Paper: arXiv 2603.23300

Authors: Mehmet Caner, Agostino Capponi, Nathan Sun, Jonathan Y. Tan

Abstract

We introduce a new agentic artificial intelligence (AI) platform for portfolio management. Our architecture consists of three layers. First, two large language model (LLM) agents are assigned specialized tasks: one agent screens for firms with desirable fundamentals, while a sentiment analysis agent screens for firms with desirable news. Second, these agents deliberate to generate and agree upon buy and sell signals from a large portfolio, substantially narrowing the pool of candidate assets. Finally, we apply a high-dimensional precision matrix estimation procedure to determine optimal portfolio weights. A defining theoretical feature of our framework is that the number of assets in the portfolio is itself a random variable, realized through the screening process. We introduce the concept of sensible screening and establish that, under mild screening errors, the squared Sharpe ratio of the screened portfolio consistently estimates its target. Empirically, our method achieves superior Sharpe ratios relative to an unscreened baseline portfolio and to conventional screening approaches, evaluated on S&P 500 data over the period 2020–2024.

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 6.0/10 (capped from 7.5: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper introduces a novel agentic AI framework combining LLM deliberation with rigorous high-dimensional portfolio theory. It demonstrates strong math by treating portfolio size as a random variable with consistency proofs, backed by realistic empirical validation on recent S&P 500 data.

Research Flowchart

  flowchart TD
    A[Goal: Enhance Portfolio Optimization<br/>via Agentic AI Screening] --> B[(Data Inputs:<br/>S&P 500 Equities & News 2020-2024)]
    B --> C[Layer 1: Dual LLM Agents<br/>Fundamental & Sentiment Screening]
    C --> D[Layer 2: Agent Deliberation<br/>Generate Agreed Buy/Sell Signals]
    D --> E[Layer 3: Precision Matrix Estimation<br/>Calculate Optimal Portfolio Weights]
    E --> F[Outcomes: Consistent Sharpe Ratio Estimation<br/>& Superior Performance vs. Baselines]

Cited References (151)

