Research has shown that stock prices are reactive rather than trending over short periods of time. Stocks are more likely to rebound from short-term fluctuations than to continue in the same direction. You will get whipsawed if you check for signals too frequently. Our proprietary models that use additional lookback periods can adjust more frequently.
It takes a while for research to work its way into the investment marketplace. We saw that with indexing. Public awareness should grow as information about momentum becomes more widely accepted.
There are other reasons momentum investing has not yet caught on. First, it is not as psychologically appealing to buy what has gone up in price as it is to buy things that appear to be cheap. Momentum may also seem too simple for those who mistakenly think more complex approaches have more merit.
Some investors have prejudices against other styles of investing or investing in indices rather than in stocks. Interestingly, some of the same behavioral factors that explain why momentum works also explain why momentum has not caught on yet: anchoring, herding, conservatism, and the slow diffusion of information. There is also a strong and universal home country bias that keeps many from investing much outside of their home country using dual momentum.
Finally, limits to arbitrage keep institutional investors from investing in new ideas because of career risk. Dual momentum, like all strategies, will have periods of underperformance. If you are wrong but doing what everyone else is doing, you have a better chance of keeping your job and your clients. But you are likely to lose clients or your job if you lose money at times when others do not.
My blog posts are a good place to start. They were written after my book was published. In many of the posts, there are links to other sources of information.
Here are two of these references that are especially relevant. They contain long-term validation of both relative and absolute momentum.
Geckzy & Samonov (2017),
Greyserman & Kaminsky (2014).
You can click on “Look Inside” to read the first chapter of this book.
You may also want to view some of my podcasts.
One risk is that history is unreliable. The future may not be as good as the past, even though there are several hundred years of out-of-sample performance supporting both relative and absolute momentum. There are behavioral reasons momentum has a good chance to continue working, since behavioral biases are hard to change.
There is some risk that momentum will attract substantial capital and not work as well in the future because of too many participants. But the index ETFs we usually use are more scalable than individual stocks. Behavioral biases can keep investors away from dual momentum.
Much of the short-term volatility of the stock market still exists with dual momentum. There are re-entry lags when a new bull market begins after a bear market. You should expect this especially from our public dual momentum models. We designed our proprietary models to reduce this transitional lag.
If you invest with dual momentum after the stock market has gone up quickly, your risk is higher for a market pullback.
There is also the risk of tracking error. No strategy outperforms its benchmarks all the time. The biggest risk factor may be that one becomes inpatient or overly influenced by short-term performance thereby abandoning dual momentum at an inoppotune time.
Future drawdowns may also be larger than those in the past.
Please see our Disclaimer page for additional risk factors.
The number of sectors depends on one’s attitude toward reward and risk. Fewer sectors mean higher expected returns, but also higher volatility. You can find parameters to create a sector rotation model suited to your individual risk preferences. Be aware, though, that data-mined parameters may not hold up that well out-of-sample.
The performance of DMSR using additional data since I published my book no longer looks as attractive as broader-based dual momentum models like GEM or our proprietary models. A recent study also shows that only 26% of sectors over their lifetime have outperformed an S&P 500 ETF.
Using data back to 1801, Geczy and Samonov (2015) show that momentum with geographically diversified stock indices outperforms momentum with sectors, individual stocks, currencies, bonds, and commodities. This may be because home country bias keeps investors from investing a lot in more attractive geographic areas, thus creating abnormal profits for those who can do so.
EMs are already included in the MSCI ACWI ex-U.S. index that GEM uses. They make up 17% of that index based on market capitalization. Investors are more prone to overreact and sell more volatile assets, like EMs, during times of crisis. Those who want to add an extra allocation to EMs should proceed cautiously. We use EMs as a separate asset in our proprietary models, since we have good risk controls and more diversification there.
Our Dual Momentum Fixed Income (DMFI) model has the best reward-to-risk characteristics. Since 1970, it has earned a higher return after expected fees and trading costs than the S&P 500. Because of its very low correlation to our other models, DMFI is useful as a portfolio diversifier as well as a stand-alone investment. See our Proprietary Models performance page or contact us for more information.
There may be some advantages in using multiple look-back periods if selected intelligently and not indiscriminately. There are tradeoffs that need to be considered when you add complexity to a model. We find it better to add other assets and more trend determinants rather than having too many look back periods.
The negative correlation you have seen between stocks and bonds has not always existed. There have been long periods of positive correlation between stocks and bonds.
