Why Trend Systems Work

Another Perspective

Trend following is a cornerstone for many successful trading strategies. But why do trend-based trading systems actually work? The reasons are rooted in fundamental, behavioral, and statistical aspects, such as:

• Prices react to long-term economic cycles and fundamental dynamics such as inflation regimes, fiscal and monetary policy, trade imbalances etc.

• Markets are influenced by human emotions such as fear, greed, herding, and loss aversion, which often results in overreaction and underreaction to news and a subsequent adjustment process.

• Returns are not normally distributed and have a fat tail. The fat tail means that there is an unusually large number of directional price moves, or runs, that are longer than would be expected if returns were randomly distributed.

The fat tail is a statistical phenomenon, backed by more than a century of empirical evidence. The extreme case is an outlier move (cocoa is back again, by the way) - an extreme price excursion into either tail of the return distribution.

Since my colleague

Moritz Heiden

describes himself as a “Professional coin flipper” in his Twitter profile, he might object to my choice of using a simple coin flipping example as a classic way to produce a random distribution, but let’s do it anyway.

In a random distribution of 100 fair coin tosses, we expect:

• ~50 sequences of 1 (head or tail followed by the opposite)

• ~25 sequences of 2 heads or tails

• ~12 sequences of 3 in a row

• ~6 sequences of 4 in a row

• ~3 sequences of 5

• ~1–2 sequences of 6

• Fewer than 1 sequence of 7 in a row

If price moves are substituted for coin flips, then heads would be a move up and tails a move down, albeit of unknown rather than certain and equal magnitude each time. While trend following would not work in our coin flipping example, it does work in the financial markets where returns are not normally distributed. Instead of 1 or 2 runs of six consecutive price moves, we may see a sequence of ten. It’s a small edge, but it is enough to make a trend following system profitable if losses are kept small and winning trades - those that successfully participate in a price trend - are allowed to run on.

Analyzing 8,212 daily bars of Japanese Yen futures (CME):

• The longest consecutive run (up or down): 18 bars

• Number of runs ≥10 bars: 27

• Frequency: 0.33%, vs. random expectation: 0.19%

In Nasdaq futures (CME) over 7,331 bars:

• Longest run: 14 bars

• Runs ≥10 bars: 29

• Frequency: 0.39%, vs. random expectation: 0.19%

Similar patterns appear across many other markets.

The chart below illustrates the difference between random run frequencies (shaded area) and actual market data (blue line), based on markets traded by Takahe Capital. You’ll notice a deficit in short runs and a surplus of longer runs—representing the fat-tailed behavior.

These longer-than-expected directional moves are what trend following systems are designed to capture.

Probability of Runs: Random vs. Actual