Seeking Alpha

varan's  Instablog

varan
Send Message
Individual investor.
  • Adaptive Asset Allocation For Lazy Portfolios 33 comments
    Jul 30, 2012 2:48 AM | about stocks: SPY, EFA, TLT, VNQ, IWB, DBC

    Selection of a few of the ETFs from a basket on the basis of prior performance, together with allocation on the basis of risk parity, may improve the performance of some of the so-called lazy portfolios (see, e,g., seekingalpha.com/article/27167-lazy-etf-portfolios-inspired-by-the-gurus ).

    We consider the following portfolios (we note that we have used TLT for the bond components of the portfolios, and this may not conform exactly to the original recommendations).

    PortfolioComponentsWeights
    IvySPY, EFA, DBC, VNQ, TLTEqually weighted
    SwensenSPY, EFA, VNQ, TLT30%, 20%, 20%, 30%
    BernsteinSPY, IWB, EFA, TLTEqually weighted

    The performance of each of these portfolios starting in the year when all the component ETFs were first available is shown in the following table (the last column lists the minimum of the annual returns during the period).

    PortfolioPeriodCAGR

    Min

    Ivy2007-20124.3%-22%
    Swensen2005-20127.4%-16%
    Bernstein2003-20128.6%-20%

    For adaptive asset allocation we update the portfolio every two months as follows:

    1. Find the two top ETFs whose performance was the best during the prior three months.

    2. If any of the two top ETFs performed worse than TLT, replace it by TLT.

    3. Find the weights of the two top ETFs using risk parity and the daily return history of the ETFs during the prior quarter.

    Note that for only two assets the risk parity problem is quite easy to solve: the weights of the assets are inversely proportional to the standard deviation of the daily returns during the prior quarter.

    The performance of this strategy applied to the lazy portfolios is shown in the table below.

    PortfolioPeriodCAGR

    Min

    Ivy2007-201217.6%-0.77%
    Swensen2005-201214.1%1.3%
    Bernstein2003-201213.1%-3.3%

    The annual returns of these portfolios during the recent years are shown in the following table.

    PortfolioYTD1 Year3 Years5 Years
    Ivy14.4%31.9%18.5%18.9%
    Swensen14.4%31.9%20.4%14.3%
    Bernstein11.6%37.1%18.8%13.5%

    It is quite noteworthy that the Morningstar Fund Selector (screen.morningstar.com/FundSelector.html) does not yield a single mutual fund that beats all of these four recent returns of any of these lazy portfolios. Surprisingly, none of the portfolios at Foliovesting (www.folioinvesting.com/rtg/ready-to-go-folios-list.jsp ) can beat all of the metrics of any of these lazy portfolios with adaptive asset allocation.

    Clearly, there is a slight improvement in the performance at the cost of twelve round-trip trades per year. Of course for taxable accounts there will be additional considerations.

    Disclosure: I am long VNQ.

    Themes: Adaptive Asset Allocation Stocks: SPY, EFA, TLT, VNQ, IWB, DBC
Back To varan's Instablog HomePage »

Instablogs are blogs which are instantly set up and networked within the Seeking Alpha community. Instablog posts are not selected, edited or screened by Seeking Alpha editors, in contrast to contributors' articles.

Comments (33)
Track new comments
  • David Jackson
    , contributor
    Comments (1277) | Send Message
     
    Very interesting!
    30 Jul 2012, 12:49 PM Reply Like
  • littletiger101
    , contributor
    Comments (9) | Send Message
     
    thank you varan. finally you have a new update.
    If I may speculate a bit, you do have spreadsheet available online to show the back-testing. If you do, can you please show us the link?
    30 Jul 2012, 11:19 PM Reply Like
  • Stephen Harlin, MD
    , contributor
    Comments (16) | Send Message
     
