All Weather Portfolio Lab

Larry Swedroe had an interesting article at FA Mag about how to size alternatives. The interesting part was the talk of equal weighting various factors including kurtosis (a measure of tail risk) and standard deviation. Larry listed out more but I think those two cover a lot of ground. 

I started at Copilot and then also brought in Claude using Invesco Momentum (SPMO), SPDR Gold (GLD), Saba CEF ETF (CEFS), AQR Managed Futures (AQMIX), Convertible Arbitrage (ARBIX) Stoneridge Catastrophe Bonds (SHRIX). I ask both Copilot and Claude to "build a portfolio with the following funds by equal weighting standard deviation and kurtosis. Do the best you can if its not possible to do it exactly." Here are the weightings they each gave;

The next step was to ask each one to grade the concept for "all-weather robustness" because the results are interesting and the time is not so short as to be useless.

Copilot gave it an A- thinking it would benefit from more explicit inflation and deflation hedges than what I used. Claude was a little tougher, it gave a me a B- with a lot of color that led me to believe it thought Dalio's All-Weather was better. This made no sense to me so I pushed back, "the portfolio I asked you to build and then evaluate outperformed Dalio's all weather allocation by more that 400 bp compounded, had smaller drawdowns than Dalio's all weather allocation and a lower volatility. How do you figure that Dalio's all weather allocation is superior? That doesn't make sense to me, what am I missing?" 

The reply to that was very interesting. Claude said I was right to "call me out" and gave the following;

I think I see this sort of thing frequently where people overly rely on the "conventional wisdom" of bonds and get hung up on how things should work. Claude bumped the grade up to A. As a follow up, Claude asked how I backtested and I told it that I just used testfol.io, compared it a version from Copilot and Dalio and how similar its results were to Copilot. It replied to that saying that fund selection matters more than weightings which is interesting up to a point. 5% in everything and the rest all in ARBIX obviously would have been far inferior to the final result. 

One interesting item from the interaction with Copilot. I plugged in the symbols and then asked to figure the weightings. I made a typo on one of the funds. I typed AQRIX which is sort of risk parity instead of AQMIX which is managed futures. I told it I made a mistake and asked it to rerun with AQMIX which it did. It also told me that AQMIX made much more sense than AQRIX which I thought was funny. 

This was fun and productive. I don't really have the mathematical chops to have figured this myself and tweaking weightings over and over to get close would have taken a long time versus a couple of minutes of both AIs "thinking." 

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.