Checking In on the All-ETF Portfolio 21 comments
-
Font Size:
-
Print
- TweetThis
In June of 2007, I published an article on how to design an all-ETF portfolio using a modern approach to asset allocation. The premise of this article was the following:
Let’s imagine that you want to build a portfolio of ETFs but want to do something better than just following a generic ‘pie chart.’ What does it mean to do something better than the pie chart approach? First, you want to see if you can get more return for the total amount of risk in your portfolio—and that means that you have to be smart about diversification. Second, you may have some other broad themes in mind. One of the key themes that I think about is how coupled I want my portfolio to be to the S&P500, for example.
What does it mean to be “smart about diversification”? The main thrust of this article—and of many of my articles—is that diversification benefits are measurable. Using a good portfolio planning tool, investors can develop portfolios with higher returns (without increasing risk) than will be possible simply by arbitrarily choosing an asset allocation (i.e. a ‘pie chart’ portfolio). I then proceeded to propose the following portfolio as an example of this theory put into practice:
While most portfolios start by specifying asset allocations based on some conventional rules of thumb, this portfolio is based entirely on guidance from Quantext Portfolio Planner [QPP], a portfolio planning tool that assists in identifying portfolios that provide the most diversification benefit. This process is explained in the original article. The portfolio that results is unlike the standard pie charts. There is no allocation to “total market” kinds of indices like the S&P500 or the EAFE index. The majority of this portfolio is focused on individual sectors, with the single largest sector being utilities. This portfolio has 30% allocated to bonds, but this is spread between TIPS and high-yield bonds—there is no allocation to a broad bond index like MSCI Aggregate.
The reason that this portfolio does not have allocations to broad market-cap weighted “all market” index funds is that they do not provide superior diversification benefits. Back in fall of 2006, I wrote an article showing that the EAFE index (EFA), and the NASDAQ 100 index (QQQQ) were both correlated to the S&P500 (SPY) at levels greater than 80%.
This portfolio leaves out allocations to broad emerging markets indices such as the MSCI Emerging Markets Index for the same reason. For years, many advisors and portfolio managers had been touting what was called the “de-coupling” between emerging markets and the U.S. markets. This effect meant that emerging markets (and developed foreign markets) would not be dragged down by a big decline in the U.S. I simply never saw evidence for “de-coupling,” which is why this portfolio has no allocation to funds like EEM.
Higher levels of correlation mean lower diversification benefits. I noted that energy and utilities, by contrast, provided a high level of diversification benefit relative to domestic equities. For more details on how this portfolio was designed, please see the original article.
The period since the all-ETF portfolio article was published has provided a good stress test for any asset allocation. The S&P500 is down about 13% for the period from June 2007 through July 2008.
QPP’s projections for this portfolio (from the original article) were the following:
The Portfolio Stats section shows the long-term QPP projections for this portfolio published in the original article, while the cells below show historical data for three years through May 2007. Several points are notable. First, the three years through May 2007 had been good for broad equity indices, with high returns and low volatility. The S&P500 had average more than 10% per year (even before we include dividends), and the annual standard deviation of the S&P500 (the standard measure of volatility) was about 7%, less than half of its average level over recent decades (see table above).
QPP showed that even this All-ETF portfolio had generated returns above its long-term expected level (i.e. trailing three year return of 14.9% vs. projected average annual return of 10.5%). Further, the projected portfolio volatility (the standard deviation) was 11.2%, as compared to 5.1% over the trailing three years (see table above). This portfolio, while good in the long-term, was due for some correction in the near term (as was the S&P500). No portfolio can out-perform forever: a trailing return greater than the projected average return indicates some potential for reversion to the mean.
The original article used data through May 2007, but the article was published in mid-June. To avoid any overlap in available data, we will look at the portfolio performance from July 1, 2007 through July of 2008. For the period from July 2007 through July 2008, we get the following results:
The All-ETF portfolio has beaten the generic 70/30 portfolio by almost 6% in average annual return (-0.5% for the All-ETF vs. -6.3% for the 70/30), with less volatility, in the 1+ year period since this article was written.
