In today’s post I want to take a quick look at the role of timing in the GTAA AGG portfolios. What impact does timing have on the performance statistics of the AGG3 and AGG6 portfolios versus not using timing at all in the portfolios. The results will surely surprise some.

The dominant impact to the performance of the AGG portfolios is relative strength, aka relative momentum. This should not come as a surprise as momentum is the most powerful and persistent market anomaly by far. The addition of timing using the 10 mo SMA signal in the AGG portfolios is meant to protect against crashes and improve risk adjusted returns and drawdowns. Well, does it? Let’s find out.

For this analysis I ran the AGG3 and AGG6 portfolios with and without the 10 mo SMA timing filter. I compare the compounded annual returns, risk, risk adjusted returns, and monthly drawdowns. I also added columns for the percentage of time the portfolio is long and the number of round trip trades per year. Note that for this analysis I used a slightly different database than I do for my historical portfolio stats. I did this to get the additional data on percentage of time long and round trip trades per year. That means you can’t compare these portfolio stats with others I’ve published. But the conclusions are the same and I think the additional data adds useful information. Now for the data. The table shows the AGG3 and AGG6 portfolios from 1973 to 2012 with and without the use of the 10mo SMA timing signal.

AGG portfolios 1973 to 2012 SMA impact

Are the results what you expected? Removing the SMA timing filter improves compounded returns and basically keeps risk adjusted performance (sharpe, sortino) the same. Drawdowns increase slightly with no SMA filter. At worst I would say the results are statistically the same. Timing does nothing over and above relative momentum in the portfolio! Stated more subtly, during the historical lookback period (1973 through 2012), there has been enough negative correlation among the 13 asset classes that there is almost always 3 to 6 on buy signal so the use of the SMA timing filter has little effect on portfolio performance. That’s evident from the percentage of time the portfolios are long. Makes intuitive sense when you think about it. This is even more clear when you narrow the look back period to just before the financial crisis. See data below for the period from 2007 to 2012. Even during this worst of periods, when everyone was complaining about ‘all correlations going to one’, the facts were otherwise.

AGG portfolios 2007 to 2012 SMA impact

A very important item to note is that these results would only apply to TAA portfolios with sufficient asset class diversification. For example, these results would not apply to the Antonacci GEM portfolio that switches between US and International stocks. In that portfolio, the crash filter is based on 12 month absolute returns, and is absolutely critical to its performance.

What about going forward? Is it worth continuing to use the SMA timing filters in the AGG portfolios? Well, it doesn’t hurt much as the data shows and in fact it does manage to reduce portfolio turnover slightly which will improve slippage in the real world. And if the future is different than the past and there is a crisis where indeed all asset correlations go to one, then the SMA timing filters will work even better than the past. So, I would say it’s not a bad insurance policy but you should expect slightly lower returns from using the SMA timing filters in the AGG portfolios.


34 Comments

Damian · June 9, 2015 at 12:11 am

Thanks for sharing this, Paul. Very interesting and, frankly, unexpected. What backtesting software and data set(s) did you use for this run?

    paul.novell@gmail.com · June 9, 2015 at 10:04 am

    Excel. Basically, an older version of Faber’s Backtested spreadsheet.

    Paul

Steve · June 9, 2015 at 6:14 am

Thanks for this analysis Paul! Very interesting. I do notice a significant reduction in DD in they 2007-2012 years for AGG6, which seems that timing did very well during the financial crisis of 2008. In your look -back to 1973, was the high DD for timing portfolios, attributed perhaps to the 87 crash?

    paul.novell@gmail.com · June 9, 2015 at 10:22 am

    In terms of drawdowns, yes, but returns will still lower.

    Yes, Oct, 1987, -20% DD. For the year though portfolio was up 7%. Without timing DD was the same. Nothing helped in 87, it was too fast.

    Paul

      Mark · June 9, 2015 at 2:05 pm

      Things moving too fast are where Stop Losses would have helped IMHO. Paul I know you are not a fan of stop losses based on comments you made when I suggested using them in another blog post. You said something to the effect of the average investor likely not wanting or knowing when to get back in if used. I respectfully disagree. I think stop losses keep the average investor from getting hit hard and not wanting to play again. If a person only has a tolerance for a 5% DD then set it at 5% vs. the month end readjustment and don’t worry about what the market does. If at the end of the month the AGG portfolio shows the ETF you sold still as a buy signal with momentum that you misread you can always get back in the game. Big DD’s are why many investors head for .01% bank interest or worse yet, their matresses. I have been doing this for awhile now and I can tell you stop losses have saved me quite a bit vs. waiting to the end of the month when I would still sell, but at a bigger loss. If big market corrections are coming (and who really knows when or where, see 1987) I prefer to be safe than sorry. I will forfeit some potential upside to avoid the landslide.

      Just my 2 cents.

      Mark

        paul.novell@gmail.com · June 9, 2015 at 7:06 pm

        Mark,

        Appreciate the comment. And I love when reasonable, smart people disagree with me. So, thanks for that.