    Also, Varan ...Faber's "Endowment" allocation had the best performance. How might your strategy impact that allocation? Thanks, again.
    14 Aug 2012, 01:56 AM Reply Like
  • Stephen Harlin, MD
    , contributor
    Comments (16) | Send Message
     
    Varan, this looks smart. I would second that request for a link to a googledoc spreadsheet. And I'm not clear on point 3, above, regarding weighting. You wrote, " 3. Find the weights of the two top ETFs using risk parity and the daily return history of the ETFs during the prior quarter." Would you mind being a little more clear on that? Grateful in advance. Stephen
    14 Aug 2012, 02:04 AM Reply Like
  • varan
    , contributor
    Comments (4532) | Send Message
     
    Author’s reply » @harlin

     

    1. I am using only the Endowment ETFs, and not his allocation. Allocation is determined by risk parity - see below.

     

    2. Compute the standard deviation of the prior quarter's daily returns of each of the top ETFs . If the standard deviation of the first ETF is A, and the second one is B, then the weight of the first ETF will be B/(A+B), and the second one will be A/(A+B). This is derived from the fact that for two assets, risk parity is achieved by setting the weights as inversely proportional to standard deviations - you can write the two weights as [AB/(A+B)](1/A) and [AB/(A+B)](1/B).
    14 Aug 2012, 09:17 AM Reply Like
  • Stephen Harlin, MD
    , contributor
    Comments (16) | Send Message
     
    Great white paper on the "How To" and why for Adaptive Asset Allocation here:

     

    http://bit.ly/O62vYy
    14 Aug 2012, 09:22 AM Reply Like
  • Stephen Harlin, MD
    , contributor
    Comments (16) | Send Message
     
    Your reply much appreciated.
    14 Aug 2012, 09:30 AM Reply Like
  • Stephen Harlin, MD
    , contributor
    Comments (16) | Send Message
     
    Varan,

     

    I've now reviewed all of Butler and Philbrick's work (white paper and published research) on Adaptive Asset Allocation and fully agree with their perspective on risk parity.

     

    Briefly, their "Darwin" Adaptive Asset Allocation is summarized by:

     

    The investment universe is comprised of the 10 major global asset classes. These include:

     

    Barclays 20+ Year Treasury Index
    Barclays 7-10 Year Treasury Index
    DB Liquid Commodities Index
    Dow Jones Global Real Estate Index
    Dow Jones U.S. Real Estate Index
    Gold
    MSCI Emerging Markets Index
    MSCI Japan Index
    S&P 500 U.S. Stock Mkt Index
    S&P Europe 350 Index

     

    At the end of every month, each asset class is ranked for near-term trailing returns. The portfolio only invests in the two top-ranked assets (equities and bonds) that month. (Five percent of net equity is held in cash or cash equivalents.)

     

    Allocation among equities and bonds is then derived from a minimum variance algorithm.

     

    Portfolio level volatility is managed by reducing position size when volatility exceeds a predetermined risk level. In addition, allocations are reduced in assets exhibiting erratic behavior while increasing allocations to assets exhibiting more predictable behavior.

     

    The portfolio is rebalanced monthly following near-term performance profiles.

     

    The authors' white paper show CAGR values significantly higher than the three gurus.

     

    - - - - - - - - -

     

    I am able to duplicate their core strategy's allocations, using a minimum variance algorithm, and I calculate expected returns which are quite close to their (published) actual returns. The strategy has many attributes, but frankly, I believe it could be improved a little bit.

     

    Notwithstanding, Butler and Philbrick's work and performance measures are the nicest I've seen using Adaptive Asset Allocation models.
    15 Aug 2012, 11:35 AM Reply Like
  • varan
    , contributor
    Comments (4532) | Send Message
     
    Author’s reply » This is not wholly surprising, since the universe of assets is larger and more comprehensive. My purpose was just to show that minor modifications can improve the performance of these portfolios. Similar improvements occur when you use this method for the lazy portfolios maintained at marketwatch.com which consist mostly of vanguard mutual funds and have been around for longer periods. I don't know about the transaction costs at vanguard, but it may well be that in tax free accounts one can get the performance improvement essentially for free with those portfolios. I will try to replicate my analysis for the Darwin portfolio, as I suspect that there might be room for some improvement by changing some parameters.

