Paper: arXiv 2510.10878

Authors: Zheng Cao, Xingran Shao, Yuheng Yan, Helyette Geman

Abstract

We propose a novel model, the Hyped Log-Periodic Power Law Model (HLPPL), to the problem of quantifying and detecting financial bubbles, an ever-fascinating one for academics and practitioners alike. Bubble labels are generated using a Log-Periodic Power Law (LPPL) model, sentiment scores, and a hype index we introduced in previous research on NLP forecasting of stock return volatility. Using these tools, a dual-stream transformer model is trained with market data and machine learning methods, resulting in a time series of confidence scores as a Bubble Score. A distinctive feature of our framework is that it captures phases of extreme overpricing and underpricing within a unified structure. We achieve an average yield of 34.13 percentage annualized return when backtesting U.S. equities during the period 2018 to 2024, while the approach exhibits a remarkable generalization ability across industry sectors. Its conservative bias in predicting bubble periods minimizes false positives, a feature which is especially beneficial for market signaling and decision-making. Overall, this approach utilizes both theoretical and empirical advances for real-time positive and negative bubble identification and measurement with HLPPL signals.

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematical models like the Log-Periodic Power Law (LPPL) and Ornstein–Uhlenbeck processes, demonstrating substantial analytical depth. It also presents detailed backtesting results (e.g., 34.13% annualized return, 2018–2024 period) and discusses generalization across sectors, indicating strong empirical and implementation focus.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Detect & Quantify Financial Bubbles"] --> B["Data & Inputs:<br>Market Data + Sentiment/Hype Index"]
  B --> C["Methodology:<br>HLPPL Model for Bubble Labels"]
  C --> D["Model Processing:<br>Dual-Stream Transformer"]
  D --> E["Outcome:<br>Time Series of Bubble Confidence Scores"]
  E --> F{"Final Findings:"}
  F --> G["34.13% Annualized Return<br>(2018-2024 Backtest)"]
  F --> H["High Generalization<br>Across Industry Sectors"]

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 the Hyped Log-Periodic Power Law Model (HLPPL), a unified framework that detects both positive and negative financial bubbles by combining LPPL dynamics with sentiment and hype indicators via a dual‑stream transformer.

Source sentence
We propose a novel model, the Hyped Log-Periodic Power Law Model (HLPPL), to the problem of quantifying and detecting financial bubbles, an ever-fascinating one for academics and practitioners alike.

Assumes 6

  • Bubble price dynamics follow a power‑law acceleration toward a finite‑time singularity. model
    Source sentence
    Suppose that the expected growth of prices follows a nonlinear feedback mechanism such that d ln p(t) / dt ∝(tc −t)^{m−1}, 0 < m < 1
  • Log‑periodic oscillations arise from investor herding and increase in frequency as the critical time approaches. model
    Source sentence
    By assuming that investor herding generates log-periodic fluctuations, the growth term is decorated with an oscillatory modulation leading to (1).
  • Trading signals are generated using fixed entry/exit thresholds on the normalized Bubble Score. execution
    Source sentence
    5. Risk Management: The strategy enforces a 15% stop-loss, caps maximum position size at 50% of capital, and incorporates transaction costs of 0.1%.
  • The backtest uses daily price and return data from CRSP and sentiment signals from WSJ, aligned at a daily frequency. data
    Source sentence
    The stock-level daily price and return data are sourced from the CRSP database via WRDS, while the news sentiment signals are derived from the Wall Street Journal (WSJ). Both datasets span the period from January 1, 2018 to December 31, 2024, and are aligned at a daily frequency.
Show 2 more
  • Only firms with at least 100 daily observations are included to ensure reliable LPPL fitting. data
    Source sentence
    Each stock requires a minimum of 100 daily observations to ensure statistical reliability of the LPPL fitting process.
  • All market‑level features (except Hype Index and FinBERT sentiment scores) are sourced from WRDS. data
    Source sentence
    All market-level features (except Hype Index and FinBERT sentiment scores) are sourced from WRDS (Wharton Research Data Services).

Data

Backtest sample: 2018 – 2024 · U.S. equities · equities · daily

  • CRSP daily stock file · CRSP via WRDS market-prices commercial
    2018-01-01 to 2024-12-31 · real estate sector stocks (SIC 6500‑6552, 6798)
    Source sentence
    The stock-level daily price and return data are sourced from the CRSP database via WRDS, while the news sentiment signals are derived from the Wall Street Journal (WSJ). Both datasets span the period from January 1, 2018 to December 31, 2024, and are aligned at a daily frequency.
  • Hype Index alternative unknown
    Source sentence
    Sector exposure indicator from Hype Index
  • FinBERT sentiment scores text-news unknown
    Source sentence
    One-hot: negative sentiment (FinBERT) Aggregated by date
  • Macro and valuation features · WRDS macro commercial
    Source sentence
    Normalized CBOE Volatility Index; market risk proxy

Recognised sources: WRDS CRSP Cboe

Costs

Modeled Costs are part of the reported headline results.

A flat transaction cost of 0.1% per trade is applied in the backtest.

Source sentence
The strategy enforces a 15% stop-loss, caps maximum position size at 50% of capital, and incorporates transaction costs of 0.1%.

Finds 4

Reported performance (Bubble Score trading strategy): Sharpe 2.49 (benchmark -0.54) · annual return 34.1% · max drawdown 11.3%

  • The machine‑learning‑enhanced HLPPL trading strategy yields an average annualized return of 34.13% with a Sharpe ratio of 1.19. performance
    Source sentence
    When implemented as a trading strategy, the machine-learning–enhanced approach generates an average annualized return of 34.13% with an average Sharpe ratio of 1.19, significantly outperforming the traditional Bubble Score rules strategy, which delivers 16.64% annualized return and a Sharpe ratio of 0.72.
  • The Transformer model achieves a mean Pearson correlation of 0.625 between predicted and actual Bubble Scores, with MSE = 0.087 and RMSE = 0.295. predictive-accuracy
    Source sentence
    the empirical analysis shows that the prediction module achieves a mean Pearson correlation of 0.625 with the Bubble Score (MSE = 0.087; RMSE = 0.295)
  • Across 32 real‑estate stocks, 30 (93.75%) produced positive returns, with an average annualized return of 16.64% and average Sharpe ratio of 0.72. performance
    Source sentence
    Total Stocks Analyzed 32 … Stocks with Positive Returns 30 (93.75%)
  • The model’s predictive performance remains stable across horizons, with correlation ranging from 0.628 on Day 1 to 0.609 on Day 5. predictive-accuracy
    Source sentence
    The observed correlations are relatively stable, ranging from 0.628 on Day 1 to 0.609 on Day 5.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 53 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.