In a fairly recent article, I suggested that a really well diversified portfolio with a standard deviation of about 10% (forward-looking, not trailing) could be expected to generate about 10% in return per year. This performance could be compared to a generic 60/40 stock/bond mix which would be expected to generate about 7.5% per year in average return for the same level of volatility (standard deviation of about 10%). This result implied a “diversification premium”—the benefit of better diversification—on the order of 2.5% per year for portfolios at this risk level.
We are looking at generally similar cases here (albeit only 30% bonds), but we can draw some qualitative comparisons. If we figure a 2.5% per year diversification premium, and the all-ETF portfolio has generated almost 6% greater return over the 1.1 year period, we can see that part of the out-performance is simply due to chance—and is not expected to persist in the long-term.
So how do things look in QPP as we go forward? I ran the All-ETF portfolio through QPP using the default settings (including three years of trailing data to initialize the model). The results are shown in the table below.
The projections from QPP are given in the Portfolio Stats table. The projected long-term return for this portfolio is 10% per year (close to the original projection). The projected volatility is a bit higher than the original calculation. Note that Reversion-to-The-Mean [RTM] is now on our side—the projected average annual return is greater than the trailing average annual return.
We could certainly adjust this portfolio so that it would have a higher projected return to projected risk, but the point of this article is to provide a look at how this portfolio has held up in the midst of a broad market decline.
These results are not the result of superior timing or any insight into how any sector would perform. These results simply demonstrate the power of portfolio theory, applied properly. While many investors have never really been exposed to this perspective on how to develop portfolios, there is considerable evidence of the effectiveness of this approach. This article may serve as one more data point—albeit a small one. I am confident, however, that this portfolio will provide consistently solid long-term performance so that when I revisit it again at a later date, the same general conclusions will hold. As a final word, I will stress that this portfolio is not a recommendation—it is simply an example. Better portfolios can be built, and every investor must carefully consider whether any portfolio is appropriate for their specific needs.
Related Articles
|
-
The Seeking Alpha 100
See allThe top 100 stock
market authors
selected for publication
Seeking Alpha 100 » -
Fastest Climbers
Authors with the largestSee all
increase in followers
in the last 7 days
Fastest Climbers » -
Top Commenters
Top commenters,See all
as ranked by their peers
Top Commenters » -
Top Instabloggers
The 100 most active
Instabloggers.-
1
TraderMark
-
2
Don Dion
-
3
Cliff Wachtel
-
4
Mike Havrilla
-
5
James Quinn
Top Instabloggers » -
1
-
Top StockTalkers
TopSee all
StockTalkers.
Top StockTalkers »
This article has 21 comments:
I'm not a big fan of Nassim Taleb. He is a terrible writer who is guilty of some egregious rhetorical fallacies. But his main point, that MPT has a terrible time coping with the unknown and that it is this kind of uncertainty that is the biggest source of investment losses, is spot on. That QPP would include Malaysia in a hypothetical best risk-adjusted portfolio is a good example of Taleb's point.
Alisha you might want to read that article before adding Emerging Markets for purposes of diversification.
You can perform backtesting in QPP--and I have published a range of articles on backtesting here at SA and on my website. I am increasingly looking at writing articles where I go back and see how the forward-looking analysis has performed in real life--and this is one such article. I like SA in part because it gives readers a real time stamp tocheck if I really said something when I said it! I find it notable that so many people are now writing about the fallacy of the "decoupling" argument after the fall--but where were these analyses before?
I address this specific Taleb-type argument in an article written in Feb:
seekingalpha.com/artic...
People who point out these individual events and say that portfolio theory can't "predict" such events are really missing the point. There are events like 9/11 that are impossible to model in the short term, but increased perception of risk in markets (via volatility and implied volatility) is a very useful signal with considerable information content. Anyway, for the detailed discussion of Taleb and model considerations, black swans, etc., see that article.
I enjoyed the original article and am glad to see the portfolio held up well under pressure. I think this sort of statistical approach is much more useful the majority of analysis on this site, which is too often based on emotion and short term trends. Have you tried using this methodology to build a portfolio that places more emphasis on return and less on standard deviation? For the long-term buy and hold investors who have time to wait out the swings, such a portfolio would be preferable. Thanks for a great article!