     

    Your use of minimum variance in the context of risk parity is quite intriguing. Generally one expects that in contradistinction to risk parity, the classical minimization of variance will lead to weights that may not include all the top assets unless you impose some additional constraints.

     

    Thanks a lot for your interest.
    15 Aug 2012, 12:18 PM Reply Like
  • varan
    , contributor
    Comments (4532) | Send Message
     
    Author’s reply » I have done an experiment with the portfolio of IEF, TLT, DBC, RWO, IYR,EEM, GLD, EWJ, SPY, and IEV.

     

    If you update the portfolio every 3 months by buying the top three of the month prior to the update, and compute the weights on the basis of standard deviations of the returns during the prior month, you get a CAGR of 17.9% during 2004-2011.
    17 Aug 2012, 01:33 AM Reply Like
  • Stephen Harlin, MD
    , contributor
    Comments (16) | Send Message
     
    Cool. Butler and Philbrick are publishing CAGRs in the 15 - 16% range, rebalancing monthly. IMHO, their principals are spot on; the strategy can be improved, however.

     

    For example, there are instruments which behave in bond-like fashion (in terms of mitigating risk), have negative correlations to equities, low realized volatilities, high liquidity - but are optionable and pay respectable dividends. Utilities are one. Volatility as an asset class is another.

     

    By pairing the equity side with optionable (but bond-like) instruments, one could increase income writing covered calls, further reduce risk with (costless) collars, initiate positions via writing puts, capture income with weekly option plays on volatility mispricing...etc.

     

    I'm currently working on models which take the two asset portfolio approach I described above ...use the minimum variance algo and the principals espoused by Butler and Philbrick ...but improve on what they call the "Portfolio Universe." Also, they are "anti-precise" and scoff at market timing for entry and exit.

     

    Timing has been shown to: 1) outperform (anti-precise models) 50% of the time, 2) improve risk-adjusted returns 66% of the time, and 3) reduce drawdowns 95% of the time by 50% (Faber, et. al.), 4) keep the investor invested 70% of the time, holding losses 15% of the time long, and 5) keep losers smaller than winners (M. Faber, et. al.).

     

    So I think that by ...keeping Butler and Philbrick's principals involving momentum, volatility, and correlation ...but widening the portfolio universe and systematically timing entries/exits ...one might improve on their already sound adaptive asset allocation model.

     

    (I have coded a calculator to obtain relative weights for a two asset allocation, embedding the minimum variance algo, and can simultaneously calculate portfolio volatility and expected returns.)

     

    One last note: Butler and Philbrick use IEF/TLT to mitigate equity risk. Between the two, IEF functions far better than TLT. During the last year, there were only 72 trading days when you'd be glad you held TLT, based on correlation and strength. In other words, TLT is usually a drag on performance.
    17 Aug 2012, 11:48 AM Reply Like
  • ATraderJournal
    , contributor
    Comments (2) | Send Message
     
    Varan, by any chance do you still have the spreadsheet that does these calculations. My intent is to use your spreadsheet and following same steps/calculations, do it in R. I would be happy to share the R program later with your readers. Regards.
    17 Aug 2012, 05:03 PM Reply Like
  • Max1985
    , contributor
    Comments (11) | Send Message
     
    Hi Steph,

     

    The paper you linked above only says top half by momentum. Where is the top 2-ranked assets part? Can you please share the link to the white paper?
    25 Aug 2012, 01:45 AM Reply Like
  • Jonathan^n
    , contributor
    Comments (5) | Send Message
     
    But RWO and DBC dont go back to 2004. So, from 2008 on, I get ~6%. Whats up?
    28 Aug 2012, 04:20 PM Reply Like
  • varan
    , contributor
    Comments (4532) | Send Message
     
    Author’s reply » DBC starts in 2007, and RWO in 2009.

