Paper: arXiv 1404.3274
Authors: Y. Lempérière, C. Deremble, P. Seager, M. Potters, J. P. Bouchaud
Abstract
We establish the existence of anomalous excess returns based on trend following strategies across four asset classes (commodities, currencies, stock indices, bonds) and over very long time scales. We use for our studies both futures time series, that exist since 1960, and spot time series that allow us to go back to 1800 on commodities and indices. The overall t-stat of the excess returns is \(\approx 5\) since 1960 and \(\approx 10\) since 1800, after accounting for the overall upward drift of these markets. The effect is very stable, both across time and asset classes. It makes the existence of trends one of the most statistically significant anomalies in financial markets. When analyzing the trend following signal further, we find a clear saturation effect for large signals, suggesting that fundamentalist traders do not attempt to resist “weak trends”, but step in when their own signal becomes strong enough. Finally, we study the performance of trend following in the recent period. We find no sign of a statistical degradation of long trends, whereas shorter trends have significantly withered.
Complexity vs Empirical Score
- Math Complexity: 5.0/10
- Empirical Rigor: 6.0/10 (capped from 9.0: transaction costs not modelled; no out-of-sample test; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 9.0)
Why this score: This paper demonstrates strong empirical rigor through extensive backtesting over two centuries and across multiple asset classes, supported by clear statistical methodologies. Its novelty lies in the unprecedented historical depth of its analysis, confirming the stability and universality of trend following. The mathematical complexity is moderate, focusing on practical signal construction and statistical significance rather than deep theoretical derivations.
Research Flowchart
flowchart TD
A["Research Goal: Establish Existence of Trend Following Anomalous Excess Returns"] --> B{"Key Methodology: Quantify Excess Returns of Trend Following"};
B --> C["Data Input: Futures (1960+) & Spot (1800+) Time Series Across 4 Asset Classes"];
C --> D{"Computational Process: Calculate Excess Returns & T-stats (Adjusted for Market Drift)"};
D --> E["Key Finding 1: Statistically Significant Excess Returns (t-stat ≈ 5 since 1960, ≈ 10 since 1800)"];
E --> F["Key Finding 2: Saturation Effect for Large Signals & Persistence of Long Trends"];
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper establishes that trend‑following strategies generate anomalous excess returns across commodities, currencies, stock indices and bonds over very long time scales.
Source sentence
We establish the existence of anomalous excess returns based on trend following strategies across four asset classes (commodities, currencies, stock indices, bonds) and over very long time scales.
Assumes 6
- Futures are self‑financed instruments, so the risk‑free rate does not need to be subtracted when computing Sharpe ratios. execution
Source sentence
Since the P&L does not include interest earned on the capital, and futures are self-financed instruments, we do not need to subtract the risk-free rate to compute the Sharpe ratio.
- The t‑stat of the P&L is given by the Sharpe ratio multiplied by the square root of the number of years the strategy is active. statistical
Source sentence
The “t-stat” of the P&L (i.e. the fact that the average return is significantly different from zero) is therefore given by the Sharpe ratio times √ N, where N is the number of years over which the strategy is active.
- The reference price level is defined as an exponential moving average of past prices with a decay rate of n months. model
Source sentence
More precisely, we first define the reference price level at time t, ⟨p⟩n,t, as an exponential moving average of past prices (excluding p(t) itself) with a decay rate equal to n months.
- Volatility is measured as the exponential moving average of absolute monthly price changes with the same decay rate n. model
Source sentence
The signal sn(t) at the beginning of month t is constructed as: sn(t) = p(t −1) −⟨p⟩n,t−1 σn(t −1) , (1) where the volatility σn is equal to the exponential moving average of the absolute monthly price changes, with a decay rate equal to n months.
Show 2 more
- The analysis uses monthly closing prices because long simulations can only be performed on monthly data. data
Source sentence
Long simulations can often only be performed on monthly data, so we use monthly closes.
- The back‑test ignores realistic implementation costs, treating the P&L as fictitious. execution
Source sentence
This 5We call this a fictitious P&L since no attempt is made to model any realistic implemen-tation costs of the strategy.
Data
diversified futures (commodities, currencies, bonds, indices · multi-asset · monthly
- Futures price series market-prices commercial
Source sentence
2.3 The results Our history of futures starts in 1960, mostly with commodities.
- Spot price series market-prices commercial
Source sentence
We use for our studies both futures time series, that exist since 1960, and spot time series that allow us to go back to 1800 on commodities and indices.
Costs
Not modeled
Source sentence
This 5We call this a fictitious P&L since no attempt is made to model any realistic implemen-tation costs of the strategy.
Finds 5
Sharpe 0.80
- The excess returns from trend following have an overall t‑stat of approximately 5 since 1960 and approximately 10 since 1800, after accounting for market drift. statistical
Source sentence
The overall t-stat of the excess returns is $\approx 5$ since 1960 and $\approx 10$ since 1800, after accounting for the overall upward drift of these markets.
- Aggregated performance of the futures sample from 1960 shows a t‑stat of 5.9, indicating high statistical significance. statistical figure from elsewhere in the paper
Source sentence
As we can see from Fig. 1, the aggregated performance P α Qα n(t) looks well distributed in time, with an overall t-stat of 5.9, which is highly significant.
- The trend‑following effect is stable across time and across the four asset classes studied. performance
Source sentence
The effect is very stable, both across time and asset classes.
- Large trend signals exhibit a saturation effect, suggesting fundamentalist traders intervene when trends become strong. performance
Source sentence
When analyzing the trend following signal further, we find a clear saturation effect for large signals, suggesting that fundamentalist traders do not attempt to resist "weak trends", but step in when their own signal becomes strong enough. The best fit to the data is provided by the hyperbolic tangent, which suggests a saturation of the signal for large values. presence of trends.
- Recent short‑term trends have withered, but long‑term trends show no statistical degradation. performance
Source sentence
Finally, we study the performance of trend following in the recent period. We find no sign of a statistical degradation of long trends, whereas shorter trends have significantly withered.
Checked by others
Supported by
- Trends and Reversion in Financial Markets on Time Scales from Minutes to Decades
Compare the findings
This paper: The trend‑following effect is stable across time and across the four asset classes studied.
That paper: Across asset classes, markets trend on time scales of a few hours to a few years and revert on shorter and longer scales.
Qualified by
- Trends and Reversion in Financial Markets on Time Scales from Minutes to Decades
Compare the findings
This paper: The trend‑following effect is stable across time and across the four asset classes studied.
That paper: Regression coefficients b (trend persistence) and c (trend reversion) are small but statistically highly significant; b decreases over decades and may vanish around 2030.
Code
No repository by the authors found in the paper.
Extracted 2026-11-15 · same card as JSON · A language model chose these items: check the source sentence before relying on one.