 1,000 simulations for each portfolio over a 6-year period. I used 6 years because that is in how long I would like to retire, but I can change this analysis as needed to be longer-term. Each simulation was created by bootstrapping 2-3 year sequences randomly from the selected period. So for example, if I was modeling the 80s, I might randomly select the returns for each asset class from: [1982, 1983, 1984], [1988,1989], [1980]. The idea was to not necessarily look at historical results as they played out, but to take the economic climate of each period and allow for some randomization/shuffling of year sequences to create random iterations.

Results



Interestingly, over the entire period the portfolios are very similar, both in terms of average returns and Sortino Ratio. For those of you not familiar, the Sortino Ratio is like the Sharpe Ratio, but instead of looking at average returns compared to volatility, it looks at average returns compared to only down-side volatility below a specific number. In this case, I was only interested in seeing volatility for annual returns below 4%.

But if you look at each specific period, you can see that the PP outperforms substantially in the 70s whereas the All Weather Portfolio looks to have both better returns and lower down-side volatility for all periods since then.

These returns have no been inflation-adjusted, but inflation adjustment is not useful for comparing two portfolios over the same period, as inflation would have the same impact on both.

Would love to hear thoughts and ideas for follow-up analysis.

[Reub]

Thanks for this! Although the returns are fairly equivalent, we have been in a long term bond bull market and having 40% in bonds, as the All Weather Portfolio does, seems like more of a risk than the PP.

[buddtholomew]

What stands out to me is the falling PP CAGR.

[Kbg]

FF,

Can you run this taking 5% out of gold and realigning it to equities?

Thanks!

[barrett]

What stands out to me is the falling PP CAGR.

I'm pretty sure that is almost entirely due to falling inflation. That's where Tyler's portfolio charts are so helpful. You can see how a portfolio is doing in real terms compared to how it has done in the past in real terms. We should also expect to see more nominally negative years for the PP... should be about one in three if inflation is at zero.

Oh, and FF9000, thanks for posting your work.

[buddtholomew]

Thanks for that Barrett.

[Tyler]

Modeling approach
Next, I also ran a bootstrapped stochastic model. For a given test period (e.g., 1972-2015 or 1980-1989, etc.), I ran 1,000 simulations for each portfolio over a 6-year period. I used 6 years because that is in how long I would like to retire, but I can change this analysis as needed to be longer-term. Each simulation was created by bootstrapping 2-3 year sequences randomly from the selected period. So for example, if I was modeling the 80s, I might randomly select the returns for each asset class from: [1982, 1983, 1984], [1988,1989], [1980]. The idea was to not necessarily look at historical results as they played out, but to take the economic climate of each period and allow for some randomization/shuffling of year sequences to create random iterations.

Nice work!

You might want to be careful about sample size. As I understand it, you're using 2-3 year chunks with single years only when necessary to add up to 6 (a 2-year chunk + a 3-year chunk requires an extra year). The problem is when you only draw those chunks from 10-year pools to keep your clean decade segregation. By my calculations, there are only 450 possible non-overlapping combinations using those criteria. So if you're doing 1000 random simulations, there's a ton of redundant data in there that can skew the averages. How do you know you're not inadvertently over-sampling the single best year for one portfolio and the single worst for the other?

But if you look at each specific period, you can see that the PP outperforms substantially in the 70s whereas the All Weather Portfolio looks to have both better returns and lower down-side volatility for all periods since then.

Here's another way to look at it. These show every rolling 6-year real compound return since 1972 (with actual sequence of returns):




I'd count the 70's for the PP, the 80's and 90's for the All-Weather, and the 2000's very close but leaning PP. Overall, the All-Weather portfolio clearly had a weakness for high inflation and rising interest rates, while the PP was way more consistent overall.

BTW, I love your analysis. It's fun to have another data-head around.

[FF9000]

Thanks for the feedback, guys! Good stuff for me to build on. FYI, the reason I like building my own models in Excel is that I like custom views that are not always offered on the admittedly much better versions (such as Tyler's) and that I honestly just love making models. Weird, I know. =P

Kbg - I need to tweak a couple of things in the model but will be happy to run a 30/25/25/20 at some point in the next few days.

barrett - Immediate next step is to tweak model to output both nominal and inflation-adjusted returns.

Tyler - you hit the nail on the head with your post. There is certainly redundant data in my approach right now, though the results are very stable across multiple runs. I'd love to get your thoughts on the most robust way to test each portfolio. I see a few different levels of depth:
1) Historical sequences of back-tested returns (what you show below).
2) A stochastic mix of all possible sequences in a given period, using various block sizes - e.g., limiting a 1980-1989 test to 450 iterations if using 2-3 year blocks. Interesting thing in this case is whether there are better cut-offs, from an economic cycle perspective, then decades. I would love to hear peoples' view on this.
3) A statistical (not bootstrapping approach), where for each asset class we determine mean, volatility, and covariance with other asset classes over a given period of time and use these to project stochastic scenarios into the future. Unfortunately this requires Principal Component Analysis (or rather, that is the only way I know how to do it), which gets quite complicated. I have an old Excel model from my professional life that does this, but I'd need to dust it off and remember how it works. I would also want to run this under different economic regimes since correlations between asset classes change over time and should probably not be "averaged". Rather, we should have the regime be randomized and the correlations for a given regime used for those iterations.

