Paper: arXiv 2404.00012
Authors: Baptiste Lefort, Eric Benhamou, Jean-Jacques Ohana, David Saltiel, Beatrice Guez, Thomas Jacquot
Abstract
This paper introduces a new risk-on risk-off strategy for the stock market, which combines a financial stress indicator with a sentiment analysis done by ChatGPT reading and interpreting Bloomberg daily market summaries. Forecasts of market stress derived from volatility and credit spreads are enhanced when combined with the financial news sentiment derived from GPT-4. As a result, the strategy shows improved performance, evidenced by higher Sharpe ratio and reduced maximum drawdowns. The improved performance is consistent across the NASDAQ, the S&P 500 and the six major equity markets, indicating that the method generalises across equities markets.
Complexity vs Empirical Score
- Math Complexity: 2.5/10
- Empirical Rigor: 6.0/10 (capped from 7.5: no out-of-sample test; see penalty flags)
- Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 7.5)
Why this score: The paper focuses on implementing and testing a practical trading strategy using existing financial indicators (stress index, volatility, credit spreads) and an NLP tool (ChatGPT for sentiment analysis), with extensive backtesting across multiple markets and reported performance metrics (Sharpe ratio, drawdowns), indicating low theoretical math but high empirical implementation.
Research Flowchart
flowchart TD A["Research Goal:<br>Enhance risk-on risk-off equity strategy"] --> B["Data Collection"] B --> C["Financial Stress Indicator<br>Volatility + Credit Spreads"] B --> D["GPT-4 Sentiment Analysis<br>Bloomberg Market Summaries"] C --> E["Computational Model:<br>Forecast Market Stress"] D --> E E --> F["Strategy Execution<br>Risk-On vs Risk-Off Signals"] F --> G["Key Outcomes<br>Improved Sharpe Ratio & Reduced Max Drawdown"] G --> H["Generalization<br>Validated on NASDAQ, S&P 500, 6 Major Markets"]
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 new risk‑on risk‑off equity strategy that combines a financial stress indicator with ChatGPT‑derived news sentiment from Bloomberg daily market summaries.
Source sentence
This paper introduces a new risk-on risk-off strategy for the stock market, which combines a financial stress indicator with a sentiment analysis done by ChatGPT reading and interpreting Bloomberg daily market summaries.
Assumes 4
- News used are the latest available before the European market opening (end‑of‑day Asian market wraps). data
Source sentence
In order to have a consistent data set, we rely on the latest available news before the European market opening, that is the end of day Asian market wraps.
- The stress index output lies in [0,1] via a normal CDF transformation. model
Source sentence
Because the stress index final result is a number between 0 and 1 thanks to the cumulative distribution function of the normal distribution, we directly get a stress index signal.
- News sentiment is scored on a simple -1, 0, +1 scale. data
Source sentence
It is a simple scale: -1 points for a negative sentiment, 0 for indecisive, and +1 for the the days th…
- Linear transaction costs of 2 basis points per trade are applied to all strategies. execution
Source sentence
To have very realistic simulations, we incorporate linear transaction costs, denoted by b equals to 2 basis points for all our strategy (b = 0.0002).
Data
- Bloomberg Daily Market Wraps text-news commercial
Source sentence
We use professional news from Bloomberg. Bloomberg provides daily market wraps, that are daily summaries of the most important news. They are available since 2010 with growing frequency after 2011
Costs
Modeled
Linear transaction cost of 2 basis points per trade (b = 0.0002) applied to all strategies.
Source sentence
To have very realistic simulations, we incorporate linear transaction costs, denoted by b equals to 2 basis points for all our strategy (b = 0.0002).
Finds 4
- The hybrid strategy yields higher Sharpe ratios and lower maximum drawdowns than benchmarks across NASDAQ, S&P 500 and six major equity markets. performance
Source sentence
As a result, the strategy shows improved performance, evidenced by higher Sharpe ratio and reduced maximum drawdowns.
