A trend following deep dive: The dynamics of dispersion
By Photis Georgiades; Howard Wong; Yash Panjabi; Tarek Abou Zeid; Otto van Hemert, Man Group
Published: 24 November 2025
Dispersion among trend followers is a nuanced, though seldom explored, topic. Intuition might suggest that there is little scope for differentiation as a trend follower. In reality, the mechanisms for capturing trends are vast, with different expressions and parameters leading to diverse outcomes.
Examining the constituents of the Société Générale Index of trend followers (SG Trend Index)1 from its inception in 2000, we found that the delta between the best and worst performer each calendar year is typically around 10 percentage points. In some years, it rises to double that.
In this paper, we examine the portfolio properties that can lead to dispersion and consider how each contributes to differentiation in trend-following outcomes.
Dispersion between trend followers
To quantify dispersion, in Figure 1, we plot the risk-adjusted (scaled to 10% annual volatility for comparability) calendar year performance for each of the (typically) 10 SG Trend Index constituents,2 with the navy and pink lines tracking the best and worst constituents for each calendar year.
Figure 1: SG Trend Index constituents, risk-adjusted annual performance (2000-2024)
Source: HFR, Inc., With Intelligence, Bloomberg, Man Group Database. Jan 2000 – Dec 2024.
In most years, the majority of constituents are directionally aligned. However, Figure 1 demonstrates that there is typically a sizeable magnitude of dispersion in the absolute performance of index constituents.
In some, albeit rarer, cases, the source of dispersion is idiosyncratic. To take one example, in 2009, one constituent stood in double-digit positive territory, while the rest of the pack logged flat or negative returns. This positive outlier turned out to be a discretionary fund which had found its way into the index and was later removed. Aside from idiosyncratic cases such as this one, dispersion generally stems from the core design parameters of a trend-following system.
Capturing performance dispersion through trend-following proxies
In a bid to model the dispersion we observe between index constituents, we have created a set of 20 trend-following proxies that we believe model the variations in the type of trend-following systems employed.
Below, we outline four binary choices to construct 16 (or 24) core trend proxies:
Speed
(a) Fast: two-to-three month holding period, one-month lookback for position sizing
(b) Slow: six-month holding period, 12-month lookback for position sizing
Inclusion of carry
(a) No carry
(b) 15% allocation to currencies and fixed income carry
Inclusion of alternative markets
(a) Purely traditional: futures and FX forwards across commodities, currencies, fixed income and stocks (around 150 markets)
(b) Including alternative markets: such as power, synthetic credit indices and interest rate swaps
Allocations
(a) Equal risk weight: to each asset class (and equal risk weight to markets within each asset class)
(b) Proportional to capacity: based on daily dollar volume and exchange limits
In addition, we consider four trend-following proxies where we apply asset class tilts in the allocations step four, bringing the total number to 20. In these proxies specifically, we ‘fix’ parameters one to three above to better isolate the impact of active allocation decisions. We utilise a medium speed (four-month holding period), excluding carry, trading traditional markets only and applying equal market risk allocations within each asset class. However, instead of applying an equal 25% allocation to each asset class as we do in the first 16 proxies, we apply a 40% overweight to one of the asset classes and 20% to the remaining three. The rationale for including these additional proxies stems from evidence that some managers consciously tilt their portfolios towards certain asset classes.
The opposite school of thought purports that all markets are equal in terms of their expected trend-following information ratio, and by that token, the goal should be to maximise diversification by spreading risk across markets. In practice, this takes into account the covariance matrix of markets, subsequently skewing allocation to those that are less correlated. For the purposes of this paper, however, we use an equal-risk-weight approach (proxy parameter 4a) to reasonably proxy for this. Additionally, parameter 4b accounts for the presence of capacity constraints, where optimising for strategies with larger assets under management would tilt the weights towards higher capacity markets.
Lastly, we subtract an annualised transaction cost of 2% for fast and 1% for slow implementations, respectively, to account for turnover differences. We also apply a 1.5% management and 20% performance fee (with high watermark and end-of-year fee crystallisation, as is common) and assume 50% of cash is unencumbered, earning the Treasury bill rate.
Using this framework, in Figure 2, we replicate the analysis from Figure 1 but instead plot the annual performance for the 20 trend-following proxy portfolios.
Figure 2: Proxy trend strategies, annual performance (2000-2024)
Source: HFR, Inc., With Intelligence, Bloomberg, Man Group Database. Jan 2000 – Dec 2024.
