Quant Approach to TAA: Equity-Like Returns with Bond-Like Volatility

It is always nice to see a model continue to perform out-of-sample. Below are the results of the "Quant Approach to TAA" paper since it left off in 2005. Equity-like returns with bond-like volatility and low drawdowns.

Obviously, much of this outperformance is due to the spectacular run in commodities and the (relatively) high allocation this model uses at 20% of the portfolio. This does not include June and the additional 5% dump in the S&P500. One could use five simple ETFs to replicate the asset classes mentioned in the paper (VTI or SPY, VEU or EFA, BND or AGG, VNQ or IYR, and DBC or GSG).

 

Comments

[quaazy1]

Not sure if I did this right, but I show a lower return for the timing model on the S&P using a 10 month trailing SMA for data from 1/1926 to current... What month did the S&P timing model signal a move to cash?

[quaazy1]

Let me restate the question... What was the most recent month that the S&P timing model signaled a move to cash?

[rayvt]

End of month, Nov 2007, for SPX/SPY.
The signal before that was end of month Sept 2004, which signalled to buy SPX.

[quaazy1]

Sorry to be so basic, but... Does that mean that your model took the
-4% hit through November??? I can't get the math to work out anywhere near what your paper suggests on a cumulative basis unless I trick the model to get out of the market BEFORE the Nov 2007 drop of -4%, for example (working back through history...)

[winyaz]

Hi,
Your timing model is interesting and I appreciate your articles on it, the endowment strategies, etc. My only concern is that in a falling market 1 month seems like an eternity to bail out of a position. Checking the 40 week sma could get you out quicker and would do much better in trending markets ... but much worse in flat markets - PCRIX would have been whipsawed repeatedly from 2004 to 2007.

Have you considered different approaches or timeframes? Perhaps weekly signals but with a filter to try to limit the whipsaws? I'm interested in thoughts on this subject.

Don

[rayvt]

quaazy1,
Yes the model had you in during November and took the 4% loss in November. BTW, I'm not Mebane--he doesn't seem to visit the comments very often. Hopefully he will be able to answer you.

winyaz,
The problem with all of life is that what's happening RIGHT NOW is what we experience, and we therefore focus too much on "now" and too little on "overall history". So we focus on today's 4% loss that it didn't get us out of, and ignore the 10% drop that it did avoid but that happened years ago.

I think Mebane has said that different time periods (from 6 months to 12 months) all had slightly different "best" category of results. I think that the exact choice of period is not important--that any period in that broad range is as good as any other period. They are all essentially the same, just different details.

The danger with adding filters & other signals is that you'll be data-mining or "over-tuning"... I've been down that road and learned my lesson. A simple filter that gives an increased risk-adjusted return is prefereable to a complex set of rules that gives a very high return -- but works only for one set of historical data.

[Indexor]

quaazy1,

Go up to the top of this page and under Mebane's picture is a section for "Bio and more articles." About 6-7 articles down you'll find the original article that this post is updating. It is titled: "A Quantitative Approach to Tactical Asset Allocation." It may just change the way you invest.

Although I've never had the pleasure of meeting Mebane, I've used a very similar system for well over 10 years. I found it uncanny that his article proved the results I've been able to achieve. As Mebane states: "The empirical results are equity-like returns with bond-like volatility and drawdown, and over thirty consecutive years of positive returns." Believe him and thank him for sharing his research.