StockViz

Our GEM backtest in Part III used a 12-month formation period to measure momentum. Here, we look at alternative formation periods with an eye on drawdowns.

6- through 12-month formation periods

Even though the 10-month version has higher returns, the 6-month one has lower peak drawdowns.

The average of all

The problem with picking one formation period out of 6 is that it smells of data-mining. What happens if you average them all out?

The average works in reducing drawdowns compared to the traditional 12-month version.

We will setup a virtual portfolio for this “averaging” strategy and post the link here when it is up and running.

Code and more charts on github.

We saw in our earlier posts on Global Equities Momentum (Part I, Part II) that by swapping the momentum equivalent of the equity indices in the GEM decision tree, one could significantly boost returns. Also, momentum trumped value.

Correlation between momentum and base indices

In the original GEM dual momentum model, the S&P 500 index was used to decide and to trade. What we claim here is that we can continue to use the S&P 500 index to decide, but we will use the momentum equivalents to trade. To back our claim, we present the correlation in the monthly returns of the base/momentum index pairs:

The indices move pretty much in tandem.

Robustness

If dual momentum is robust, then our strategy piggybacks on its robustness through the decision tree. Where we differ is in the way we express the trade. And our backtest shows that GEM is superior to buying and holding the underlying indices themselves both in terms of returns and drawdowns:

Annual returns:

Instruments

Implementing this strategy is fairly straightforward. You need to track the following ETFs:

• SPY: for S&P 500
• BIL: for US T-bills
• IDEV: World ex-US
• MTUM: US Momentum
• IMTM: World ex-US Momentum
• AGG: Aggregated bond

You will be long one of the last three ETFs above at any given point in time:

We will setup a virtual portfolio for this strategy and post the link here when it is up and running.

Code and more charts on github.

In our previous post on Global Equities Momentum, we explored how we could potentially replace the indices used in the GEM decision tree with their momentum counterparts to boost returns. Corey Hoffstein (@choffstein) pointed out that given the excess turnover of momentum strategies, measuring their trend maybe adding too much noise. Also, could using value indices, given their lower turnover, make more sense?

We setup the following backtest to fix the first problem and explore the second.

1. We will use the S&P 500 index to make the first decision of the GEM model: Should we invest in equities or bonds?
2. Once we get past #1, we will use different sets of indices to make the next one: USA or ex-USA? And trade the same.
3. We will use the MSCI USA PRIME Value index to represent US Value and MSCI ACWI ex USA PRIME Value index and MSCI WORLD ex USA PRIME Value index, in turn, to represent international value.

USA/All World ex-USA Value GEM

USA/Developed World ex-USA Value GEM

1. The GEM models both show vastly better returns and shallower drawdowns compared to buying and holding the underlying indices alone.
2. There is not a lot of difference between the two GEM models.
3. However, there are no equivalent ETFs for investors interested in implementing either of these GEM models.

In contrast, USA Momentum/MSCI World ex-USA Momentum:

Not only does the momentum GEM vastly outperform the value GEMs, it can be easily implemented with the MTUM and IMTM etfs.

Before jumping into any of these strategies, it is worth asking: Is this just data mining? How can we be sure that these backtests are statistically valid? By the same token, how can we be sure that even the dual-momentum model is robust? Gary Antonacci’s GEM backtest goes as far back as 1971 but we only have MSCI index data starting from 1995 or later. Besides, we are not sure if it is even possible to construct a reasonable momentum index going that far back. So, caveat emptor!

Code and more charts on github.