     

    During 2007-2012 I get 13% CAGR. 2009-2012 yields 14%. But for both, much of the returns comes from the 32% of 2009. More specifically;

     

    2007 34.2%
    2008 9.3%
    2009 32.2%
    2010 9.9%
    2011 7.1%
    2012 4.3%
    28 Aug 2012, 04:44 PM Reply Like
  • Jonathan^n
    , contributor
    Comments (5) | Send Message
     
    So your basket does not contain RWO in 2008, 2009 and you add that in along the way?

     

    No wonder my results havent been matching...
    29 Aug 2012, 03:36 PM Reply Like
  • varan
    , contributor
    Comments (4532) | Send Message
     
    Author’s reply » yes.

     

    Whenever it is available it is used, otherwise not.

     

    thanks.
    29 Aug 2012, 04:18 PM Reply Like
  • ATraderJournal
    , contributor
    Comments (2) | Send Message
     
    Interesting strategy and comments. Thanks. In recent years pair stocks and bonds have low correlation. On other hand in 90's I think bonds and stocks have much higher correlation. Another example would be current high correlation scenario in Spain between bonds and stocks. So it might be useful to have correlation level as one of the factors.

     

    Adaptive asset allocation is an interesting paper. I tried to recreate but couldn't get similar results. So currently searching to see if there is an excell/R implementation on net. I agree with their perspective on risk parity. On other hand I think performance cost associated with volatility targetting as described in paper might be more useful to planners then individual investors. Reason being for financial planners/fund managers, the clients treat the amount they see in month end statement as cash in bank even if though it is likely an open equity. So for planners/managers it is important that open equity curve is smooth even if it costs little bit on performance side. On other hand, someone investing their own money likely be ok with ignoring open equity and use closed equity as cash in bank.

     

    Regards,
    Contact-(http://bit.ly/OCZqow)
    17 Aug 2012, 12:18 AM Reply Like
  • Stephen Harlin, MD
    , contributor
    Comments (16) | Send Message
     
    UBS just rebalanced their global asset allocation recommendations, increasing weights in equities and bonds:

     

    45.1% Equities
    27% Government bonds
    11.6% Corporate bonds
    7.3% Commodities
    5% Real estate
    1.25% Inflation-indexed bonds
    1.75% Cash

     

    Looks benchmark.
    17 Aug 2012, 06:55 PM Reply Like
  • borraa
    , contributor
    Comments (2) | Send Message
     
    Varan
    Do you use ETF Replay to do the analyses you do or something else?
    22 Apr 2013, 01:14 AM Reply Like
  • borraa
    , contributor
    Comments (2) | Send Message
     
    Varan
    In your comments made about 15 Feb 2012 (not sure what site I found this on),
    "For this basket, a simple relative strength strategy works very well. If you buy the top two (among SPY, GLD, TLT and SHY) on the basis of the returns of the previous three months, and hold for three months and repeat, you get the annualized (2005-2012) return of over 16% with no year of loss, maximum 5% drawdown, a Sharpe ratio of over 2, and the three factor alpha of 14%. Pretty good. Much better than buying and holding, and probably also superior to the 10 month SMA approach, as you have no more than eight round trip trades a year."

     

    What platform did you use in arriving at these back-test results ?
    ETF Replay or something else?
    22 Apr 2013, 01:27 AM Reply Like
  • varan
    , contributor
    Comments (4532) | Send Message
     
    Author’s reply » Thanks.

     

    I use a platform that I have developed.

     

    For the backtest you mentioned, here are the results:

     

    2005 16.80%
    2006 11.81%
    2007 17.24%
    2008 13.97%
    2009 17.78%
    2010 16.05%
    2011 25.46%

     

    CAGR 16.95%

     

    NET 2.99

     

    For 2012, the result was not that good. 2013 is still current and we shall see.