If we have [$BLACK_SWAN], the short-term losses to any portfolio are likely to be far greater than Quantext Portfolio Planner or any other portfolio model can estimate. This kind of massive disruptive event is not what these models are capable of estimating.
...
Ultimately, portfolio models are perhaps most useful because they give investors objective tools to:
1. examine diversification benefits in their portfolios
2. estimate total portfolio risk (albeit not very well for market crashes, wars, epidemics, etc.)
3. estimate total portfolio returns in light of assumptions
It is point number two that I was trying to highlight. MPT is all about creating portfolios with the best return for a given amount of risk. But if we are doing a poor job of estimating risk, then it is not at all clear that we are, in fact, maximizing expected returns per unit of risk. Tools such as QPP may thus be giving us a false sense of security when they recommend, say, a concentrated position in Malaysia over a larger more diversified basket of emerging market stocks.
Yes, but..you have taken a brief statement from a longer point. In the short term, portfolio theory models will under-estimate the probability of "black swan" events. If you expand the time horizon--even including these events--the models do well. If you are a short-term trader who needs to model the amplitude and duration of daily-monthly fluctuations in extreme one-off events, this is not an ideal model.
With regard to systemic risk in Malaysia, why are you so confident that this risk is not priced into the implied volatility? you MAY be right, but what is the evidence? This actually relates quite closely to the work that I did on credit ratings and implied volatility (also on SA).
Geoff
Yes, the MPT, APT and follow-up portfolio theories rely on normal (arithmetic) distributions, henceforth ignoring fat-tail events. Converting to a stable (log) distribution exposes the tails to more accurately define the actual risk.
These traditional models rely on mean-variance which is why they predict 10% annualized returns (based on the past 80 years. It is ~ 7% over the past 180 years). Using mean-variance, your clients should have doubled their money since 1998; but in fact the S&P 500 is still down after 10 years. Making a bet on the law of large numbers is great if you have 50 years for a pay-off; I don’t. Markets move through economic cycles that reward equities positively on average 18.5 years (bullish) and penalize equities negatively on an average 17 years (bearish). It can take up to 20 to 30 years for an investor to break-even if invested at the market highs.
An upgrade can be made to mean-variance with a newer Noble prize winning formula (circa 2003) called GARCH (Generalized Auto-Regressive Heteroskedacity) that tracks the clustering of data.
The third problem with traditional models is that they assume markets are static as well as correlation. We all know that correlations increase as markets become more volatile. Why not have real time correlation models? You can upgrade linear correlation to a dynamic correlation by switching to a copula dependency model.
Upgrade your 1) risk analysis using stable distributions, 2) time-series using GARCH, and 3) correlation dependency using copula dependency. These upgrades significantly add to performance and are the underlying concepts of Extreme Value Theory.
First, Mandelbrot wrote his book four years ago and we had already built the same concepts several years before. So the fact he hadn't seen the models work doesn't mean they don't exist.
Second, I like your idea’s but I’m highlighting the weaknesses in your models and offering improvements to help; not to be critical. You choose to indirectly advertising QPP so I choose to freely suggest how models can be improved.
So the question is:
a) Do you believe normal distributions are better than stable distributions? I welcome this discussion.
b) Do you believe linear correlation is better than copula dependency? I welcome this discussion.
c) Do you believe a concept from 1952 called mean-variance can possibly be better than another Nobel winning formula created this decade? I welcome this discussion as well.
I’m not advertising our firm but will share our asset allocation mix as of July 22, 2008, the date of our last rebalancing.
2.00% Vanguard Health Care VIPERs VHT
2.00% Vanguard Industrials VIPERs VIS
5.00% Short QQQ ProShares PSQ
3.00% iShares MSCI Brazil (Free) Index EWZ
2.00% iShares FTSE/Xinhua China 25 Index FXI
3.00% iShares S&P Latin America 40 Index ILF
2.00% BLDRS Emerging Markets 50 ADR Index ADRE
3.00% PowerShares DB Agriculture DBA
2.00% iShares Dow Jones US Real Estate IYR
3.00% iShares GS $ InvestTop Corp Bond LQD
71.50% iShares Lehman 1-3 Year Treasury Bond SHY
1.50% Money Market Fund MMA
Because asset allocation models all follow a 3 step process I will describe the advantages in each step:
1) Step 1 (univariate model): a Stable ‘t’ distributions with GARCH features better determines the current risk & forecasted return of a security as opposed to a rear-view mirror approach based on normal distributions.