  1. Can ChatGPT Forecast Stock Price Movements? Return Predictibality and Large Language Models (2026)
  2. Information-Acquisition and Under-Diversification (2010)
  3. Assesing the Benefits of Optimized Agentic-AI Systems for Asset Pricing (2026)
  4. The memorization problem: Can we trust LLMs’ Economic Forecasts? (2026)
  5. Mimicking in Finance (2026)
  6. Large Language Models: An Applied Econometric Framework (2025)
  7. Sparse spanning portfolios and under-diversification with second-order stochastic dominance — → 2402.01951
  8. Financial machine learning (2023)
and 143 more
  1. Judgment under uncertainty: heuristics and bias (1974)
  2. Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment (1983)
  3. The other side of value: The gross profitability premium (2013)
  4. Value investing: The use of historical financial statement information to separate winners from losers (2000)
  5. Bad news travels slowly: size, analyst coverage, and the profitability of momentum strategies (2000)
  6. Financial ratios, discriminant analysis and the predictions of corporate bankruptcy (1968)
  7. Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency (1993)
  8. The Cross-Section of Expected Stock Returns (1992)
  9. AI’s predictable memory in financial analysis (2025)
  10. Portfolio Analysis in High Dimensions with Tracking Error and Weight Constraints (2026)
  11. AI Agents vs. Agentic AI: A Conceptual taxonomy, applications and challenges (2026)
  12. Deep Learning Based Residuals in Non-linear Factor Models: Precision Matrix Estimation of Returns with Low Signal-to-Noise Ratio (2025)
  13. A Practioner’s Guide to AI-ML in Portfolio Investing (2025)
  14. Expected returns and Large Language models (2023)
  15. Agentic Context Engineering:Evolving Contexts for Self Improving Language Models (2025)
  16. Alpha Agents: Large Language Model Based Multi-Agents for Equity Portfolio Constructions (2025)
  17. Large Language Models: A Deep Dive (2024)
  18. Deep Learning, Foundation and Concepts (2024)
  19. Large language Models in Finance: A Survey (2024)
  20. Attention is All you Need (2017)
  21. Can LLM-based Financial Investing Strategies Outperform the Market in Lomg-Run (2025)
  22. Factor augmented sparse throughput Deep Relu Neural networks for high dimensional regression (2024)
  23. Learning latent factors from diversified projections and its applications to over-estimated and weak factors (2022)
  24. Estimation risk and the implicit value of index tracking (2022)
  25. Lest we forget: Learn from out-of-sample forecast errors when optimizing portfolios (2022)
  26. Portfolio Optimization with Tracking Error Constraints (2003)
  27. A mean-variance analysis of tracking error (1992)
  28. Why constrain your mutual fund manager? (2004)
  29. Portfolio Optimization under Tracking Error and Weights Constraints (2011)
  30. Covariance Versus Precision Matrix Estimation for Efficient Asset Allocation (2016)
  31. Forecasting large realized covariance matrices: The benefits of factor models and shrinkage (2018)
  32. Incorporating Global Industrial Classification Standard into Portfolio Allocation: A Simple Factor-Based Large Covariance Matrix Estimator with High Frequency Data (2016)
  33. Estimation of large dimensional conditional factor models in finance (2020)
  34. Sharpe Ratio Analysis in High Dimensions: Residual Based Nodewise Regression in Factor Models (2023)
  35. A diagnostic criterion for approximate factor structure (2019)
  36. The current state of the arbitrage pricing theory (1992)
  37. Large covariance estimation through elliptical factor models (2018)
  38. Risks of large portfolios (2015)
  39. A generalized approach to portfolio optimization: Improving performance by constraining portfolio norms (2009)
  40. Arbitrage, factor structure, and mean-variance analysis on large asset markets (1983)
  41. Sparse and Stable M arkowitz Portfolios (2009)
  42. Skill, scale, and value creation in the mutual fund industry (2021+)
  43. High-dimensional minimum variance portfolio estimation under statistical factor models (2021)
  44. Bridging Factor and Sparse Models (2021)
  45. High dimensional inverse covariance matrix estimation via linear programming (2010)
  46. High-dimensional covariance matrix estimation in approximate factor models (2011)
  47. High-dimensional covariance matrix estimation using a factor model (2008)
  48. Time-varying risk premium in large cross-sectional equity data sets (2016)
  49. A B ernstein type inequality and moderate deviations for weakly dependent sequences (2011)
  50. Time Series Analysis (1994)
  51. Matrix Analysis (2013)
  52. Matrix Algebra (2005)
  53. Vast volatility matrix estimation using high frequency data for portfolio selection (2012)
  54. Robust performance hypothesis testing with the S harpe ratio (2008)
  55. Portfolio selection with parameter and model uncertainty: A multi-prior approach (2007)
  56. Risk reduction in large portfolios: Why imposing the wrong constraints helps (2003)
  57. Optimal portfolio choice with parameter uncertainty (2007)
  58. Mean-variance portfolio optimization when means and covariances are unknown (2011)
  59. Markowitz meets Talmud: A combination of sophisticated and naive diversification strategies (2011)
  60. Portfolio Selection (1952)
  61. Nonlinear shrinkage of the covariance matrix for portfolio selection: Markowitz meets Goldilocks (2017)
  62. Improved estimation of the covariance matrix of stock returns with an application to portfolio selection (2003)
  63. A well conditioned estimator for large dimensional covariance matrices (2004)
  64. Analytical nonlinear shrinkage of large-dimensional covariance matrices (2020)
  65. Approaching mean-variance efficiency for large portfolios (2019)
  66. A Constrained l1 minimization approach to sparse precision matrix estimation (2011)
  67. Confidence regions for entries of a large precision matrix (2018)
  68. A nodewise regression approach to estimating large portfolios (2021)