[Tyler]

I'm all about making your own models. Good for you!

The trick with setting "regime" cutoffs is that I'd argue they vary by asset. Gold clearly behaved differently after 1972 with the repeal of Bretton Woods. Maybe you can draw a line for treasuries with rising rates before 1981 and falling rates after. You can make the case that the 80's and 90's were the perfect storm for US stocks. And maybe cash can be split when it had a positive and negative real return. But finding clean cuts that make sense for all of them is a lot more difficult.

Also, randomly mixing up the returns as if reactions to certain market events happen in a vacuum seems intuitively questionable to me. Markets tend to operate in cycles and are not completely random, and as we've seen with the PP certain assets play off of one another. Your idea to account for correlations does make some sense in that regard, but as you mentioned correlations do change so picking the right one for the right condition ahead of time is easier said than done.

Basically, I've personally always struggled with understanding how stochastic methods model an actual functioning economy rather than simply randomly mixing numbers with no economic basis behind them. But I admit that may be a gap in my own understanding and not a fault with the methods. If anyone can point me to research on that, I'd be interested in learning more.

Taking a step back for a moment, what are you trying to learn?

[FF9000]

Good question - what am I trying to learn? I ask myself that sometimes.

In a nut shell, I have very specific financial goals and I want to pick the portfolio that gets me to my goals within a set period of time with a ~99% confidence interval. Now, of all portfolios that get me to my goal in 99% of scenarios, I want to pick the one with the highest upside (Sharpe ratio, Sortino ratio, etc.), so that I give myself as much potential upside as possible so as to reach my goals faster and/or be wealthier than I expect at the time I want to quit working.

All of this over-analyzing is to get really comfortable - emotionally and intellectually - with the selection I make. I love the idea of the Permanent Portfolio, but just as with any allocation or philosophy, there are people who take short-cuts in demonstrating its performance - Harry Browne not exempt in somecases (as much as I love the guy). Once I am intellectually comfortable with an allocation, I'll invest my money, rebalance it every once in a while, and forget about it, focusing on other parts of my life. But in this pivotal period where I am actually deeply thinking about how I will invest my money for the future, I want to leave no stone unturned.

In a way, I am chasing a certainty that doesn't exist. But just going through the process helps me get comfortable. I dislike the feeling of committing my money to an allocation just because some expert said to, so I need to get to that level of comfort independently. =)

Wow, that wasn't in a nutshell at all, was it now?

[Tyler]

I've been there!

I can't answer for you how to solidify your portfolio confidence, as figuring that out for yourself is an important step. Be sure to share your thought process, as I'm sure that will interest many people.

The one piece of advice I'd offer on your final approach to your retirement goal is to not get too fixated on portfolio perfection. Find something that's "good enough" that you can stick with without worry, and move on to more important things like your savings rate and your personal happiness. Speaking personally, when I felt like I was about that far away from retirement I studied my investments a hundred different ways but felt pretty indecisive. Then I really embraced the MMM and ERE mindset. By focusing on my spending and savings, I cut years off of my plan and greatly increased my confidence. A very high savings rate coupled with the very low volatility of the PP (to grow my money at a healthy rate while minimizing retirement date risk) was just the combination I needed to get me excited and push me over the top.

YMMV, of course. But there's much more to good money management than investment returns, and the other variables are pretty darn powerful and a lot more directly in your control.

[FF9000]

I feel like we are speaking the same language, Tyler. Savings is the primary driver of my quest towards Financial Independence. I am a big fan of MMM. While I can't (or rather, choose not to) live his frugal lifestyle given that I live in NYC, I am about as frugal as possible given my situation and save 50% of my gross salary despite a relatively high effective tax rate. That part actually gets me more excited because I can control it to a much higher extent than I can control market returns.

Off-topic, but it's so funny how everything comes full circle. Last year I actually read "How I Found Freedom in an Unfree World", having no idea who the author was. I was very surprised when I found the Permanent Portfolio completely independently a year later, and connected the dots. Harry is truly a role model in that sense - he shares my philosophy on life as well as my investing goals.

[FF9000]

I have updated the data to be inflation-adjusted. When looking at real returns, you can see that the "decreasing CAGR" for the PP goes away.





I have to say, if I was looking at this objectively, I would have to ask why PP over AW? The only real answer I can come up with is that 40% LTT > 25% LTT, which isn't ideal if rates start to rise. The other answer could be that "gold is going to go up", but that is pure speculation.

[FF9000]

And to answer what would happen if the PP went from 25/25/25/25 to 30 in Equity and 20 in Gold, see Blue vs. Red below.