2) Step 2 (bivariate model): a copula dependency more accurately examines the current relationship between two securities as opposed to linear correlation based upon long-term averages.
3) Step 3 (multivariate model): ranking the bivariate models with Monte Carlo modeling to forecast produces an optimal mix based on current market conditions as opposed to Monte-Carlo modeling based upon long-term averages.
So you can see from our current allocation the risk/return trade-off for securities is unattractive in the current market conditions and the recent volatility is keeping investment out of securities which have traditionally low or negative correlation.
So you see I have only upgraded the basic attributes of asset allocation and continue to follow the same 3-step process as you and the other solution providers.
Again, I wasn’t trying to change the conversation over to EVT but was trying to address the blindness in risk modeling and offering a solution to the problem.
All the best -
You're web site is not working.
No model is perfect and there are many ideas for better models. A forum like SA is a great way to show what a model says and then to go back and revisit it later to see how it did.
I do favor simpler models when they work and QPP has held up very well for what it is designed for--and this article is a case in point. The ultimate test of statistical models is that they work operationally.
If you have an operational model that is of value, I am sure that many people here would be interested in what it says. But how do we establish value? In a broad forum, it is by publishing positions / suggested allocations that can be benchmarked later.
There are so many complexities in how even a model like QPP deals with stationarity and correlation that this is probably not the forum. QPP does not assume stationary statistics or linear correlation, for example. I often show correlation matrices but QPP is not a so called Delta Normal Mean Variance model. I am not sure where you got that idea.
Geoff
I am not sure what you are asking here...
Geoff
I have read your past posts and we obviously differ in several areas; but only because I’ve been in the business 25 years, have built models since the mid 80’s that have put me in the top 1% for two decades, and mostly read whitepapers, not trade magazines. In the future I would welcome a debate on your previous posts. In previous articles you cited LTCM as the exception, not the rule, I disagree. You note the underperformance of active managers, yet who determines who is active or who is a closet indexer? Two Yale professors recently came out with a new model that determines if a manager is truly active (based on 23 years of research). Truly active managers soundly beat the index after fees and expenses! BTW, I do not consider active managers to be the same thing as market timers. Taleb may be egotistical but he is right none-the-less.
I will love to resume this conversation in 10 years after you’ve experienced what is to come. My word of caution is to avoid absolutes, avoid getting sucked into the academic theories, and be careful who you choose to debate. I’ve chosen to fight the popular opinion my entire career; it’s a tough road. My best word of advice is for you to never stop asking ‘why’. As Kennedy once said “How could I have been so wrong as to trust the experts?” Don’t let a little criticism stunt your progress. I applaud your efforts and cherish your enthusiasm.
It would be far more helpful if you published one or more articles about what you think is the best approach, how it works, back tests, etc. Simply claiming that your model is better and that you have out-performed by some measure over some period of time is not compelling. Also, publishing a portfolio and then revisiting it later and seeing how it has performed--as I did here--is a useful data point. QPP is consistent with the opinions of some of the very best minds in institutional research (Roger Ibbotson, David Swensen, etc.). This does not make it right, of course, but it is meaningful in my opinion. There may be better models in the world, but there are none that I know of that are as well documented both in stress testing and in practice.
Geoff
Thanks for the link to the article. I never would have come to the conclusion that the dealings we do here have such a STRONG impact on business abroad. In relation to S&P500, my favorite ETFs for diversification, ADRE and EEM have an r of no less than .6 over the last five years. Thats quite a pitiful performance. However, as a nonactive trader, I still like the idea of global markets ETFs for my portfolio purposes, and Geoff's article and the comments that ensue, are quite helpful and entertaining.
On Aug 04 05:21 PM Delta David wrote:
> Emerging Markets = Diversification? www.indexuniverse.com/...
>
>
> Alisha you might want to read that article before adding Emerging
> Markets for purposes of diversification.