     

    2012 -2.82%
    2013 3.55%
    22 Apr 2013, 01:44 AM Reply Like
  • varan
    , contributor
    Comments (4532) | Send Message
     
    Author’s reply » Due to the lack of interest, haven't updated this in a while.

     

    The returns so far:

     

    Ivy

     

    2007 17.79%
    2008 14.61%
    2009 24.27%
    2010 7.70%
    2011 15.85%
    2012 18.84%
    2013 14.10%
    2014 19.12%
    2015 -3.79%

     

    CAGR 15.04%

     

    Swensen
    2005 12.26%
    2006 16.99%
    2007 2.42%
    2008 1.69%
    2009 30.29%
    2010 10.90%
    2011 25.49%
    2012 22.33%
    2013 22.37%
    2014 19.12%
    2015 -2.28%

     

    CAGR 15.08%

     

    Bernstein
    2003 24.20%
    2004 9.35%
    2005 13.20%
    2006 15.15%
    2007 7.66%
    2008 -1.17%
    2009 23.08%
    2010 3.15%
    2011 22.22%
    2012 19.95%
    2013 24.35%
    2014 17.56%
    2015 2.01%

     

    CAGR 14.25%
    31 May, 10:54 AM Reply Like
  • tmdoherty
    , contributor
    Comments (678) | Send Message
     
    Thanks for updating this varan. I hope you will continue to keep this up to date.

     

    Applying risk parity to these portfolios does indeed seem to improve them significantly. But it is hard to know how much the improvement is without side-by-side comparative statistics. Can you please update more comprehensive performance metrics in comparative fashion? For example:

     

    1. What were the Sharpe ratios and maximum drawdowns?

     

    2. What were the rolling 12-month returns, and especially the 95% confidence interval limits of those returns?

     

    3. Some think these portfolios have good defensive performance. So it would be particularly helpful to know how the risk-parity modified portfolios held up during periods of market stress, such as the 2008/2009 financial crisis, the 2011 credit downgrade, and various pullbacks/corrections.

     

    Thanks,

     

    TMD
    31 May, 04:20 PM Reply Like
  • tmdoherty
    , contributor
    Comments (678) | Send Message
     
    Hi Varan,

     

    I believe there is a new trade scheduled for tomorrow (June 1). If so, the selections I calculate for the Risk-Parity Modified Swensen Portfolio for the next 2 months are:

     

    VGTSX 56%
    VTSMX 44%

     

    Is this correct?

     

    Thanks,

     

    TMD
    1 Jun, 02:59 AM Reply Like
  • varan
    , contributor
    Comments (4532) | Send Message
     
    Author’s reply » It trades every two months. Its on Jan/Mar/May/Jul/Sep/Nov schedule.

     

    Thanks.
    1 Jun, 08:50 AM Reply Like
  • tmdoherty
    , contributor
    Comments (678) | Send Message
     
    Thanks much.

     

    What are the results like if you use the other cycle (Feb/Apr/Jun/Aug/Oct/Dec) ?

     

    According to my calculations, the strategy is up about 20% YTD using that cycle.

     

    TMD
    1 Jun, 02:07 PM Reply Like
  • varan
    , contributor
    Comments (4532) | Send Message
     
    Author’s reply » Wow! I will have to take look. Kind of a pain to modify the calcaluation but your result seems to justify the effort. Thanks

     

    After GMR ( with QQQ replacing ILF and with added EFA ) this is one of the best methods.

     

    Interesting that the annual top ten dividend growth stocks that I track elsewhere would also have done better had I bought them in Feb.
    1 Jun, 05:39 PM Reply Like
  • tmdoherty
    , contributor
    Comments (678) | Send Message
     
    RE dividend growth stocks...are you referring to your Instablog on "DG5-10"? That could use a performance update. Interesting that there have been no comments. I would have thought the rabid hounds of DG hell would have been all over that.