Source: Graham Capital Management Research Note, September 2017
Long-term bonds also have considerable duration risk. They are best suited for those who can hold them to maturity. During times of crisis, investors usually prefer safe harbor assets with little risk of any kind. See here for more on why long-term bonds may not be the best safe harbor asset. Our Dual Momentum Fixed Income (DMFI) model has done exceptionally well without incurring the risk of long duration bonds.
You can apply dual momentum to the bond market using the simple logic in my book.
The model in my absolute momentum paper (included as an appendix in my book), moves to U.S. Treasury bills when bonds are weak. We do something similar in our proprietary models.
There are 16 major leading economic indicators. With so many to choose from, it is easy to put together something that looks good in hindsight. The unemployment rate has been a useful stock market timing indicator back to1970. But it gives worse results than the trend alone before then. No one knows what the future will be like.
The stock market itself is a leading economic indicator. There can be potential data snooping and selection bias issues when you add additional indicators or parameters.
Our public models like GEM were designed to protect smaller investors from extreme drawdowns while giving them a chance to earn better than buy-and-hold eturns over the long run. They were not desinged ot be optimal models.
The proprietary models we license to advisors and family offices use more assets than our public models. They also use multiple look-back periods and variations on trend-following. Besides an advanced version of GEM (A-GEM), we fixed income and global balanced proprietary dual momentum models (E-GBM and DMFI) to accommodate different risk preferences. Our newest proprietary model, NASDAQ Breadth & Trend (QBAT), uses daily rather than monthly data. Besides dual momentum, it uses breadth, sentiment, intermarket validation, and mean reversion. It is our most advanced and most adaptable model. See the Proprietary Models page or contact us for more information on our proprietary models.
That is a question I think all model developers and investment professionals should answer. I invest over 97% of my funds in a mix of my own proprietary models. They provide good returns and all the diversification I need.
Most 401K plans include a fund for non-U.S. stocks. You could use that. If your 401k has an emerging market fund, you could create a synthetic ex-U.S. all country world fund by allocating 15-20% to the emerging market fund and 80-85% to a non-U.S. developed markets fund. You can still derive signals using an ex-U.S. all-country world stock ETF as shown in my book.
Our research shows that these do not improve GEM’s results. Factor-based funds are often subject to data mining and selection biases. Real-time results from factor-based funds have not been encouraging. See our blog post here.
There are several of them mentioned in the reviews of my book on Amazon and elsewhere. Two of them are here and here. If you are still skeptical, you can do your own backtesting with ETFs or mutual funds using Portfolio Visualizer. You should use an S&P 500 fund to determine “single absolute momentum”. Mutual funds have a longer history than ETFs. You could use VFINX, VWIGX, and VBMFX to get results back to 1987. These results are somewhat different from what you would get using ETFs or indices.
I use the All Country World ex-US Index along with the S&P 500 in order to be all-inclusive and to avoid selection bias. Selection bias is where you select a subset of the data in order to make your results look better or, in this case, worse. EAFE omits Canada and emerging markets, which represent a meaningful 24% of the world’s equity market.
ACWI ex-US and EAFE performances may appear similar. But ACWI ex-US gains more in up markets and loses more in down markets than EAFE. We designed dual momentum to capture the upside while avoiding much of the downside. So there is a big difference in performance between the two indices when using dual momentum that is not apparent from looking at their non-filtered performance.
EAFE is also slower than ACWI ex-U.S. in switching to non-U.S. stocks when they become strong versus U.S. stocks. This means EAFE can give different switching times.
It is also misleading to say there is little difference in the performance of EAFE and ACWI ex-U.S. by looking at all the data since 1970 or earlier, since MSCI didn’t track emerging markets before 1988. There was no difference in thier performance before 1988 since ACWI ex-U.S. did not exist before that. But since then, dual momentum captured the out performance of ACWI ex-U.S. while reducing downside exposure.
Holding all else constant, GEM earned 130 basis points more in annual return using MSCI ACWI ex-U.S. rather than MSCI EAFE from when the ACWI ex-U.S. index was created by MSCI in 1988. There is no reason to invest with MSCI EAFE which produces considerably lower returns than MSCI ACWI ex-U.S. in GEM once you apply dual momentum.
One reason is that there is a lot more index data history. Another is that non-U.S. stock ETFs and mutual funds (as well as index data using net instead of gross returns) withhold 30% of their dividend income for foreign taxes. This is recoverable by some investors as a tax credit. To get accurate relative strength signals, you would need to deduct 30% of the dividends from U.S. ETFs.
Expense ratios can also distort the relative performance between assets. Momentum decisions should be based on asset performance. The markets do not care about your taxes or other expenses.