- Dynamic SI+News strategy achieves the highest Sharpe (0.89) and lowest max drawdown among tested strategies for the NASDAQ. performance
Source sentence
Dynamic SI+News 0.89 0.62 9.2% 13% 8.6
- Across all major equity markets, Dynamic SI+News attains the top Sharpe (0.85) and best Calmar ratio, outperforming SI‑only and VIX strategies. performance
Source sentence
Dynamic SI+News 0.85 0.44 6.8% 13% 8.5
- The SI‑only strategy was selected 71% of the time over 2011‑2024, indicating it often outperforms the combined SI+News approach. performance
Source sentence
the SI-only strategy was selected 71% of the tim…
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 20 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
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.
Cited References (53)
- Giving Content to Investor Sentiment: The Role of Media in the Stock Market (2007) — link
- Textual analysis of stock market prediction using breaking financial news: The AZFin text system (2009)
- BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (2018) — link
- FinBERT: Financial Sentiment Analysis with Pre-trained Language Models (2019) — link
- GPT-4 Technical Report (2023) — link
- A review of ChatGPT AI’s impact on several business sectors (2023)
- S enti W ord N et 3.0: An Enhanced Lexical Resource for Sentiment Analysis and Opinion Mining (2010) — link
- FinEntity: Entity-level Sentiment Classification for Financial Texts (2023) — link
and 45 more
- Practical Text Classification With Large Pre-Trained Language Models (2018) — link
- When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks (2011)
- BloombergGPT: A Large Language Model for Finance (2023) — link
- The Impact of Macroeconomic News Sentiment on Interest Rates (2022) — link
- Trading strategies to exploit blog and news sentiment (2010)
- DeepVaR: a framework for portfolio risk assessment leveraging probabilistic deep neural networks (2023)
- Stock Market Prediction Based on Financial News Text Mining and Investor Sentiment Recognition (2022) — link
- Googling investor’s sentiment, financial stress and dynamics of European market indexes: a Markov chain analysis (2020) — link
- Deep learning for sentiment analysis: A survey (2018)
- The response of different investor types to macroeconomic news (2019)
- The evolution of Fintech: A new post-crisis paradigm (2015)
- Can ChatGPT Decipher Fedspeak? (2023) — link
- How to Learn and Teach Economics with Large Language Models, Including GPT (2023) — link
- Language Models and Cognitive Automation for Economic Research (2023) — link
- Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models — → 2304.07619
- Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence (2023) — link
- Ceo behavior and firm performance (2020) — link
- Large language models can rate news outlet credibility (2023) — link
- Sentiment Analysis Applied to News from the Brazilian Stock Market (2022) — link
- A Deep Learning Approach with Extensive Sentiment Analysis for Quantitative Investment (2023) — link
- Predicting Price Trends Using Sentiment Analysis: A Study of StepN’s SocialFi and GameFi Cryptocurrencies (2023) — link
- Investment Strategy via Lead Lag Effect using Economic Causal Chain and SSESTM Model (2022) — link
- Investment strategy via analyst report text mining (2023) — link
- Risk-On / Risk-Off: Financial market response to investor fear (2015)
- Time-varying relationship of news sentiment, implied volatility and stock returns (2016) — link
- Application of Machine Learning With News Sentiment in Stock Trading Strategies (2023) — link
- Equilibrium Data Mining and Data Abundance (2020) — link
- Textual sentiment in finance: A survey of methods and models (2014)
- Negative news and investor trust: The role of $ Firm and # CEO Twitter use (2018)
- Can ChatGPT Compute Trustworthy Sentiment Scores from Bloomberg Market Wraps — → 2401.05447
- Sentiment Analysis in Financial News: A Review (2016)
- Social Media and News Sentiment Analysis for Advanced Investment Strategies (2023)
- Sentiment Analysis in Financial Texts (2023)
- A Deep Learning Approach with Extensive Sentiment Analysis for Quantitative Investment (2023)
- Sentiment Correlation in Financial News Networks and Associated Market Movements (2023)
- A new financial stress indicator: properties and conditional asset price behavior (2014)
- Natural language based financial forecasting: a survey (2018)
- Mutual Fund Performance (1966)
- Adjusting for Risk in Portfolio Performance Measurement (1975)
- Asset allocation: Management Style and Performance Measurement (1992)
- The Sharpe ratio (1994)
- Calmar ratio: A smoother tool (1991)
- Measuring economic policy uncertainty (2016)
- In search of attention (2011)
- All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors (2008)
2 of 53 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.