Overall, the results show that the proxy portfolios are reasonably robust in capturing both the yearly pattern of returns and dispersion, with the delta between the best and worst performers for each calendar year approximately in line with that of the actual SG Trend Index constituents. We can therefore leverage the insights from our proxy portfolios to better understand the practical dispersion between index constituents.
Analysing the trend proxy portfolios and binary parameters
Taking a more holistic view of the long-term performance of our 20 proxy portfolios, in Figure 3, we observe that dispersion in the initial period from 2000 to 2006 is relatively less pronounced compared with the period following the onset of the Global Financial Crisis (GFC) in early 2007. Most trend-followers did similarly well irrespective of design. Indeed, the absolute average pairwise correlation between the portfolios was 0.87 in the early period, falling to 0.78 from 2007 onwards.
Figure 3: Proxy trend strategies cumulative performance (2000-2024)
Source: Man Group Database. Jan 2000 – Dec 2024. The performance of the proxy portfolios is hypothetical and back tested. See Important Information.
The more meaningful dispersion in the post-GFC period was principally driven by design choices. 2009-19 was widely regarded as the ‘CTA winter’, where traditional trend struggled amid the currents of the Federal Reserve put. Portfolios that capitalised on the proliferation of liquid, tradable alternative markets, which basked in plentiful trends, along with those that followed a slower, more passive approach naturally navigated this period better. Although, this is not a persistent effect through time, as certain properties or design choices perform better in different environments.
For many investors, trend following’s role in a portfolio is as a diversifying, crisis-offset allocation, and therefore the design parameters that drive dispersion in such periods are a crucial consideration for allocators. To explore this, in Figure 4, we decompose long-term performance to uncover which of the parameters give rise to improved crisis alpha properties.
We define crisis periods as the peak-to-trough time periods when the S&P 500 experiences a 15% or worse drawdown.
Figure 4: Crisis versus non-crisis performance for the 20 trend proxies (2000-2024)
Source: Man Group Database. Jan 2000 – Dec 2024.
Our first observation is that all of the proxy portfolios sit below the diagonal line, underscoring that trend following’s crisis Sharpe is comparatively better than its all- periods Sharpe. This accords with empirical evidence that trend following’s performance is most positive in the worst quintile of equity returns.
Where each of these portfolios sit below the diagonal line, however, is influenced by the different binary parameters of each proxy.
Speed
Starting with speed, we observe a clustering of faster trend (pink markers) further along the x-axis. This indicates that faster portfolios have historically tended to perform better during periods of crisis.
Alternative markets
Trading alternative markets, represented by the triangles in Figure 4, leads to better overall performance, as shown by the clustering of triangles in the top right-hand direction.
Notably, however, there is an interesting interplay with the effects of speed, with the additivity of alternative markets more apparent in slower trend systems.
Asset class tilts
Active asset class tilts can also drive meaningful bifurcation in trend-following outcomes.
Notably, the relative performance between any of the asset class-tilted portfolios can be significant in any given year. 2019 is a pertinent example of this, when being overweight fixed income led to a marked improvement in performance relative to most other implementations. Despite this, there is no single asset class tilt that consistently outperforms year-after-year.
Carry
Trend followers often complement their core trend allocation with a satellite allocation to diversifying, non-trend content. In doing so, they aim to provide diversifying performance during more challenging periods for trend signals. Carry is a natural fit to achieve this, given that stable, rangebound markets, which are not conducive to most trend signals, are often associated with beneficial periods for carry.
The size of allocation to carry is variable among trend followers and can periodically have meaningful impact on performance. In 2016 and 2023, for example, an allocation to carry was beneficial. As it pertains to crisis performance, however, carry does not have a material impact, with other parameters being more prominent drivers of dispersion.
Conclusion
Our analysis illustrates that even among seemingly similar trend-following CTAs, subtle differences in key parameters – such as speed, market set, carry and allocation methods – can drive considerable performance dispersion, particularly during periods of heightened market stress. Our 20 proxy portfolios offer a behind-the-scenes insight into the mechanics of the SG Trend Index universe, while also serving as a useful framework for internal benchmarking and supporting continued refinement in risk management.
This is an abridged version of a paper originally published on Man Institute.
1. Constituent data is sourced from third-party databases with which Man Group has a data licence. Past performance is not indicative of future results. The performance of SG Trend Index constituents is not necessarily representative of the performance of any Man Group product.