  69. Asymptotically honest confidence regions for high dimensional parameters by the desparsified conservative lasso (2018)
  70. Inference for high dimensional instrumental variables regression (2018)
  71. High Dimensional Linear GMM (2019)
  72. The large sample distribution of the maximum Sharpe ratio with and without short sales (2016)
  73. New light on portfolio allocation problem (2002)
  74. Probability inequalities (2010)
  75. Oracle inequalities for high dimensional vector autoregressions (2014)
  76. On the conditions used to prove oracle results for the Lasso (2009)
  77. Regression shrinkage and selection via the lasso (1996)
  78. The adaptive lasso and its oracle properties (2006)
  79. Simultaneous analysis of Lasso and Dantzig selector (2009)
  80. Sup-norm convergence rate and sign concentration property of Lasso and Dantzig estimators (2008)
  81. On asymptotically optimal confidence regions and tests for high-dimensional models (2014)
  82. On asymptotically optimal confidence regions and tests for high-dimensional models (2013) — link
  83. High-dimensional graphs and variable selection with the lasso (2006)
  84. High dimensional inverse covariance matrix estimation via linear programming (2010)
  85. Model selection and inference: Facts and fiction (2005)
  86. Honest confidence regions for nonparametric regression (1989)
  87. Confidence sets based on sparse estimators are necessarily large (2009)
  88. The nonexistence of certain statistical procedures in nonparametric problems (1956)
  89. Sparse models and methods for optimal instruments with an application to eminent domain (2012)
  90. Sparse models and methods for optimal instruments with an application to eminent domain (2010) — link
  91. Inference on Treatment Effects after Selection among High-Dimensional Controls (2011)
  92. Inference on Treatment Effects after Selection among High-Dimensional Controls (2014)
  93. Square root lasso: pivotal recovery of sparse signals via conic programming (2011)
  94. Estimation and testing under sparsity (2016)
  95. Statistics for High Dimensional Data (2011)
  96. Oracle inequalities for convex loss functions with non linear targets (2013)
  97. A selective overview of variable selection in high dimensional feature space (2010)
  98. High dimensional instrumental variable regression and confidence sets (2012)
  99. The adaptive and thresholded lasso for potentially misspecified models (2011)
  100. Variable selection via nonconcave penalized likelihood and its oracle properties (2001)
  101. Asymptotic properties of bridge estimators in sparse high-dimensional regression models (2008)
  102. Adaptive elastic net for generalized methods of moments (2014)
  103. Oracle efficient variable selection in random and fixed effects panel data models (2013)
  104. High dimensional sparse econometric models: An introduction (2011)
  105. Inference for high-dimensional sparse econometric models (2011) — link
  106. Econometric Theory (2000)
  107. Least squares after model selection in high-dimensional sparse models (2013)
  108. Pivotal estimation via square-root lasso in nonparametric regression (2014)
  109. Uniform post selection inference for LAD regression and other z-estimation problems (2013) — link
  110. Confidence intervals for low dimensional parameters in high dimensional linear models (2014)
  111. Hypothesis testing in high-dimensional regression under the gaussian random design model: Asymptotic theory (2013) — link
  112. Confidence intervals and hypothesis testing for high-dimensional regression (2014) — link
  113. POET: Principal Orthogonal Complement Thresholding (POET) Method (2016) — link
  114. nlshrink: Non-Linear Shrinkage Estimation of Population Eigenvalues and Covariance Matrices (2016) — link
  115. lars: Least Angle Regression, Lasso and Forward Stagewise (2013) — link
  116. Regularization paths for generalized linear models via coordinate descent (2010)
  117. Regularization parameter selections via generalized information criterion (2010)
  118. Large covariance estimation by thresholding principal orthogonal complements (2013)
  119. Nonlinear shrinkage estimation of large-dimensional covariance matrices (2012)
  120. Tuning parameter selection in high dimensional penalized likelihood (2013)
  121. Confidence regions for entries of a large precision matrix (2018)
  122. Optimal versus naive diversification: How inefficient is the 1/N portfolio strategy? (2007)
  123. Performance hypothesis testing with the Sharpe and Treynor measures (1981)
  124. Performance hypothesis testing with the Sharpe ratio (2003)
  125. High-Dimensional Portfolio Selection with Cardinality Constraints (2023)
  126. High-dimensional minimum variance portfolio estimation based on high-frequency data (2020)
  127. In-Sample and Out-of-Sample Sharpe Ratios of Multi-Factor Asset Pricing Models (2024)
  128. Vast Portfolio Selection With Gross-Exposure Constraints (2012)
  129. Finbert: Financial sentiment analysis with pre-trained language models (2019) — link
  130. Separating winners from losers among lowbook-to-market stocks using financial statement analysis (2005)
  131. Quality minus junk (2019)
  132. Determining the number of factors in approximate factor models (2002)
  133. Deep neural networks for estimation and inference (2021)
  134. Good debt or bad debt: Detecting semantic orientations in economic texts (2014)
  135. On persistence in mutual fund performance (1997)
  136. Empirical asset pricing via machine learning (2020)
  137. The characteristics that provide independent information about average US monthly stock returns (2017)
  138. Measuring massive multitask language understanding (2020) — link
  139. Artificial intelligence asset pricing models (2025)
  140. Missing values handling for machine learning portfolios (2024)
  141. Shrinking the cross-section (2020)
  142. Optimal versus naive diversification: How inefficient is the 1/N portfolio strategy? (2009)
  143. Markowitz versus the Talmudic portfolio diversification strategies (2009)

1 of 151 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.