Trend Following
Risk Reduction and Return Impacts
Quantitative researchers studying markets have found in paper after paper that historical price trends seem to offer information about future returns: specifically, assets that have been going up often continue to go up and assets that are going down often continue to go down. Practitioners most commonly use trend following to reduce drawdowns, though there are some that argue that trend following also improves returns.
Yet this finding is somewhat uncomfortable. The idea that we should buy when prices are higher and sell when prices are lower seems to defy economic intuition. And relying on historical price patterns to make investment decisions seems either too simple to not have been arbitraged away or too close to pseudoscience to be implementable.
We wanted to study trend following for ourselves to attempt to reconcile the academic literature with our skepticism. Assessing trend following is difficult because there are so many different ways to define when an asset is rising or when it is falling. Without a standard for measuring trend––a necessary precondition for assessing its strategic value––much of the rigor in such an analysis is lost. So we must first choose a standard to determine the trend of an asset, then determine the effectiveness of a trend-following strategy across a wide range of asset classes.
We started by looking at how trend following would have performed out of sample on a set of individual assets. There are multiple ways to define trend, and these approaches generally yield similar results. We chose to use autoregressive (AR) and moving average (MA) models that measure the strength of mean reversion and trend in recent monthly returns to forecast the next month’s return. Our first step was to train an ARMA model on each asset using in-sample data before Dec 31, 1999. This gave us a trend-following predictor that was specific to each asset. Our second step was to test these predictors on new data from Jan 2000 to Dec 2020 to see how well they would have performed out of sample. We looked at how trend following impacted both returns and drawdowns.
Figure 1: Out-of-Sample Trend Following Performance (Before Trading Costs), 2000–2020
Source: FRED, Bloomberg, Ken French website, and Verdad Analysis. When trend following sells an asset, we assume that all of the proceeds are reinvested in one-month T-bills.
Trend-following’s impact on returns is inconsistent when tested over the last 20 years. Recent research has suggested that more frequent reversals of trend have decreased the returns of trend following. Fewer extended periods of trend lead to lower returns. Our findings are not an indictment of trend-following strategies, but they do suggest, unsurprisingly, that a simple, rigid approach with monthly data is not sufficient to create a successful strategy.
But the simple approach does highlight very clearly that trend following’s real advantage is in reducing drawdowns. The improvement in drawdowns is both more consistent and more significant: trend following improved drawdowns in the asset classes we studied by an average of 9.5 percentage points. As a risk signal, it works very well, especially for assets that tend to trend. Take large growth as an example. Below we have plotted the sell signals for large growth over the past 20 years.
Figure 2: Large Growth Sell Signals (Blue Bars) vs. Index Performance
Source: Ken French website, Verdad Analysis. Blue bars indicate when the trend-following model triggers a sell decision. Index performance is on a log scale so that direction shows more clearly.
The sell signals for large growth were first triggered in February 2000, July 2007, and November 2018, all times when investors would have wanted to be reducing risk. So as a risk-off signal, trend following seems to work well and deserves a place alongside other risk signals such as a rising high-yield spread.
But for as well as it works in large growth, it works poorly in assets like small-cap value, despite the apparent 14% reduction in drawdowns. If we chart the sell signals for small-cap value, we see that the signal fires too often.
Figure 3: Small Value Sell Signals (Blue Bars) vs. Index Performance
Source: Ken French website, Verdad Analysis. Blue bars indicate when the trend-following model triggers a sell decision. Index performance is on a log scale so that direction shows more clearly.
So while trend following indeed reduces drawdowns dramatically in the summary data, an examination of the required trading to execute the strategy makes it impracticable. Small value stocks tend to have more violent trend reversals, and we were not surprised by this result, as we have written about it previously.
So, like most investment approaches, trend is not a panacea, but our approach of using a standardized trend measure to test trend out of sample reveals clearly that trend following is an excellent risk signal for assets that trend. Again, this is not a surprise, as we have used simple 200-day moving average rules for the S&P 500 in our work on countercyclical investing, but here we allowed the ARMA model to analyze the historical data and choose the appropriate time frame and tested it out of sample. And our conclusion remains the same: trend deserves its place as a risk-off signal.
Acknowledgment: Ivan is a rising sophomore at Harvard studying math and statistics. He earned an A in Math 55, which is widely considered the hardest undergraduate math class in the country. He takes a strong interest in applying statistical methods to big questions in finance, technology, and biotechnology. He hopes to intern next summer as a quantitative researcher in the financial/tech sector. I'd be delighted to introduce you to him if you're hiring.