Real Returns

[stuper1]

With interest rates where they are now, you'd better have an awfully good crystal ball if you are seriously considering putting 40% of your dough in LTTs. I'm going the other way myself toward the Golden Butterfly approach. 20% in cash, gold, and LTTs seems just about right to me.

[Dieter]

>> 15% VFITX (T-bills)

I think this should be:

15% VFITX (Intermediate-term treasuries)

Definitely great stuff -- thanks for sharing your data.

And as per my apparent new Mantra, accounting for emergency fund (cash)....

Or does All-Weather folks say to not keep separate cash / emergency fund on hand?

[MachineGhost]

There's a 300-year Monte Carlo backtest of the PP on the forum somewhere if only the stupid ass search function would actually work. I don't think all of the messages were indexed after the migration. Would someone see if theres a way to force a re-indexing???

Anyway, that's a bullshit Dalio "all weather" portfolio and looks like its been taken from huckster Robbin's epic tome.

I just don't feel like flogging a dead horse, so I'm gonna bow out from hereafter.

[MachineGhost]

Off-topic, but it's so funny how everything comes full circle. Last year I actually read "How I Found Freedom in an Unfree World", having no idea who the author was. I was very surprised when I found the Permanent Portfolio completely independently a year later, and connected the dots. Harry is truly a role model in that sense - he shares my philosophy on life as well as my investing goals.

I just want to point out the Browne evolved from most of the adolescent tripe he wrote in "Unfree World" back in 1973. So it would naturally appeal to you in your late 20's, but you too will grow and evolve and realize how silly most of it was. So take it with a grain of salt. Freedom and money is great, but it has rapidly diminishing returns.

[ochotona]

With interest rates where they are now, you'd better have an awfully good crystal ball if you are seriously considering putting 40% of your dough in LTTs. I'm going the other way myself toward the Golden Butterfly approach. 20% in cash, gold, and LTTs seems just about right to me.

Just as a general rule, 40% in any one asset is too much. It really doesn't matter what that asset is. Swensen maintains that any one class should be between 5%-30%.

[stuper1]

Just as a general rule, 40% in any one asset is too much. It really doesn't matter what that asset is. Swensen maintains that any one class should be between 5%-30%.

Do you think the Golden Butterfly violates this rule by having 40% in stocks split 20% to large cap and 20% to small value?

[FF9000]

The real issue is that we are not able to back-test the PP against periods prior to 1972 due to gold. Even LTT can be back-tested further back, given the 92% correlation between LTT and ITT from 1972-2015 - so ITT could be used as a substitute.

If there were a proxy for gold pre-1972, we could easily compare portfolios over longer periods including rising rate environments. Contrary to the general consensus, total bond returns have not always been negatively correlated with interest rates, and their total returns have been positive in rising rate environments before.

[MachineGhost]

Here's the 300-year Monte Carlo backtest on the PP and nine other portfolios:

http://www.wallstreetcourier.com/v/data ... niques.pdf

This was the paper that got me interested in Most Diversified Portfolio since it outperformed the PP. That is just Correlated Risk Parity under another name and I ultimately decided it still wasn't worth the bother. What I do now is just stick to 25%x4 as a strategic weight and use trend following. Simple is best.

[FF9000]

Here's the 300-year Monte Carlo backtest on the PP and nine other portfolios:

http://www.wallstreetcourier.com/v/data ... niques.pdf

This was the paper that got me interested in Most Diversified Portfolio since it outperformed the PP. That is just Correlated Risk Parity under another name and I ultimately decided it still wasn't worth the bother. What I do now is just stick to 25%x4 as a strategic weight and use trend following. Simple is best.

What is your trend-following system, exactly? Could not find it on the forum spelled out anywhere.

That's an interesting PDF but I didn't see much on methodology for how they got their data, and in the end, it is a marketing piece for their members-only portfolio.

[MachineGhost]

What is your trend-following system, exactly? Could not find it on the forum spelled out anywhere.

It is an enhanced 200-day absolute momentum. InsuranceGuy runs his own PP portfolio using both absolute and relative momentum. I am a headless chicken about downside risk so I focus overwhelmingly on that to the exclusion of almost anything else, even returns. I rather miss out than have a sure loss. The PP can be very risk unfriendly... it had a -26% MaxDD in 1980 if you were rebalancing in April (gold peaked in January and troughed in April) but that was a "Tight Money" environment, the worst of all possible world's for risk assets.

[FF9000]

What is your trend-following system, exactly? Could not find it on the forum spelled out anywhere.

It is an enhanced 200-day absolute momentum. InsuranceGuy runs his own PP portfolio using both absolute and relative momentum. I am a headless chicken about downside risk so I focus overwhelmingly on that to the exclusion of almost anything else, even returns. I rather miss out than have a sure loss. The PP can be very risk unfriendly... it had a -26% MaxDD in 1980 if you were rebalancing in April (gold peaked in January and troughed in April) but that was a "Tight Money" environment, the worst of all possible world's for risk assets.

Right, but what's your exact approach? Do you go all into cash for the 25pct equities allocation if it's under the 200 day sma? Do you check monthly? Etc. Thanks!