I should start this Post by saying that I have been a follower of Gary Antonacci, his papers and blog (http://www.dualmomentum.net/), for a long time. I can honestly say that Gary Antonacci has been one of the people responsible for me accepting momentum as a valid strategy for portfolio construction – despite my initial skepticism and reluctance to do so (to put it mildly). Many of his concepts/ideas are built into the Ranking spreadsheet that I make available through this site.
Although I was familiar with Antonacci’s work I decided to buy his book, “Dual Momentum”, in the hope that I might find at least one new idea that I might be able to use to improve the current spreadsheet. I will therefore start with my review/reaction to the book.
For anyone not familiar with momentum investing the book provides an excellent history of the academic work done in this area and a convincing argument that it can be used as a valid methodology for portfolio construction and investment. However, I don’t feel that the background and build up is especially well supported by the “practical” Global Equity Momentum (GEM) model presented as an example of his “Dual Momentum” strategy. I will elaborate on this statement in the material that follows but, before this, I would also comment that I found some of his statements, on issues such as the requirement for diversification, to be unclear and a little confusing. If I were a cynic I might suggest we should expect a follow-up – Dual Momentum-2 – since he has not covered all the work described in his already published papers. So, although I was not expecting too much, I did find the book disappointing.
Before presenting any back-test results I would like to make some important observations/comments:
- Antonacci’s back-tests cover a period of 40 years (1974 – 2013) – obviously a highly desirable time frame covering a range of market conditions;
- For any “system” to be considered “robust” I require it to be effective over a wide range of market conditions (time “windows” within the larger “picture”) and applicable to selections of “reasonable” assets in different (global) markets;
- There should be suitable “tradable” assets available, with good liquidity, to enable practical application of the system.
In the following back-tests I take a look at the performance of a 3 asset portfolio following the Global Equities Momentum (GEM) Dual Momentum strategy described in the book. In this example I have used ETFs (VTI, VEA and TLT) to replace S&P 500, ACWI ex-US and Aggregate Bonds as used by Antonacci in his book. I could have used SPY rather than the broader VTI, VEU rather than VEA to include emerging market equities, AGG rather than TLT to represent bonds … or numerous other options – but the differences are not likely to be all that significant. The first point to make here is that using ETFs confines me to an ~8 year test period rather than the longer (more desirable?) 40 year period covered by Antonacci in his book.
The first back-test employs Antonacci’s GEM system, but applied to the ETFs noted above. This test adopts the strict rules of Antonacci’s system in that 100% of funds are allocated to one of the 3 ETFs (VTI, VEA or TLT), based on prior 12 month Relative Momentum.
The performance of the Antonacci “Dual Momentum” system is shown by the light blue line in the above figure and is compared with the performance of the Vanguard 70/30 Equity/Bond Retirement Fund (VTTVX) over the same ~8 year period. A small change was made to the Ranking SS used on this site to ensure that TLT was used as the “go to” asset if equities (VTI or VEA) were not performing well (i.e. had negative Absolute Momentum). For comparison, I have shown the results (red dashed line) generated using the “standard” SS settings that allocate funds to SHY if no assets are ranked higher than SHY. There is only 1 month in which allocations were different (SHY was chosen over TLT since 12 month SHY momentum was higher than TLT – i.e. TLT had negative Absolute Momentum) therefore differences are not significant (it could just as easily have gone the other way).
The good news is that the turnover for this system is very low (very few switches between ETFs and low transaction costs). Also, performance during the 2008 downturn in equity markets, that resulted in large drawdowns (~41%) in VTTVX, was good. However, the system was slow to detect/respond to the upturn in equity markets that resulted in system drawdown of ~29% over a ~3 year period between 2009 and 2011. Psychologically, this would probably have made it difficult to stick with the system over this period. System volatility was a reasonable 14% although annual returns (CAGR) were only 5.34% (resulting in a 0.38 Sharpe ratio) compared with a 6.12% (CAGR) return for VTTVX i.e. the system did not beat the performance of VTTVX over this 8 year period.
This is where it gets difficult to assess the value of back-tests – if we look at the 40 year record, as presented by Antonacci (and many other researchers who analyze data over long time periods – a highly desirable goal) then results look good. However, unfortunately, most of us are getting to that age where a 40 year time frame isn’t exactly on our horizon – and an 8 year time frame is far more meaningful. Ideally we want a system that looks good over a 40 year horizon, but that also looks good in any 8 year “window” within that period.
Antonacci mentions alternate look-back periods briefly, but does not show any performance charts, only a table showing 40 year annual returns for 3, 6, 9 and 12 month periods – this shows the 12 month look-back showing the highest annualized returns.
The following figure shows the performance of the same system, over the same time period, using a 6-month look-back period (light blue dashed line) and using the “standard” 3 month/6 month weighted periods used in most of our analyses on this site (red line):
Certainly, the performance of the latter 2 options appears to be superior to the 12 month look-back option – but is this sustainable? We need to be careful in our judgment/assessment and not be too critical of Antonacci’s choice of the 12 month look-back period. Since market behavior/conditions change with time I believe we really need an “adaptive” capability – but how to achieve this is not easy to define.
The 6 month (100% weighted) and 3/6 month (50%/30% weighted – 20% Volatility) options obviously involve more transactions, and associated costs, but this would certainly appear to be justified based on the above performance.
12 month (100%): CAGR = 5.34%; Volatility = 14.03%; Return/Risk (Sharpe) = 0.38; MDD = 29.16%
6 month (100%): CAGR = 13.4%; Volatility = 14.11%; Return/Risk (Sharpe) = 0.95; MDD = 15.14%
3/6 month (50/30/20): CAGR = 16.03%; Volatility = 13.14%; Return/Risk (Sharpe) = 1.22; MDD = 11.57%
In my next post on the subject of Dual Momentum I will address the issue of diversity that Antonacci seems to avoid in his book – but has addressed in his papers.
November 27 Addendum: Following comments from Hartford the following figure shows the performance of the GEM Dual Momentum model using AGG (Aggregate Bond) rather than TLT as the Bond ETF:
In the end, there is little difference in total return over the ~8 year period, however, volatility, draw-down and Return-Risk characteristics are more favorable/attractive:
12 month (100%) using AGG: CAGR = 5.97%; Volatility = 10.98%; Return/Risk (Sharpe) = 0.54; MDD = 17.07%