     

    RE the risk-parity modified Swensen strategy, I want to verify that I am calculating things the same way you do. I measure returns as the price at the close of the last trading day of this month vs. the price at the close of the last trading day of the month 3 months ago. So for the current trade for the "off-cycle" trades (I realize the current trade is actually next month), that would be the closing price on May 29 vs. the closing price on Feb. 27.

     

    I calculate returns as the entry price at the close of the first trading day of the month, vs. the closing price on the last trading day of the month two months later. So for the current trade, that would be the closing price on June 1 vs. the closing price on Aug 31.

     

    Risk parity weights are calculated from the SD of the daily returns over the previous 3 calendar months. So, for the current example, that would be Feb. 27 through May 29 inclusive.

     

    Please let me know if this is how you are calculating things.

     

    Thanks,

     

    TMD
    1 Jun, 06:20 PM Reply Like
  • varan
    , contributor
    Comments (4532) | Send Message
     
    Author’s reply » The entire purpose of the DG-10 is to just update it once a year as a counterweight to the verbal effluivia on those types of stocks so rampant. I made a performance update on 1/1/2015 and will update next on 1/1/2016. YTD the returns if bought on 2/1 are twice the returns if bought on 1/1.

     

    Re. Risk parity, everything seems to be the same except that I assume all transaction to occur at closing on the first trading day of the respective month. That is more convenient as Yahoo Finance gives the adjusted data only for the closing price. You can calculate the other prices but this way is simpler.

     

    Thanks.
    1 Jun, 07:32 PM Reply Like
  • tmdoherty
    , contributor
    Comments (678) | Send Message
     
    Thanks varan.

     

    I discovered an error in my calculations of the "off cycle" version of your strategy, and found that the YTD returns are about flat (in large part because of selection of VUSTX twice in a row, both times for losses). Then I examined performance over time (since Oct. 2000). I calculate a CAGR of 5.6% with a max DD of -33.6% and a Calmar ratio (actually a MAR ratio) of 0.17.

     

    Mean/median returns of all 3,523 rolling 12-month return periods was 8.0%/10.2% respectively, and the 95% CI limits of those returns ranged from -19.4% to +35.4%.

     

    Over the same period, the CAGR of the funds (VTSMX, VGSIX, VGTSX, VIPSX, and VUSTX) was 5.2% to 11.6%, but the max DDs were very high. The Calmar ratios of these funds ranged from 0.07 (VGTSX) to 0.41 (VUSTX).

     

    So the strategy did not perform as well as two of the component funds:

     

    CAGR/Max DD/Calmar:

     

    VIPSX: 5.3%, -15.1%, and 0.35

     

    VUSTX: 7.6%, -18.4%, and 0.41

     

    My conclusion is that this off-cycle strategy is just not worth it, and performs much worse than the performance statistics you cite above. Not sure why that should be.

     

    Would be great if you could do the same analysis to confirm the above.
    4 Jun, 05:36 PM Reply Like
  • varan
    , contributor
    Comments (4532) | Send Message
     
    Author’s reply » I tried the off month strategy for 2015 using the ETFs that I have listed and reached the same conclusion as yours. If you can list the funds that you are using I should be able to extend this back to 2000. Should post the results Monday/ Tuesday as some weekend diversions will occupy me I hope.
    4 Jun, 06:14 PM Reply Like
  • tmdoherty
    , contributor
    Comments (678) | Send Message
     
    Here are the funds I used:

     

    VTSMX
    VGSIX
    VGTSX
    VIPSX
    VUSTX
    4 Jun, 08:47 PM Reply Like
Full index of posts »
Latest Followers

StockTalks

More »

Latest Comments


Instablogs are Seeking Alpha's free blogging platform customized for finance, with instant set up and exposure to millions of readers interested in the financial markets. Publish your own instablog in minutes.