Paper: arXiv 2410.18098

Authors: Guilherme Thomaz, Denis Maua

Abstract

Mutual funds aim to generate returns above market averages. While predicting their future portfolio allocations can bring economic advantages, the task remains challenging and largely unexplored. To fill that gap, this work frames mutual fund portfolio prediction as a Next Novel Basket Recommendation (NNBR) task, focusing on predicting novel items in a fund’s next portfolio. We create a comprehensive benchmark dataset using publicly available data and evaluate the performance of various recommender system models on the NNBR task. Our findings reveal that predicting novel items in mutual fund portfolios is inherently more challenging than predicting the entire portfolio or only repeated items. While state-of-the-art NBR models are outperformed by simple heuristics when considering both novel and repeated items together, autoencoder-based approaches demonstrate superior performance in predicting only new items. The insights gained from this study highlight the importance of considering domain-specific characteristics when applying recommender systems to mutual fund portfolio prediction. The performance gap between predicting the entire portfolio or repeated items and predicting novel items underscores the complexity of the NNBR task in this domain and the need for continued research to develop more robust and adaptable models for this critical financial application.

Complexity vs Empirical Score

  • Math Complexity: 5.5/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper applies recommender system models (like autoencoders and NBR methods) to financial data but the mathematics involves standard matrix factorization and probabilistic models rather than advanced derivatives. Empirical rigor is solid due to the creation of a benchmark dataset from SEC, Open FIGI, and Yahoo Finance, and the use of the RecBole library for reproducible evaluation.

Research Flowchart

  flowchart TD
  A["Research Goal: Predict Novel Items in Mutual Fund Portfolio"] --> B["Methodology: Formulate as Next Novel Basket Recommendation<br>Use Public Data to Build Comprehensive Benchmark"]
  B --> C["Computational Process: Apply & Evaluate Multiple Recommender System Models<br>Comparing SOTA vs. Simple Heuristics vs. Autoencoders"]
  C --> D{"Data Inputs: Mutual Fund Historical Portfolios<br>Equities, Holdings, Time Series"}
  C --> E["Key Findings & Outcomes"]
  E --> F["Prediction of novel items is more difficult<br>than repeated items or full portfolio"]
  E --> G["SOTA NBR models are outperformed by<br>simple heuristics for novel items"]
  E --> H["Autoencoder-based approaches show<br>superior performance on novel items"]
  E --> I["Domain-specific characteristics are critical<br>for financial recommender systems"]

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 applying Next Novel Basket Recommendation (NNBR) techniques to mutual fund portfolio prediction and creates a benchmark dataset for this task.

Source sentence
We introduce a novel application of NNBR techniques to the domain of mutual fund portfolio prediction, addressing the unique challenges and opportunities presented by this context.

Assumes 2

  • Only assets with a valid CUSIP, listed on US stock markets, labeled as Common Stocks, with a Long pay‑off profile and at least $10,000 allocation are included. data
    Source sentence
    Only assets with a valid CUSIP identifier, listed on US stock markets and labeled as "Common Stocks" by Open FIGI are retained. Furthermore, we keep assets with a "Long" pay-off profile and an allocation of at least 10,000 USD.
  • The median portfolio size is 69 items, which may require undersampling for some NBR models. data
    Source sentence
    This is a particular problem in the mutual fund context, where the median portfolio size is 69 items.

Data

  • benchmark dataset · SEC, Open FIGI, Yahoo Finance benchmark public
    2020-2021 · US mutual funds with partial allocation in US‑listed stocks
    Source sentence
    We create a comprehensive benchmark dataset using publicly available data from the SEC, Open FIGI, and Yahoo Finance, focusing on mutual funds with at least a partial allocation in U.S.-listed stocks from 2020 to 2021.