The ones I have seen impose their own ideas without having good reasons for their changes. You can easily determine the signals yourself with price charts using the information in Chapter 8 of my book. Alternatively, you can use a free Morningstar, Portfolio Visualizer, or ETF Screen portfolio to determine the GEM signals.
I show dual momentum both ways in my book. It may be easier to apply relative momentum first if you are using a variety of assets. If you are only using U.S. and non-U.S. stocks, then it is usually better to apply absolute momentum first. On page 98 of my book, I determine absolute momentum first using the S&P 500 index, since the often U.S. leads world equity markets.U.S.
Studies here, here, and here show that stocks perform best early in the month. This is when institutional investors make changes to their portfolios. Prices then are most representative of their true value. Rebalancing during that time frame has shown the best results with our models. Here are the Sharpe and Sortino ratios for GEM based on the trading day of the month.
Our published results are based on month-end closing prices. I have tested using an equal allocation to stocks and bonds when the switching signal is close. Results are about the same as following the signal precisely. To reduce feelings of regret no matter what happens, you might use this approach. Another possibility is to wait a day or two and see which way the signal goes.
The idea behind absolute momentum is to look at the return of the S&P 500 above the risk-free return. There is no reason to bear the risk of stocks when you can earn a better return without so much downside risk.
Accelerating Dual Momentum is highly data mined using a limited amount of data. You can read more about this in my blog post, “Perils of Data Mining.”
Dual momentum has a trend-following component that lags when a new bull market begins. That is the cost for avoiding the carnage of the preceding bear market. It is unfair to look only at a new bull market without also considering the preceding bear market. You need to consider at least a full market cycle to make a proper evaluation of performance. GEM has outperformed its benchmark portfolio since 2008, which includes a full market cycle.
You also should understand why GEM has outperformed over the long run. First, there are the profits from switching between U.S. and non-U.S. stocks. Since 2009, U.S. stocks have continuously outperformed non-U.S. stocks. So there has been little opportunity to profit from relative strength. Second, this recent secular bull market was one of the longest in history. This means there was also little opportunity for absolute momentum profits. It is difficult for any trend-following approach to keep up with the stock market when it is so strong.
GEM gives comfort to investors who fear a 50% market decline. You don’t cancel your fire insurance because your house hasn’t burned down over the past 10 years.
Our proprietary models use variations of dual momentum, as well as other filters. They are more adaptive to changing market conditions.
Since 1971, 73% of GEM’s gains have been long term, while nearly 100% of its losses have been short term. Long-term capital gains are now taxed at a historically low rate but might be higher when ETF investors realize their profit. Many investors don’t know that over 40% of the long-run return of the S&P 500 comes from dividends. These are taxed each year, even if they are earned in an ETF. ETFs also have to pass along taxable gains from futures, options, short positions, and stocks that settle in cash, Those who want to defer all income, including interest and dividends, can do so by applying dual momentum to a retirement account or by using a low-cost variable annuity.
One reason is that they use net rather than gross returns for non-U.S. index data. Net returns have a 30% reduction in dividends to account for foreign tax withholding. This puts GEM at a disadvantage to models that are not invested in foreign ETFs. If you are going to account for taxes, you should adjust all indices and ETFs instead of penalizing only models that invest globally.
The main reason for the performance difference, though, is their use of MSCI EAFE rather than MSCI ACWI ex-U.S. The latter includes Canada and emerging markets, while the former does not. The site says there is little difference between the two indices. That is true only if you look at both bull and bear markets. But dual momentum tries to bypass bear markets while participating fully in bull markets.
Using MSCI EAFE instead of MSCI ACWI ex-U.S. decreases GEM’s diversification and lowers its return by 130 bps annually since MSCI ACWI ex-U.S. became usable in 1989. This site does not penalize other models accordingly. They can use ETFs that provide emerging market profits.
Because you can use ensemble methods to use many things does not mean you should do so. Tobias Moskowitz, a top momentum researcher, said, ”Momentum is a phenomenon that exists at 6 to 12-month horizons. Beyond 12 months, momentum wanes…” Most academic papers on momentum use look back periods of 6 or 12 months. The ensemble paper you mentioned shows a fall-off in results with look back periods under 3 months and over 12 months.
There is some benefit in using a modest number of look back periods. We use different look back periods in our own proprietary models. But we select them based on their appropriateness reasons, not randomly. Reducing specification risk by incorporating other factors or additional trading approaches makes more sense than using an array of highly correlated look back periods.
Their statistical evidence is also suspect. You cannot correctly judge the probability of something happening after it has already happened. You need to look at results out-of-sample. A 12-month look back has more out-of-sample validation than any other look-back period. It was first introduced by Cowles & Jones in 1937 and validated by Jegadeesh & Titman in their seminal 1993 study. Other examples and additional information are in this blog post.
