Paper: arXiv 2308.06260

Authors: Oleksandr Romanko, Akhilesh Narayan, Roy H. Kwon

Abstract

In this paper, we explore potential uses of generative AI models, such as ChatGPT, for investment portfolio selection. Trusting investment advice from Generative Pre-Trained Transformer (GPT) models is a challenge due to model “hallucinations”, necessitating careful verification and validation of the output. Therefore, we take an alternative approach. We use ChatGPT to obtain a universe of stocks from S&P500 market index that are potentially attractive for investing. Subsequently, we compared various portfolio optimization strategies that utilized this AI-generated trading universe, evaluating those against quantitative portfolio optimization models as well as comparing to some of the popular investment funds. Our findings indicate that ChatGPT is effective in stock selection but may not perform as well in assigning optimal weights to stocks within the portfolio. But when stocks selection by ChatGPT is combined with established portfolio optimization models, we achieve even better results. By blending strengths of AI-generated stock selection with advanced quantitative optimization techniques, we observed the potential for more robust and favorable investment outcomes, suggesting a hybrid approach for more effective and reliable investment decision-making in the future.

Complexity vs Empirical Score

  • Math Complexity: 5.0/10
  • Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.0)

Why this score: The paper employs advanced mathematical finance concepts like Markowitz mean-variance optimization, efficient frontiers, and Sharpe ratio calculations, which establishes moderate-to-high math complexity. It demonstrates strong empirical rigor through comprehensive backtesting on multiple time periods, comparison with benchmarks and fund data, and detailed performance metrics evaluation.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>AI-Assisted Portfolio Selection"] --> B["Data Input:<br>S&P 500 Market Index"]
  B --> C["AI Generation:<br>ChatGPT Stock Selection"]
  C --> D{"Computational Process:<br>Portfolio Optimization"}
  D --> E["Method A:<br>Optimize AI Selection"]
  D --> F["Method B:<br>Quantitative Benchmark"]
  E & F --> G["Comparison:<br>Against Market Funds"]
  G --> H["Key Findings:<br>Hybrid Approach Optimal<br>AI Selection + Quant Weighting = Best Results"]

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 using ChatGPT to generate a stock universe from the S&P 500 and then applying traditional portfolio optimization to those stocks, creating a hybrid approach for investment selection.

Source sentence
... when stocks selection by ChatGPT is combined with established portfolio optimization models, we achieve even better results.

Assumes 4

  • GPT‑4 was trained on data prior to September 2021, so in‑sample parameters are computed using historical data before that date. data
    Source sentence
    At the time of writing, the GPT‑4 model was trained on data prior to September 2021. Consequently, we used historical data up prior to September 2021 as our in‑sample data for portfolio selection models
  • The stock universe is limited to constituents of the S&P 500, specifically the 485 stocks for which complete price histories are available. market
    Source sentence
    From all stocks included in the S&P500 market index, we obtained complete historical timeseries (both in‑sample and out‑of‑sample) for 485 stocks using Yahoo Finance.
  • ChatGPT’s recommendations reflect the frequency and positivity of mentions in its training text, implying a sentiment‑based selection bias. model
    Source sentence
    The underlying hypothesis is that the frequency and positivity of mentions related to a given stock in its training data
  • Portfolios are re‑balanced only at the start of each out‑of‑sample testing period, with no intra‑period trading. execution
    Source sentence
    Out‑of‑sample data used for testing and comparing portfolios span three distinct periods:

Data

  • Historical price series for S&P 500 constituents · Yahoo Finance market-prices public
    485 S&P 500 stocks
    Source sentence
    From all stocks included in the S&P500 market index, we obtained complete historical timeseries (both in‑sample and out‑of‑sample) for 485 stocks using Yahoo Finance.
  • GPT‑4 model via API · OpenAI text-news commercial
    Source sentence
    In this study, we utilized the GPT‑4 model through its API.

Recognised sources: Yahoo Finance

Costs

Not modeled A trading or execution result is reported without costs.

Finds 3

  • Portfolios built from ChatGPT‑selected stock universes outperform those derived from cardinality‑constrained optimizations, showing ChatGPT excels at stock pre‑selection. performance
    Source sentence
    the portfolios constructed from the ChatGPT trading universes outperform those derived from cardinality‑constrained optimizations, implying that ChatGPT is good at stock pre‑selection.
  • Combining ChatGPT‑selected stocks with traditional portfolio optimization yields better risk‑reward profiles than using either approach alone. performance
    Source sentence
    when stocks selection by ChatGPT is combined with established portfolio optimization models, we achieve even better results.
  • ChatGPT alone is less effective at assigning optimal weights to stocks within a portfolio. negative
    Source sentence
    ChatGPT is effective in stock selection but may not perform as well in assigning optimal weights to stocks within the portfolio.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 15 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 5 papers in the archive (5 by independent authors)

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.