Paper: arXiv 2309.08800

Authors: Yichi Zhang, Mihai Cucuringu, Alexander Y. Shestopaloff, Stefan Zohren

Abstract

In multivariate time series systems, lead-lag relationships reveal dependencies between time series when they are shifted in time relative to each other. Uncovering such relationships is valuable in downstream tasks, such as control, forecasting, and clustering. By understanding the temporal dependencies between different time series, one can better comprehend the complex interactions and patterns within the system. We develop a cluster-driven methodology based on dynamic time warping for robust detection of lead-lag relationships in lagged multi-factor models. We establish connections to the multireference alignment problem for both the homogeneous and heterogeneous settings. Since multivariate time series are ubiquitous in a wide range of domains, we demonstrate that our algorithm is able to robustly detect lead-lag relationships in financial markets, which can be subsequently leveraged in trading strategies with significant economic benefits.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper presents advanced mathematics including Dynamic Time Warping, multireference alignment connections, and lagged multi-factor models, but also demonstrates robust empirical validation with backtest-ready trading strategies and performance metrics.

Research Flowchart

  flowchart TD
  A["Research Goal: Detect Lead-Lag Relationships in Multivariate Time Series"] --> B["Input: Multivariate Financial Time Series Data"]
  B --> C["Preprocessing: Stationarity & Alignment"]
  C --> D["Methodology: Dynamic Time Warping DTW<br>with Multireference Alignment MRA"]
  D --> E["Process: Clustering & Temporal Shift Estimation"]
  E --> F["Output: Estimated Lead-Lag Matrices"]
  F --> G["Outcome: Identified Cluster-Level Lead-Lag Structures"]
  G --> H["Application: Trading Strategies<br>Significant Economic Benefits"]

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 proposes a computationally scalable DTW-based framework with clustering for lead‑lag detection in high‑dimensional time series.

Source sentence
We introduce a computationally scalable framework for lead-lag detection in high-dimensional time series based on DTW, with clustering used as a denoising step.

Assumes 4

  • Each asset is assumed to have β = 1 exposure to the market. model
    Source sentence
    we assume each asset has β = 1 exposure to the market
  • PnL is rescaled to a target annualized volatility of 0.15. execution
    Source sentence
    We rescale the PnL by their volatility to target equal risk assignment, and set our annualized volatility target σtgt to be 0.15.
  • Days with >10% zero returns and assets with >50% zero returns are dropped. data
    Source sentence
    we drop the days for which more than 10% of the equities have zero returns as well as the equities for which more than 50% of days have zero returns
  • Extreme excess returns are winsorized at ±0.15. data
    Source sentence
    we winsorize the extreme value of excess returns for which any value is larger than 0.15 or smaller than -0.15

Data

  • Wharton’s CRSP Equity · Wharton’s CRSP market-prices commercial
    2000/01/03‑2019/12/31 · 679 equities
    Source sentence
    Wharton’s CRSP Equity Daily 679 2000/01/03 2019/12/31 5211
  • Wharton’s CRSP ETF · Wharton’s CRSP market-prices commercial
    2006/04/12‑2019/07/01 · 14 ETFs
    Source sentence
    Wharton’s CRSP ETF Daily 14 2006/04/12 2019/07/01 3324
  • Pinnacle Data Corp Futures · Pinnacle Data Corp market-prices commercial
    2000/01/05‑2020/10/16 · 52 futures contracts
    Source sentence
    Pinnacle Data Corp Futures Daily 52 2000/01/05 2020/10/16 5166

Costs

Unclear No cost treatment was found in the passages the extractor read; that is not evidence that costs are not modeled.

Finds 3

  • DTW KMed Med achieved the highest Sharpe ratio (SR = 0.93) after 2008, outperforming other methods. performance
    Source sentence
    DTW KMed Med emerged as the most profitable strategy, clearly outperforming others with an SR of 0.93.
  • Before 2008, DTW KMed Mod and DTW KMed Med had higher Sharpe ratios than other algorithms in the Gβ strategy. performance
    Source sentence
    In the Gβ strategy, before 2008, the Sharpe ratio (SR) of DTW KMed Mod and DTW KMed Med outperformed the other algorithms.
  • The proposed DTW‑based algorithms show promising Sharpe ratios on financial data, indicating economic benefits versus the benchmark. performance
    Source sentence
    Our proposed algorithms show promising Sharpe Ratios when applied to financial data sets, indicating their potential economic benefits compared to the benchmark.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 8 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive (1 by independent authors): Follow the Leader: Enhancing Systematic Trend-Following … (uses its method or data)

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.