        When I said that stop losses don’t work, what I meant is that they don’t improve compound returns, sharpe, and sortino ratios over and above the standard rules of the AGG system. That’s what the data shows any way you run the numbers. My comments were specific to the AGG systems. If you can show me that stop losses improve annual returns, sharpe, and sortino during the historical period over and above AGG3/6 rules then I’m all ears. Now, the use of stop losses may improve results over an investor’s actual implementation and reaction to the volatility of the AGG3/6 system. I buy that – depends on the investor. Every investor should find a system that works for them.

        Also, in some systems, like the Quant Trending Value system I post about, using trailing stop losses and more frequent rebalancing periods does boost returns, sharpe, and sortino so I use them there.

        Paul

          Mark · June 10, 2015 at 10:08 am

          Glad to hear it. Paul, I fully respect and appreciate the work you do and have been following you for over a year now. I did not want to come off as being “that guy” so glad I hopefully didn’t. At the end of the day, when you are managing your own money, you have to have a strategy and system that lets you sleep at night (and your wife too!) For me, I have seen too many colleagues subscribe to the buy and hold and end up holding through the big losses and then finally getting out at the bottom and never getting back in. I know GTAA addresses this through month end re-balances, I just like the safety net of adding stop losses in the event the market tanks during the month. My stop losses have only triggered once so far, so we are more aligned on this than it may appear. However if the market drops 20% mid-month I am out way ahead of that. I then wait for the buy signal and get back in or stay out depending on what happens. While the historical look back data may not support my approach, I can tell you the comfort of knowing I am not going to get pummeled by a 1987, 2000, or 2008 is comforting. When you have sizable investments in different strategies, stomaching big losses is not for the faint of heart. Recovery from a 20% dip requires a lot more than a 20% gain. Dated article, but still very true and timely. http://observationsandnotes.blogspot.com/2011/02/importance-of-avoiding-large-losses.html

          On a separate note for those of us who have money in 401Ks that we manage separate from personal accounts, have you looked at a “quant strategy” using ETFs since we can’t buy individual stocks inside 401K. Was thinking of developing and applying GTAA logic to XLY,XLK,XLI,XLB,XLE,XLP,XLV,XLU,XLF to have a “quant like” portfolio. For those that don’t know these ETFs they represent the S&P Sectors. It is interesting to watch the dynamics of the sectors inside the S&P vs. just the closing bell number. There is lots of movement inside the S&P that is not always correlated. Interested in your thoughts.

          As always, very much appreciate the work you do and the opportunity to exchange thoughts in this forum.

          Thanks!
          Mark

          paul.novell@gmail.com · June 10, 2015 at 11:14 am

          Totally agree with what you’ve seen with the buy and hold types. I’ve seen it way too much as well.

          You make a good point about 401Ks and their limitations. A sector rotation model like you describe is easily implementable. In fact, Antonacci has one in his book “Dual Momentum”. The performance numbers are pretty good. I can throw something together in P123 and see how it does then maybe do a post on it.

          Paul

Abraham · June 9, 2015 at 1:57 pm

Hi Paul,
looking at your spreadsheet, I notice you compute momentum as the average of 1,3,6, and 12 month performance. One month performance is generally considered mean-reverting so I’d expect its contribution to future momentum estimation should, if anything, be inverted.
What’s your opinion on this?
Thanks.

    paul.novell@gmail.com · June 9, 2015 at 7:17 pm

    Abraham,

    This is an area I’m doing some research into but I can say a few things about it now. If you run the numbers for the AGG3 and AGG6 portfolio using index data from 1973 to 2014 with and without the 1 month performance, the numbers are basically the same. Without the 1 month performance the numbers are a bit worse but statistically the same. But if I run the numbers taking into account actual implementation with ETFs and fees/slippage I do get better performance by dropping 1 month performance. Problem is you can’t test real word performance with ETFs going back earlier than 2004 or so. So, basically I can’t say anything conclusive on it.

    Paul

Bryan · June 9, 2015 at 6:34 pm

Paul,

Great post, but I am still a little confused on how to implement. If you are removing the 10 month SMA, what metric are you using to determine which efts are invested in the AGG3 or AGG6 portfolio?

Thanks,
Bryan

    paul.novell@gmail.com · June 9, 2015 at 7:32 pm

    Their performance ranking (average of 1,3,6,12 month total return).

    Paul

      david hilliard · June 10, 2015 at 5:04 am

      I am also not up to speed on the implementation of the buy signals. Do you happen to have a link or post date for a explanation of the buy signal. Thanks for the helpful blog. David

        paul.novell@gmail.com · June 10, 2015 at 10:54 am

        I covered the rules in this old post. Also, the original paper can be found on Faber’s website.

        The rules are pretty simple.
        1. Rank the ETFs by the average total return over the last 1,3,6, and 12 months.
        2. For AGG3(6), buy the top 3(6) equal weight if and only if they are above the 10 mo SMA
        3. If any of the top ranked ETFs are below the 10mo SMA then that part of the portfolio stays in cash.