High stock valuation levels can mean lower expected stock returns, and low bond yields usually point to lower future bond returns. But stocks and bonds still fluctuate and create trading opportunities. In 2000, there were also high stock market valuations and low bond yields. But our GEM model had a compound annual return of 11.6% over the next 10 years, whereas a 60/40 stock/bond portfolio returned only 2.5%.
Any anomaly can lose profitability if it becomes too widely followed. However, the behavioral basis behind momentum is strong and persistent. Human nature does not readily change.
The same biases that make dual momentum work well also keep investors away from it. This is especially true of institutional investors, who often have an aversion to tactical approaches.
Shleifer and Vishny (1997) show that asset managers are afraid of strategies that deviate from popular benchmarks because investors may leave following periods of underperformance. This can create career risk. Home country bias is also a strong disincentive to invest outside one’s own country. It keeps investors from having much in non-U.S. stocks when they are stronger than U.S. stocks.
The performance of actively managed funds has, in aggregate, been inferior to the performance of passively managed funds. People have known this since the 1960s. Yet over 70% of all domestic equity funds are still actively managed. Dual momentum investing may very well show the same disconnect. Our Dual Momentum Fixed Income (DMFI) has been around for more than 10 years. It has had a higher return than the S&P 500 with the volatility of short-term bonds since 1970, yet it has attracted very little interest.
Academic research shows that over the long run, momentum for stocks works best with look-back periods of 3 to 12 months. Look-backs at the longer end of this range reduce transaction costs and increase the likelihood of long-term capital gains.
A 12-month look back was found to work well by Cowles & Jones in 1937. It has held up well ever since. Staying with this look-back period reduces concerns about data mining and seasonality bias. See our blog post “Perils of Data Mining” for more on this. Our proprietary models use multiple look-back periods based on what assets they invest in.
It makes sense to skip the last month when you are applying momentum to individual stocks. This is because individual stocks can overreact to news and then mean revert. If you use stock indices or other asset classes like we do, you do not need to skip the last month.
See here for why we prefer to use momentum with indices instead of stocks. According to this study, stock momentum profits have been insignificant since 1999. According to another study, momentum profits disappeared in the early 1990s. There is also the matter of transaction costs. See here for the most recent research on that. For more on stock momentum, see our review of the book Quantitative Momentum.
Value and momentum operate on different time frames. Value is based on longer-term mean reversion, while momentum relies on intermediate-term past performance.
Over the long run, dual momentum performs better. Historically, stocks provide the best return. So we want to be in them as long as their trend is positive. GEM uses absolute momentum to tell us that. Once we decide to be in stocks, GEM uses relative momentum to tell us whether to be in U.S. or non-U.S. stocks.
Moving averages and other forms of trend following can also work. But Zakamulin showed that absolute momentum outperformed 3 different types of moving averages on 155 years of stock market index data. Absolute momentum and reverse exponential moving averages were the only two methods that outperformed buy-and-hold with statistical significance. Our proprietary models incorporate additional ways of determining trends.
On page 98 of my book, I first determine absolute momentum using the S&P 500 index since the U.S. leads world equity markets. I also cite a supporting reference. By doing so, you may occasionally be in aggregate bonds when the trend in U.S. stocks is down, even when non-U.S. stocks are the strongest asset. On page 101 of my book, there is a flowchart that applies relative momentum before absolute momentum for those who prefer doing momentum that way. This may be what you were using.
Investors once thought stop-losses reduce return as they reduce risk exposure. But more recent research shows that stops can enhance return if they are used with care. We found dual momentum more effective than stops in reducing risk exposure and enhancing expected return. Stops add little or no value to dual momentum.
Dual momentum works best when volatility is not too high. Individual countries can have extreme volatility. This can make it difficult to get in and out of positions quickly enough without giving up considerable profit. This is also why we do not use stock sectors with our proprietary models.
Country index funds also have less liquidity and higher bid-ask spreads than broader-based index funds. That can create price impact and scalability issues, just as it does when you use momentum with individual stocks.
Most leveraged ETFs have daily resets, which makes them best suited for day trading. Tracking error can be considerable, with highly leveraged (3X) ETFs held over longer periods.
Our QBAT proprietary model is adaptive enough to use a 2X leveraged ETF at times. There is still considerable short-term volatility with dual momentum. Intra-month drawdowns greater than 20% have occasionally occurred, and larger ones may happen in the future. Investors should carefully consider that before using any kind of leverage.