Paper: arXiv 2401.03737
Authors: Georgios Fatouros, Konstantinos Metaxas, John Soldatos, Dimosthenis Kyriazis
Abstract
This paper introduces MarketSenseAI, an innovative framework leveraging GPT-4’s advanced reasoning for selecting stocks in financial markets. By integrating Chain of Thought and In-Context Learning, MarketSenseAI analyzes diverse data sources, including market trends, news, fundamentals, and macroeconomic factors, to emulate expert investment decision-making. The development, implementation, and validation of the framework are elaborately discussed, underscoring its capability to generate actionable and interpretable investment signals. A notable feature of this work is employing GPT-4 both as a predictive mechanism and signal evaluator, revealing the significant impact of the AI-generated explanations on signal accuracy, reliability and acceptance. Through empirical testing on the competitive S&P 100 stocks over a 15-month period, MarketSenseAI demonstrated exceptional performance, delivering excess alpha of 10% to 30% and achieving a cumulative return of up to 72% over the period, while maintaining a risk profile comparable to the broader market. Our findings highlight the transformative potential of Large Language Models in financial decision-making, marking a significant leap in integrating generative AI into financial analytics and investment strategies.
Complexity vs Empirical Score
- Math Complexity: 2.5/10
- Empirical Rigor: 7.5/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper uses relatively straightforward mathematical concepts for an LLM-based framework, but provides detailed backtesting on a specific dataset (S&P 100 over 15 months) with reported performance metrics (alpha, cumulative returns, risk profile).
Research Flowchart
flowchart TD A["<b>Research Goal</b><br>Can LLMs Beat Wall Street?<br>AI in Stock Selection"] --> B["<b>Framework: MarketSenseAI</b><br>GPT-4 + Chain of Thought<br>+ In-Context Learning"] B --> C["<b>Data Integration</b><br>Market Trends, News,<br>Fundamentals, Macro Factors"] C --> D["<b>AI Processing</b><br>Expert-like Reasoning<br>Signal Generation & Evaluation"] D --> E["<b>Validation</b><br>S&P 100 Stocks<br>15-Month Backtest"] E --> F["<b>Key Findings</b><br>72% Cumulative Return<br>10-30% Excess Alpha<br>Controlled Risk"] style A fill:#e1f5fe style F fill:#e8f5e8
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 MarketSenseAI, an innovative GPT‑4‑based framework for stock selection that integrates chain‑of‑thought and in‑context learning to generate actionable, explainable investment signals.
Source sentence
This paper introduces MarketSenseAI, an innovative framework leveraging GPT-4’s advanced reasoning for selecting stocks in financial markets.
Assumes 4
- Transaction costs are assumed to be 5 basis points per trade. execution
Source sentence
Values in parenthesis represent the total returns after transaction costs (5bps/trade).
- The investment universe is limited to the S&P 100 constituents. market
Source sentence
MarketSenseAI’s performance leverages stocks listed in the S&P 100 index, comprising the 100 largest and most established companies in U.S. equity markets.
- Equal‑weighted indices are used to counterbalance disparate stock performances. statistical
Source sentence
Our analysis places special emphasis on equal-weighted indices, which helps counterbalance this disparity and illustrates MarketSenseAI’s potential.
- GPT‑4 explanations improve signal accuracy, reliability and acceptance. model
Source sentence
A notable feature of this work is employing GPT-4 both as a predictive mechanism and signal evaluator, revealing the significant impact of the AI-generated explanations on signal accuracy, reliability and acceptance.
Data
S&P 100 stocks · equities
- News articles text-news unknown
Source sentence
A total of 163,483 articles published from December 1, 2022, to February 29, 2024, averaging 4.57 articles per day per stock
- Fundamentals fundamentals unknown
Source sentence
Financial data were gathered from 612 quarterly reports of the S&P 100 stocks, starting from the second quarter of 2022. This data set produced 608 unique fundamentals summaries, averaging about 6 per stock.
- Stock price data market-prices unknown
Source sentence
Historical daily stock prices (adjusted close) from January 1, 2022, to February 29, 2024, were analyzed to compute stock price dynamics.
- Descriptions other unknown
Source sentence
Concise descriptions of each stock and its sector, employed by Algorithm 1 to identify similar stocks.
Costs
Modeled
5 basis points per trade are subtracted from reported returns.
Source sentence
Values in parenthesis represent the total returns after transaction costs (5bps/trade).
Finds 4
total return 72.0%
- The vanilla MarketSenseAI equally‑weighted strategy achieved a total return of 35.48% (32.94% after transaction costs) with a Sharpe ratio of 2.49 and a Sortino ratio of 3.87, outperforming the equally‑weighted S&P 100. performance
Source sentence
The strategy following the complete set of MarketSenseAI’s signals (MS) equally weighted, yields an impressive total return of 35.48% (32.94% after transaction costs) with a Sharpe and Sortino ratio of 2.49 and 3.87, respectively.
- Capitalization‑weighted MarketSenseAI portfolios delivered the highest risk‑adjusted scores, with MS‑L‑Cap achieving a total return of 66.22% (65.25% after costs) and a Sharpe ratio of 2.90. performance
Source sentence
Values in parenthesis represent the total returns after transaction costs (5bps/trade).
- GPT‑4‑ranked top‑stock strategies (MS‑Top5‑GPT and MS‑Top10‑GPT) produced total returns of 50.96% (49.67% after costs) and 49.09% (48.07% after costs), respectively, markedly higher than naive trend‑following. performance
Source sentence
The MS-Top5-GPT and MS-Top10-GPT strategies, which select the top 5 and top 10 GPT-4 ranked stocks respectively, achieved remarkable total returns of 50.96% (49.67% after costs) and 49.09% (48.07% after c…
- Across the 15‑month test on S&P 100 stocks, MarketSenseAI generated excess alpha between 10% and 30% and cumulative returns up to 72%, comparable risk profile to the broader market. performance
Source sentence
Through empirical testing on the competitive S&P 100 stocks over a 15-month period, MarketSenseAI demonstrated exceptional performance, delivering excess alpha of 10% to 30% and achieving a cumulative return of up to 72% over the period, while maintaining a risk profile comparable to the broader market.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 17 related papers). Coverage is partial, so this is neither replication nor refutation.
12 papers in the archive : Can LLM-based Financial Investing Strategies Outperform the … ; Large Language Model Agent in Financial Trading: A Survey ; Bridging Language Models and Financial Analysis
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.