Paper: arXiv 2501.16772

Authors: Sara A. Safari, Christof Schmidhuber

Abstract

We empirically analyze the reversion of financial market trends with time horizons ranging from minutes to decades. The analysis covers equities, interest rates, currencies and commodities and combines 14 years of futures tick data, 30 years of daily futures prices, 330 years of monthly asset prices, and yearly financial data since medieval times. Across asset classes, we find that markets are in a trending regime on time scales that range from a few hours to a few years, while they are in a reversion regime on shorter and longer time scales. In the trending regime, weak trends tend to persist, which can be explained by herding behavior of investors. However, in this regime trends tend to revert before they become strong enough to be statistically significant, which can be interpreted as a return of asset prices to their intrinsic value. In the reversion regime, we find the opposite pattern: weak trends tend to revert, while those trends that become statistically significant tend to persist. Our results provide a set of empirical tests of theoretical models of financial markets. We interpret them in the light of a recently proposed lattice gas model, where the lattice represents the social network of traders, the gas molecules represent the shares of financial assets, and efficient markets correspond to the critical point. If this model is accurate, the lattice gas must be near this critical point on time scales from 1 hour to a few days, with a correlation time of a few years.

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced statistical and econometric modeling, including cubic polynomial regression and references to statistical physics concepts like lattice gas models, indicating high mathematical density. Empirically, it is exceptionally rigorous, leveraging decades of multi-asset, multi-frequency historical data (from tick to medieval) and providing detailed methodology for data aggregation and parameter estimation, making it highly backtest-ready.

Research Flowchart

  flowchart TD
  A["Research Goal: Analyze Market Trend Reversion<br>across Time Scales"] --> B["Methodology: Multi-Scale Empirical Analysis"]
  
  B --> C{"Data Inputs"}
  C --> C1["Minutes: 14 Yrs Futures Tick Data"]
  C --> C2["Days: 30 Yrs Daily Futures"]
  C --> C3["Months: 330 Yrs Monthly Prices"]
  C --> C4["Years: Medieval Data"]
  
  C --> D["Computational Process<br>Identify Trending vs Reversion Regimes"]
  
  D --> E{"Key Findings"}
  E --> E1["Trending Regime: 1hr - Years<br>Weak trends persist (herding)<br>Strong trends revert (intrinsic value)"]
  E --> E2["Reversion Regime: <br>Shorter & Longer scales<br>Weak trends revert<br>Significant trends persist"]
  E --> E3["Lattice Gas Model Fit<br>Critical point near market efficiency<br>Correlation time: few years"]

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 empirically establishes scaling facts about financial market trends across time scales that theoretical models must reproduce.

Source sentence
In the current paper, we focus on establishing the empirical facts, i.e., the observed scaling bahaviour that such theoretical models must reproduce

Assumes 6

  • Intraday trending behavior is independent of asset class and time zone. model
    Source sentence
    Hereby, we implicitly assume that the intraday trending behavior of markets is independent of the asset class and the time zone
  • Markets are efficient so any trading profit just compensates for risk. market
    Source sentence
    We assume that markets are efficient in the sense that any trading profit just compensates for this risk.
  • Market returns cannot be assumed to be independent, identically distributed normal random variables. statistical
    Source sentence
    Since market returns cannot be assumed to be independent, identically distributed normal random vari-ables, we cannot trust standard formulas for the t-statistics
  • End‑of‑minute prices are sufficient for intraday analysis, without using full tick data. data
    Source sentence
    The analysis is based on end-of-minute prices (last traded price in each minute), rather than the whole set of intra-minute tick data. These end-of-minute prices are sufficient for our purposes
Show 2 more
  • Bootstrapping with 5000 samples provides reliable error estimates for regression coefficients. statistical
    Source sentence
    The error of the coefficients and their t-statistics were computed by bootstrapping, using 5000 bootstrapping samples
  • Adjusted R‑squared is estimated via cross‑validation (15‑fold for daily, 12‑fold for minute data). statistical
    Source sentence
    ... The adjusted R-squared, reported here in basis points (0.01%), has been estimated out-of-sample by 12-fold cross-validation

Data

  • Futures tick data (intraday) · TickData market-prices commercial
    2010-2023 · global futures across equities, interest rates, currencies, commodities
    Source sentence
    Our data were provided by TickData (https://www.tickdata.com) and cover the 14-year period 2010-2023.
  • Daily futures prices market-prices unknown
    30 years · futures across equities, interest rates, currencies, commodities
    Source sentence
    The analysis covers equities, interest rates, currencies and commodities and combines 14 years of futures tick data, 30 years of daily futures prices
  • Monthly asset price series market-prices unknown
    330 years · diversified set of markets
    Source sentence
    330 years of monthly asset prices
  • Yearly financial data since medieval times market-prices unknown
    since medieval times · various asset classes
    Source sentence
    yearly financial data since medieval times
  • Monthly data from the Global Financial Database · Global Financial Database market-prices commercial
    equity indices, interest rates, currencies across many countries
    Source sentence
    We use monthly data from the “Global Financial Database”, covering the diversified set of markets shown in table 7.

Costs

Not modeled A trading or execution result is reported without costs.

Source sentence
There are trading costs. Moreover, there is the risk that the price does not revert but runs away due to an underlying trend or a buy/sell order by a large investor. We assume that markets are efficient in the sense that any trading profit just compensates for this risk.

Finds 4

  • Across asset classes, markets trend on time scales of a few hours to a few years and revert on shorter and longer scales. statistical
    Source sentence
    Across asset classes, we find that markets are in a trending regime on time scales that range from a few hours to a few years, while they are in a reversion regime on shorter and longer time scales.
  • Regression coefficients b (trend persistence) and c (trend reversion) are small but statistically highly significant; b decreases over decades and may vanish around 2030. statistical
    Source sentence
    b and c are statistically highly significant out-of-sample, although they seem to have decreased over the decades
  • Trends only persist when the horizon is between a few days and one or two years. statistical
    Source sentence
    While the strength of reversion c seems to be approximately constant, we find that the persistence of trends b depends on the time horizon of the trend
  • The lattice‑gas network model must be near its critical point on time scales from 1 hour to a few days, with a correlation time of a few years. theoretical
    Source sentence
    If this model is accurate, the lattice gas must be near this critical point on time scales from 1 hour to a few days, with a correlation time of a few years.

Checked by others

Supported by

  • Time Series Momentum (2011)
    On the question: What is the typical time horizon over which return trends or momentum persist before reversing?
    Compare the findings

    This paper: Trends only persist when the horizon is between a few days and one or two years.

    That paper: Return persistence lasts 1 to 12 months and partially reverses over longer horizons, consistent with sentiment theories of under‑reaction and delayed over‑reaction.

    Both find persistence up to ~12 months; A extends persistence to 1‑2 years, while B shows reversal beyond 12 months.

  • Two centuries of trend following (2014)
    On the question: Does a trend‑following effect exist across asset classes and time horizons?
    Compare the findings

    This paper: Across asset classes, markets trend on time scales of a few hours to a few years and revert on shorter and longer scales.

    That paper: The trend‑following effect is stable across time and across the four asset classes studied.

    Both find a trend effect: A reports stability across four asset classes; B shows trends on scales of a few hours to a few years, supporting A.

Qualified by

  • Two centuries of trend following (2014)
    On the question: Do trends generate persistent returns across different markets and time horizons?
    Compare the findings

    This 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.

    That paper: The trend‑following effect is stable across time and across the four asset classes studied.

    Both find a trend effect; A reports it stable across four asset classes, while B finds a small but significant persistence (b) that declines over decades and could vanish around 2030.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.