Absolute momentum or time series momentum, as trends are called in academia, can be used to time virtually any price series (asset prices, composite indices, equity curves, factors and so on) in such a simple manner, that to be able to gain an advantage over the market that way, seems a bit suspect. Not to mention, that it seems in clear violation of classic financial theories like the efficient market hypothesis.
As I use a trend overlay throughout my portfolio, I thought, I better try to figure out, if the concept can be trusted to be useful in practice or if it is just a figment of people´s imagination. Does it work, how will the practical implementation look and feel and will it likely continue to be a viable approach in the future?
To dig into the subject, we need to be clear about what we are talking about. There are two types of momentum, that lead to very different results in practice:
- The classic momentum factor – relative momentum or relative strength, which has been deeply researched with the seminal work by Jegadeesh and Titman written in 1993. Relative momentum compares the price series of different assets to each other and goes long the strongest assets and short the weakest assets or simply long the strongest assets in a long- only portfolio. It has historically increased returns substantially, but it also led to higher volatility and drawdowns.
- Absolute momentum or trend only looks at a single price series and determines the direction in which it is moving. Trend following goes long the assets in an upward direction and short the assets moving downwards. In a long- only application it moves to cash when the trend points downwards. This is what this article looks at primarily. An important difference to relative strength momentum is, that long- only trend following historically has not significantly enhanced returns, but rather reduced drawdowns and volatility. It is primarily a protective strategy and it´s value lies in providing higher risk-adjusted returns over time by avoiding major drawdowns.
Methods for determining trends
There are different ways to figure out the trend a market is in, but the most important factor is the time frame you look at. Several studies have come up with the following time frames separating trending and mean reverting tendencies in the structure of many markets:
- Short-Term Momentum (1-month) – exhibits a reversal in returns
- Intermediate-Term Momentum (3-15 months) – exhibits a continuation in returns: the trend following strategies, I use, fall into that time frame. I call methods, that use a look-back period around one year long-term trend following, looking at 3-6 months determines the medium-term trend.
- Very Long-Term Momentum (1 to 5 years) – exhibits a reversal in returns: Trend strategies start to break down as mean reversion takes over and the market returns to fundamental values.
My main takeaway here is, that trend has shown up historically over a robust, wide time range in the intermediate term. Mean reverting (e.g. value) strategies might be especially profitable for investors with a long time horizon. Short term strategies are more challenging as more and more random noise confuses the signal the shorter the time frame becomes and mean reversion becomes prevalent.
Any method used to determine trend will catch the movement (if it materializes) in a very similar fashion. I would say, it really doesn’t matter much, choose the method that intuitively makes most sense to you – differences in outcome will largely be due to chance. Apart from the look- back period, the second factor that does make a difference is the sensitivity of the trend signal. The more the signal is smoothed (by averaging prices or looking at prices less frequently), the less sensitive it is to changes in trend, but the less it will suffer from whipsaws – the nemesis of any trend methodology. Here are some common methods you can use in a long-only portfolio (exit or shorting rules usually are the reverse of the entries):
- Comparing two price points is the simplest: If the asset shows a positive excess return (higher than treasury bills) over the past X (3-15) months go or stay long. Sensitivity depends on the price frequency used: monthly prices will be less sensitive than daily prices for example.
- Break out methods: if the asset´s price is higher than the X-day/ month high go long.
- Moving average: if the asset´s price is higher than the X-day/ month moving average go long.
- Moving average crossover: if the shorter-term MA is higher than the longer term MA go long. This has the effect of smoothing signals when using daily prices.
- Trend lines: same basic idea as a moving average, but not easily quantifiable, subject to discretion and therefore not very useful.
In general it makes sense to align the look-back period and the sensitivity of your signal with your strategy´s intentions – e.g. do you want to be in the market for the long-term or change positions more frequently? Most studies use a 12-month look-back, monthly prices and often a fixed holding period.
For the long-term tactical management of the adaptive global asset allocation in my portfolio, I found the use of a single moving average with a slightly longer time period, looking at daily prices (I use a 275-day moving average simply to be a little bit different from the commonly used 200-day MA), a good compromise between sensitivity and whipsaws and very simple to implement. The idea is to protect the portfolio from the really big drawdowns above -15% to -20% and avoid very high turnover High volatility and tail risks increasingly cluster in markets that are already downtrending significantly.
A similar result can be reached with a slightly shorter-term moving average using monthly prices (e.g. this simple, but effective method uses a 10-month moving average).
Trying to filter the quality of a trend using volume, noise filters or similar methods, adds complexity, but not necessarily reliability – I prefer to stay simple.
One approach that makes intuitive sense to me, is determining the look back period of the trend signal as a function of the volatility of the asset – a more volatile asset would need a bit more wiggle room to avoid whipsaws. This could be achieved by a different length moving average depending on how volatile the asset is. In practice implementation could be a bit of a hassle, but I found an application that has done something along that line and tested a sector rotation model with promising results using a trailing stop loss model. This will be worth more research.
What to expect in practice from using a long-term trend following overlay in a portfolio
Market timing doesn’t work, is something you hear very often and when you backtest a trend following method that tactically times a long-only diversified portfolio you can see where this is coming from: absolute returns are not significantly better using a long-term trend timing method, used on a portfolio of single equities returns are even likely to underperform buy and hold.
But before trashing the whole idea let´s go step by step to analyze expected results over time and how they can indeed be quite useful.
Running a simple simulation using the tool portfoliovisualizer on US equities exemplifies how this tends to play out over time (Here you can find an independent test on 235 indices with very similar results across the board and across different historical periods): I looked at the S&P 500 SPY ETF from 1995 to mid 2017 (see chart below): fully invested when price is above 275-day moving average, otherwise in cash. Returns are not significantly higher than buy and hold, but volatility goes down significantly and drawdowns a lot, increasing risk-adjusted returns – which is what really matters.
What happens over time is, that the method goes to cash in bear markets, where it outperforms. It cannot outperform during bull markets as it acts identical to the buy and hold asset – it´s 100% invested. It underperforms when the market goes down a fair bit (below the moving average), but quickly recovers – a whipsaw. It also takes some time to get invested after a prolonged downturn and therefore misses the first part of the new rally. These features are inherent in the method and will always play out in that fashion – the real question is whether the outperforming or the underperforming periods will gain the upper hand over time.

