Paper: arXiv 2508.02758

Authors: Yihao Ang, Qiang Wang, Qiang Huang, Yifan Bao, Xinyu Xi, Anthony K. H. Tung, Chen Jin, Zhiyong Huang

Abstract

Synthetic time series are essential tools for data augmentation, stress testing, and algorithmic prototyping in quantitative finance. However, in cryptocurrency markets, characterized by 24/7 trading, extreme volatility, and rapid regime shifts, existing Time Series Generation (TSG) methods and benchmarks often fall short, jeopardizing practical utility. Most prior work (1) targets non-financial or traditional financial domains, (2) focuses narrowly on classification and forecasting while neglecting crypto-specific complexities, and (3) lacks critical financial evaluations, particularly for trading applications. To address these gaps, we introduce \textsf{“CTBench”}, the first comprehensive TSG benchmark tailored for the cryptocurrency domain. \textsf{“CTBench”} curates an open-source dataset from 452 tokens and evaluates TSG models across 13 metrics spanning 5 key dimensions: forecasting accuracy, rank fidelity, trading performance, risk assessment, and computational efficiency. A key innovation is a dual-task evaluation framework: (1) the \emph{“Predictive Utility”} task measures how well synthetic data preserves temporal and cross-sectional patterns for forecasting, while (2) the \emph{“Statistical Arbitrage”} task assesses whether reconstructed series support mean-reverting signals for trading. We benchmark eight representative models from five methodological families over four distinct market regimes, uncovering trade-offs between statistical fidelity and real-world profitability. Notably, \textsf{“CTBench”} offers model ranking analysis and actionable guidance for selecting and deploying TSG models in crypto analytics and strategy development.

Complexity vs Empirical Score

  • Math Complexity: 4.5/10
  • Empirical Rigor: 6.0/10 (capped from 8.5: transaction costs only partially modelled; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 8.5)

Why this score: The paper presents a comprehensive benchmark with extensive empirical evaluation on real cryptocurrency data (452 tokens), multiple market regimes, and financial metrics (Sharpe, MDD, VaR, etc.), indicating high data/implementation rigor. The mathematics is present (definitions of log-returns, rolling windows, and model descriptions) but focuses more on evaluation framework and practical insights rather than dense theoretical derivations.

Research Flowchart

  flowchart TD
  A["Research Goal: <br>Benchmark TSG models<br>for Crypto Time Series"] --> B["Data Curation<br>(452 tokens, 4 regimes)"]
  B --> C["Methodology<br>(Dual-Task Framework)"]
  C --> D["Evaluation Metrics<br>(13 metrics across 5 dimensions)"]
  D --> E["Computational Process<br>(Benchmark 8 Models)"]
  E --> F["Key Findings<br>(Trade-offs & Guidance)"]

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 CTBench, a comprehensive benchmark for cryptocurrency time series generation, including a curated dataset and dual-task evaluation framework.

Source sentence
To address these gaps, we introduce CTBench, the first comprehensive TSG benchmark tailored for the cryptocurrency domain.

Assumes 3

  • Cryptocurrency markets operate continuously 24/7 with extreme volatility and irregular liquidity. market
    Source sentence
    However, in cryptocurrency markets, characterized by 24/7 trading, extreme volatility, and rapid regime shifts, existing Time Series Generation (TSG) methods and benchmarks often fall short
  • Residuals between real returns and reconstructed returns follow mean‑reverting dynamics for the Statistical Arbitrage task. statistical
    Source sentence
    In this task, residuals between real 𝑹train and reconstructed returns ˆ𝑹train are assumed to follow mean-reverting dynamics.
  • Trading strategies are evaluated under two fee regimes: zero fee and a flat 0.03% fee. execution
    Source sentence
    Zero Trading Fee 0.03% Trading Fee

Data

  • Crypto-Centric Datasets · Binance exchange market-prices public
    January 2020 to December 2024 · 452 cryptocurrencies (spot pairs against USDT)
    Source sentence
    Hourly 24/7 OHLC data from 452 cryptocurrencies, curated and processed via a standardized pipeline for consistency and reliability.

Recognised sources: Binance

Costs

Unclear - sources disagree The platform classifier and the paper-text extraction give different answers, so no value is asserted.

Platform classifier: Partial

Paper text extraction: Modeled - 0.03% trading fee (flat fee) used in evaluation alongside a zero‑fee baseline

Source sentence (the authors' words)
Zero Trading Fee 0.03% Trading Fee

Finds 5

  • Diffusion‑TS achieves the highest predictive accuracy in the 2021 bull market but yields a negative CAGR and low Sharpe, indicating an accuracy‑alpha gap. performance
    Source sentence
    its negative CAGR and low Sharpe ratio highlight an accuracy–alpha gap
  • TimeVAE provides a balanced trade‑off, delivering solid forecasting accuracy together with robust returns. performance
    Source sentence
    TimeVAE strikes a compelling balance, delivering solid forecasting accuracy and robust returns
  • COSCI‑GAN produces promising CAGRs and Sharpe ratios above five under trend‑sensitive strategies. performance
    Source sentence
    COSCI-GAN thrives under trend‑sensitive strategies such as LOTQ and PW, producing promising CAGRs and Sharpe ratios above five
  • Models with smoother residual signals (e.g., TimeVAE, Diffusion‑TS) exhibit smaller rank erosion when a 0.03% fee is introduced, suggesting lower turnover. statistical
    Source sentence
    smoother models like TimeVAE and Diffusion-TS show smaller rank erosion
  • LS4 shows strong fee resilience, compounding almost monotonically in the Statistical Arbitrage task, especially during regime shifts. performance
    Source sentence
    LS4 compounds almost monotonically, highlighting its superior fee resilience

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 36 related papers). Coverage is partial, so this is neither replication nor refutation.

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.