During springtime 2015 I read Antonaccis book "Dual momentum" and was very impressed by the results and easiness of the described Global Equites Momentum strategy, GEM. Each month, if stock market is strong, invest in the best 12-month performer of US and World ex-US, otherwise invest in aggregate bond. Particularly I liked the idea of diversification in the time dimension. Instead of the usual stock/bond mix, you hold bonds only when stock market is weak.
In August 2015 I dropped my 30% bond allocation and started a momentum portfolio, following GEM. Unfortunately a strong decline in stock market began almost the day after my investment, and according to the strategy I had to wait a month before I could rebalance into bonds. This was an awful experience. After some additional whip-saws during autumn/winter, I abandoned the strategy, and instead tried other more complex asset allocation strategies. But the easiness of the strategy made me return to GEM in October 2016.
GEM is technically easy to execute, but emotionally hard for many (most?) investors, including me, due to the whip-saw losses, slow response to sudden market changes and long periods of underperformance, despite the fact that the overall performance, both in- and out of sample in backtests, is excellent.
In August 2017 I read a very interesting blog post "Timing Luck and Portfolio Tranching", discussing the impact of trading on different days of the month, and showed the advantage of a combined approach. My immediate reaction was that this is a solution to the original slow response of GEM to sudden changes. Therefore I have since then tranched my portfolio in four slices, and checks/trades now weekly instead of monthly, enabling me to adjust to new market conditions on a weekly basis. At the same time and for the same reason, I replaced the original 12 month lookback period with a combined 3/6/12 month lookback. These changes have really helped me stick to the strategy.
In the beginning of 2018 American ETF's unfortunately became unavailable for EU-based investors, due to an EU-regulation. My initial approach was to emulate ex-US with Europe, Asia Pacific ex-Japan, Japan and Emerging Markets, but this turned out to mean a lot of transactions when swapping to and from ex-US stock, which for me led to some mistakes, wrong amounts of were sold and bought. It was also time consuming.
Instead I started to play with the idea to implement GEM with all of the regional assets and hold the 1-2 best performers, instead of either US or World ex-US. Inspired by various blogs and papers with backtest results I implemented this in May 2018. The detailed rules have been varied from time to time since then (gold was added in the beginning but later removed, the possibility to hold a single asset (read US) was added, etc.), but in essence the strategy has worked well for me and is easy to execute. And it is in my opinion still similar to the original GEM.
Recently (May 2019) I read a very interesting report about using different lookback periods. Instead of picking one or a few among 1-18 months, use all of them as a combined strategy. I like this idea of diversification of the parameters, since despite reading quite a lot about which lookback period that works best, I have never been convinced. During different time periods different lookbacks perform better. That's the main reason why I used 3/6/12 until now, together with the fact that it is more responsive to changes compared to 12m only.
The new idea (at least for me) is to use all lookback periods 1-18 month and run them as parallel portfolios. The result is a similar risk adjusted performance with lower draw downs, compared to original GEM, and you avoid the risk that a single or a few chosen lookbacks will be a bad choice the decades to come. To me this makes a lot of sense, and since a spreadsheet can be used to calculate the signals, I have decided to implement this approach.
Combining 18 lookback periods with 4 weekly tranches, gives 72 parallel portfolios. This can however in practice be run as a single portfolio where the combined signals are calculated weekly, see here. As expected the signals are more scattered for the shorter lookbacks and more stable for the longer. I have introduced a 6% threshold for new assets in order to avoid too frequent trading with small amounts, and will use the same filter for rebalancing between assets. Possibly the 6% threshold should be increased. Another possibility might be to skip the first months. Time will tell. (Edit 20/8/19: Threshold changed to 8%)
In August 2015 I dropped my 30% bond allocation and started a momentum portfolio, following GEM. Unfortunately a strong decline in stock market began almost the day after my investment, and according to the strategy I had to wait a month before I could rebalance into bonds. This was an awful experience. After some additional whip-saws during autumn/winter, I abandoned the strategy, and instead tried other more complex asset allocation strategies. But the easiness of the strategy made me return to GEM in October 2016.