        Paul

          david hilliard · June 10, 2015 at 10:57 am

          thanks very much David

          Paul Duke · June 14, 2015 at 12:51 pm

          Paul,
          First of all – thanks for the great site and valuable information! Having just gotten up to speed by reading Faber’s papers and this site (great stuff). I still have a question regarding implementation of AGG3(6). I see your reference to “Equal Weight” for AGG3(6)….as in, 33% of one’s portfolio in each of the top 3 selected assets (as referenced in above post). Can you point me to the reference to “Equal Weighting?
          Thanks so much for your site, time and energy.

          A fellow “Paul”

          paul.novell@gmail.com · June 16, 2015 at 10:26 am

          Hi Paul, in Faber’s paper on the topic he references his relative strength study when discussion the GTAA aggressive portfolios. If you go to his relative strength study you will find the equal weighting in the ‘buy rule’ section.

          Paul

David · June 10, 2015 at 2:17 am

Hi Paul,

Interesting post, as always !

Thanks,

David

Mark · June 10, 2015 at 11:29 am

Paul —> You make a good point about 401Ks and their limitations. A sector rotation model like you describe is easily implementable. In fact, Antonacci has one in his book “Dual Momentum”. The performance numbers are pretty good. I can throw something together in P123 and see how it does then maybe do a post on it.

Mark —> That would be awesome. Looking forward to it!

    Mark · June 10, 2015 at 11:46 am

    Found this article on StockCharts.com referencing two of our favorite guys — Faber and O’Shaunessey. If you haven’t read it, might summarize/compliment the work by Antonacci in Dual Momentum.

    http://stockcharts.com/school/doku.php?id=chart_school:trading_strategies:sector_rotation_roc

      paul.novell@gmail.com · June 10, 2015 at 2:06 pm

      Thanks.

    Mark · August 10, 2015 at 8:33 am

    Hey Paul….Trust you are doing well and having a great summer! Did you ever have a chance to run a model around the S&P sectors?

    Thanks
    Mark

      paul.novell@gmail.com · August 21, 2015 at 10:01 am

      Hey Mark, not anything above what Faber’s sector model shows. IMO, to do better then that model you need to add sectors beyond the traditional SP sectors. Like biotech for example.

      Paul

John · June 10, 2015 at 1:07 pm

In your post of April 2, 2015 your table shows a sortino ratio of 5.28 for AGG3. Why the big difference?

    paul.novell@gmail.com · June 10, 2015 at 2:06 pm

    John, as I said in the post, you can’t compare the results in this post with what I’ve posted in the past. But since you ask, the biggest difference is that in most of my historical posts I use annual data. The database used for this post uses monthly data.

    Paul

Steve · June 11, 2015 at 5:29 pm

My 401k provider discourages movement in and out. They assess a 2% withdrawal penalty for most funds if not held for 60 days. Other funds need to be held for 30 days. This makes it impossible to follow GEM unless I modify the rules.

    paul.novell@gmail.com · June 16, 2015 at 10:18 am

    That is not too uncommon.

    Paul

Angela Smith · June 17, 2015 at 7:34 am

This website is excellent for both my husband and I.

    paul.novell@gmail.com · June 17, 2015 at 12:27 pm

    Thanks Angela. Glad its helpful.

    Paul

Gary · June 19, 2015 at 6:26 am

Paul, I’ve been following your blog (and your wife’s) for about a year now. I know you have commented in the past on firms offering low cost automated investing such as Betterment and Wealthfront. Do any of these firms employ asset allocation strategies that you are discussing here?

    paul.novell@gmail.com · June 21, 2015 at 11:56 am

    Hey Gary, no they don’t. They basically offer versions of buy and hold portfolios.

    Paul

Will · June 28, 2015 at 5:50 am

Hi Paul,

Thanks for another great post!

Related to timing, how about the situation where you have a pile of money coming to you, e.g., from a mandated 401K distribution, or inheritance, or bonus, etc. The question is WHEN to invest it? Clearly if you happen to get it at a market high and invest it then, you will not recover for years. On the other hand, if you are lucky and get it at market bottoms, you win big time. But, in general, is there an approach to maximize one’s probability of succeeding? I am wondering about legging in over time for example? Or, is it better to go all in right away? Or, is there one portfolio that will be better in this situation than another? What is your thinking in this situation?

Will

    paul.novell@gmail.com · June 30, 2015 at 7:02 pm

    Hi Will,

    The data says to invest right away. Waiting usually, but not always, lowers performance. The timing filters are there to protect against big losses. But in my experience, it’s more about your emotions as an investor than what the data says is the right thing to do. I usually find that people are more comfortable legging into a portfolio over a series of months, say in thirds or quarters.
    There is no right answer – it depends what you feel comfortable with.

    Paul

Mark · July 11, 2015 at 3:54 pm

While I am still an advocate of not being exposed in the market, examples like this show that timing is way less important than consistency and staying in the game. Good read.

http://awealthofcommonsense.com/worlds-worst-market-timer/

Comments are closed.