The features are clearly visible in the comparison of annual returns over time: large bear market drawdowns are almost completely avoided – as in 2002, 2003 and 2008. In the recovery phase after major bear markets the trend portfolio lags behind – as in 2003 and 2009. Whipsaws cause significant underperformance – as happened in 2011, 2015 and 2016. This underperformance doesn’t look like much in the chart above, but can be rather hard to tolerate in practice. A good mental model would be to view these stretches of underperformance as paying an insurance premium, that protects against severe drawdowns.

Shortening the test period to the time of the current bull market (everything else being equal) in the chart below, you can see that the initial underperformance coming out of the gate in 2009 only got worse all the way until today. You would have looked at a buy and hold investor with envy as he made twice as much money over the last 8 years.
Of course, nobody could have known in advance that the corrections in 2015 and 2016 were just blips on the screen rather than more severe downturns. The advantage of the trend following method will only become apparent in the next downturn, where it will likely outperform enough (depending on the size of the downturn) to come out ahead in the long run. In the 80s and 90s the same underperformance happened – it simply is the nature of the method. But, as long as there are significant bear markets, it should work over each cycle – and I don´t see why there suddenly shouldn’t be.
This long underperformance is a strong reason for the continuing viability of the method, as few investors are able to tolerate it and stick with a trend strategy – it implies that today is actually a great time to implement a trend strategy, because it is likely to outperform after a period of underperformance.

Limitations of trend following
The evidence that long-term trend following as a tactical overlay on a diversified global asset portfolio, consisting of broad based ETF, offers distinct advantages to an investor, seems to be quite solid. This has been shown to work all the way down to sector ETF, but Gary Antonacci in his book „Dual Momentum“, covering all aspects of the the subject, makes it work very efficiently and well with just three ETF (US equities, Developed World equities, bonds). I have found the best compromise for me to be about 20 broad based ETF covering all asset classes (each asset class is divided into several sub-class ETF e.g. equities into US, Europe, Asia, emerging markets and international sectors).
In some areas, though, problems appear:
- Trend following using individual stocks
- Over shorter time frames
- On the short side of equities and other markets with an upward bias (this is simply because the assets are upwards biased, skewing the probabilities in favor of the long side and also because your maximum win shorting an asset cannot exceed 100%, effectively a method for cutting your winners, which goes against the essential concept underlying trend following – see below)
A good explanation, I found, is that viable single stock and short term trend strategies are much more difficult to find, because these exhibit a much higher level of noise in comparison to the valid trend signal. It all boils down to the assets volatility in comparison to the size of the trend – if that difference becomes too small, the whipsaw losses overwhelm the profitability of the strategy. Add to that additional fees, slippage etc..
When you add many stocks together, the volatile individual security´s noise cancels out and the market direction (the signal) becomes much clearer.
If you want to use a tactical trend overlay on a portfolio of individual stocks, it´s best to monitor the trend of the portfolio´s benchmark (e.g. S&P 500) and use that to scale general equity exposure up or down. Eliminating the short side also leads to a higher chance of success on single stocks than a long/short strategy, but it it is unlikely to actually perform better than simply using an index due to higher volatilities and cost – I don´t hold any individual stocks in my portfolio for various reasons.
Both momentum effects, relative and absolute momentum, are very well documented in academic studies (see comprehensive charts and tables here), as well as in practical applications. Trend following works best in the most challenging market conditions, protecting your portfolio from severe drawdowns. But you have to be prepared to lag behind a buy and hold approach for extended periods of time (fortunately only as a relative loser as this should generally coincide with profitable times). Prolonged temporary underperformance is a feature virtually every outperforming strategy has (e.g. investing in different factors). Trend following can prove immensely valuable, as it may save you from panic or a blow up in bear markets, enable you to deploy your preserved capital at the most opportune times (when capital usually is scarce) and help you stay the course over time. It provides additional diversification across time.
Why does it work?
A simple question with complex answers. There is a thorough discussion of many rational and behavioral explanations of trend following in Gary Antonacci´s book. They boil down to the markets displaying an initial underreaction to change followed by a delayed overreaction, due to investor´s behavior.
A more technical explanation lies in the trend following adage: let your winners ride and cut your losers short. Because financial markets are not normally distributed, but exhibit long tails (i.e. more extremely large wins and losses), cutting the left tail – avoiding large losses – while letting the right tail run will lead to a winning outcome over time.
In this context, I will look at the systematic application of trend following using long/short futures over a large number of markets, as is the hallmark of the CTA industry in a future post.
9 comments