GEM is technically easy to execute, but emotionally hard for many (most?) investors, including me, due to the whip-saw losses, slow response to sudden market changes and long periods of underperformance, despite the fact that the overall performance, both in- and out of sample in backtests, is excellent.
In August 2017 I read a very interesting blog post "Timing Luck and Portfolio Tranching", discussing the impact of trading on different days of the month, and showed the advantage of a combined approach. My immediate reaction was that this is a solution to the original slow response of GEM to sudden changes. Therefore I have since then tranched my portfolio in four slices, and checks/trades now weekly instead of monthly, enabling me to adjust to new market conditions on a weekly basis. At the same time and for the same reason, I replaced the original 12 month lookback period with a combined 3/6/12 month lookback. These changes have really helped me stick to the strategy.
In the beginning of 2018 American ETF's unfortunately became unavailable for EU-based investors, due to an EU-regulation. My initial approach was to emulate ex-US with Europe, Asia Pacific ex-Japan, Japan and Emerging Markets, but this turned out to mean a lot of transactions when swapping to and from ex-US stock, which for me led to some mistakes, wrong amounts of were sold and bought. It was also time consuming.
Instead I started to play with the idea to implement GEM with all of the regional assets and hold the 1-2 best performers, instead of either US or World ex-US. Inspired by various blogs and papers with backtest results I implemented this in May 2018. The detailed rules have been varied from time to time since then (gold was added in the beginning but later removed, the possibility to hold a single asset (read US) was added, etc.), but in essence the strategy has worked well for me and is easy to execute. And it is in my opinion still similar to the original GEM.
Recently (May 2019) I read a very interesting report about using different lookback periods. Instead of picking one or a few among 1-18 months, use all of them as a combined strategy. I like this idea of diversification of the parameters, since despite reading quite a lot about which lookback period that works best, I have never been convinced. During different time periods different lookbacks perform better. That's the main reason why I used 3/6/12 until now, together with the fact that it is more responsive to changes compared to 12m only.
The new idea (at least for me) is to use all lookback periods 1-18 month and run them as parallel portfolios. The result is a similar risk adjusted performance with lower draw downs, compared to original GEM, and you avoid the risk that a single or a few chosen lookbacks will be a bad choice the decades to come. To me this makes a lot of sense, and since a spreadsheet can be used to calculate the signals, I have decided to implement this approach.
Combining 18 lookback periods with 4 weekly tranches, gives 72 parallel portfolios. This can however in practice be run as a single portfolio where the combined signals are calculated weekly, see here. As expected the signals are more scattered for the shorter lookbacks and more stable for the longer. I have introduced a 6% threshold for new assets in order to avoid too frequent trading with small amounts, and will use the same filter for rebalancing between assets. Possibly the 6% threshold should be increased. Another possibility might be to skip the first months. Time will tell. (Edit 20/8/19: Threshold changed to 8%)
In essence the result is, in my opinion, a GEM-like portfolio with increased diversification in number of assets, in the time dimension, as well as in the lookback period.
In practice I run my momentum portfolio as three sub-portfolios. The first one, with regional stock assets and global/emerging markets bond market, is like original GEM USD based. The second one is very similar to the the first but SEK based, i.e. comparisons are based on SEK-calculated performance. And the third one, also SEK based, switches between Swedish and Global stock. Since total Swedish stock market cap is only a few per cent of the the global this is, at least in spirit, similar to GEM, but with US stock replaced by SE stock. This provides even further diversification in the process/strategy dimension, without, I hope and expect, loss in risk adjusted performance. Again, time will tell.
My momentum strategy is described here and the corresponding weekly investment signals herehere.
References
GEM/Global Regional Assets:
Portfolio tranching:
Multiple lookback periods:
Inga kommentarer:
Skicka en kommentar