Be Careful With Backtesting

We have a lot of fun here with backtesting portfolios. The process can be informative but not definitive. If you are going to include backtesting in your investment process, it is important to understand the limitations beyond the boilerplate that past performance doesn't guarantee future results.

There was news going around that Daffy.org was adding Bitcoin in 5% or 10% weightings to a couple of its portfolios, presumably to improve returns for otherwise "normal" looking portfolios. Along the lines of Taleb's barbell strategy of getting most of a portfolio's return from a narrow slice of the portfolio and putting the rest in very conservative holdings, I wanted to model in Daffy's 5-10% weighting idea to try to get a market equaling result with just a small allocation to risk assets. 

You can see Portfolios 2 and 3, Portfolio 1 is just 100% Vanguard Balanced Index (VBAIX) which is a proxy for a 60/40 portfolio.  

The results are thought provoking but, and there is a big but, it is built on a couple of phenomenal years for Bitcoin. In 2017, Bitcoin went from about $1000 up to $5000 and in 2020 Bitcoin went from about $9000 to $34,000. That could happen again of course but building a portfolio modeling that sort of gain, without realizing there might be unrepeatable is one flaw of modeling. It is important to understand at least some of what drove returns. An uncharacteristically great year can throw off the entire backtest.

If Bitcoin had traded exactly like  T-bill then Portfolios 2 and 3 would have compounded below 3%. The way to deal with this IMO would be to allocate a little less to Bitcoin and have at least some exposure to plain vanilla equities. If Bitcoin does go up 5-fold in one of of the next ten years, great, but if it didn't, the portfolio could have a decent CAGR even if it was less than VBAIX.

Making the above changes, replacing the AQR fund with VOO and shaving Bitcoin down to 2% gave a CAGR of 8.16% versus 8.13% for VBAIX but the standard deviation was 5.59% versus 10.10% for VBAIX.

I saw a reference to Meb Faber's Ivy Portfolio which I believe he first published in 2011. It allocates as follows.

It looks like a derivation of Harry Brown's Permanent Portfolio which allocates 25% each to equities, long bonds, cash and gold. Here's a comparison of the two along with 100% in VBAIX.

The Ivy had the lowest return and the highest standard deviation. I don't doubt whatever work was done on the front end to create the Ivy but if I wrote about it when the paper first came out, I would have bagged on it pretty hard. I have never been a fan of the huge percentages into real estate and commodities that some folks swear by. For a while in the 2000's, REITs did very well and there were a lot of calls to put 20-25% into REITS and commodities too for that matter. That felt like a terrible idea to me back before the crisis and then all the way through. 

People hope REITs will offer protection against equity volatility. The history of that isn't very good. In 2008, Vanguard REIT ETF (VNQ) was down 37% and in 2022 it was down 26%. Having exposure to a REIT or two, sure, why not if there is one or two that you like but a huge percentage just isn't going to work out. I believe commodities do offer some protection against equity volatility because the correlation is low to negative. A point we make here all the time though is that if equities are the thing that goes up the most, most of the time, why would you want a huge allocation to something with a negative correlation to the thing that goes up the most, most of the time? 

Some exposure to negatively correlated assets should smooth out the ride, yes but Ivy is more volatile for having too much in commodities, IMO. This gets us to another flaw of modeling which is tunnel visioning on how something should work versus the real world. I don't know if my belief in small exposures to diversifiers is necessarily forward looking, it is from the standpoint of trying to avoid being over exposed to future calamities but I just had not seen REITs offer the protection that some people thought they did back almost 20 years ago which was backward looking. 

Backtesting and modeling are useful tools but don't lose sight of the forest when you're looking at the trees. 

The information, analysis and opinions expressed herein reflect our judgment and opinions as of the date of writing and are subject to change at any time without notice. They are not intended to constitute legal, tax, securities or investment advice or a recommended course of action in any given situation.