Recognised sources: Yahoo Finance

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • Predicting novel items in mutual fund portfolios is inherently more challenging than predicting the entire portfolio or only repeated items. predictive-accuracy
    Source sentence
    Our findings reveal that predicting novel items in mutual fund portfolios is inherently more challenging than predicting the entire portfolio or only repeated items.
  • State‑of‑the‑art NBR models are outperformed by simple heuristics when both novel and repeated items are considered together. performance
    Source sentence
    While state-of-the-art NBR models are outperformed by simple heuristics when considering both novel and repeated items together...
  • Autoencoder‑based approaches demonstrate superior performance in predicting only new items. predictive-accuracy
    Source sentence
    ...autoencoder‑based approaches demonstrate superior performance in predicting only new items.
  • The best NNBR model (EASE) achieves an NDCG of only 0.21, far below the allocation heuristic’s NDCG of 0.956 for NBR tasks. performance
    Source sentence
    It is important to note that the NBR and NBRR NDCG metrics for the current allocation heuristic are 0.956 and 0.961, respectively, while the NDCG for the best NNBR model, EASE, is only 0.21.
  • Simple allocation heuristic performs strongly on NBR and NBRR tasks, suggesting simple, interpretable models can be highly effective. performance
    Source sentence
    The strong performance of the current allocation heuristic in NBR and NBRR tasks suggests that simple, interpretable models can be highly effective for predicting the core holdings of mutual funds.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 26 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.

Cited References (76)

  1. Investment Basics Explained With Types to Invest in (2023) — link
  2. Stocks: What They Are, Main Types, How They Differ From Bonds (2022) — link
  3. US Mutual Funds Industry: Market Size (2023)
  4. 2023 Investment Company Fact Book (2023)
  5. What We Do (2023)
  6. Form N-PORT (2023)
  7. EDGAR API Documentation (2023)
  8. Systemic risk from investment similarities (2019) — link
and 68 more
  1. Old dog, new tricks: a modelling view of simple moving averages (2018) — link
  2. What Is an Investment Fund? Types of Funds and History (2023)
  3. How Often Do Mutual Funds Report Their Holdings? (2023)
  4. Front-Running Definition, Example, and Legality (2023)
  5. BPR : Bayesian Personalized Ranking from Implicit Feedback (2009)
  6. On Sampling Strategies for Neural Network-based Collaborative Filtering (2017)
  7. Improving Pairwise Learning for Item Recommendation from Implicit Feedback (2014)
  8. Optimizing Top-n Collaborative Filtering via Dynamic Negative Item Sampling (2013)
  9. IRGAN : A Minimax Game for Unifying Generative and Discriminative Information Retrieval Models (2017)
  10. Adversarial Sampling and Training for Semi-Supervised Information Retrieval (2019)
  11. Reinforced Negative Sampling for Recommendation with Exposure Data (2019)
  12. Simplify and Robustify Negative Sampling for Implicit Collaborative Filtering (2020)
  13. Overview: Negative Sampling on Recommendation Systems
  14. Node2vec: Scalable Feature Learning for Networks (2016) — link
  15. Fund2Vec: Mutual Funds Similarity Using Graph Learning (2022) — link
  16. Hierarchical structure in financial markets (1999) — link
  17. Learning Embedded Representation of the Stock Correlation Matrix using Graph Machine Learning (2022) — link
  18. Planar Maximally Filtered Graph (PMFG) (2023)
  19. Boost then Convolve: Gradient Boosting Meets Graph Neural Networks (2021) — link
  20. LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation (2020) — link
  21. Inductive Representation Learning on Large Graphs (2018) — link
  22. Graph Attention Networks (2018) — link
  23. BPR: Bayesian Personalized Ranking from Implicit Feedback (2009)
  24. Ranking with Ordered Weighted Pairwise Classification (2009) — link
  25. Why Are Deep Learning Models Not Consistently Winning Recommender Systems Competitions Yet? A Position Paper (2020) — link
  26. Sets2Sets: Learning from Sequential Sets with Neural Networks (2019) — link
  27. Recency Aware Collaborative Filtering for Next Basket Recommendation (2020) — link
  28. Market basket analysis with networks (2011) — link
  29. A Next Basket Recommendation Reality Check (2023) — link
  30. MUTUAL FUND RECOMMENDATION SYSTEM WITH PERSONALIZED EXPLANATIONS (2022) — link
  31. Explainable Mutual Fund Recommendation System Developed Based on Knowledge Graph Embeddings (2022) — link
  32. Next-Generation Personalized Investment Recommendations (2022) — link
  33. Fund2Vec: Mutual Funds Similarity using Graph Learning (2021) — link
  34. LightGBM: A Highly Efficient Gradient Boosting Decision Tree (2017)
  35. The Costs and Benefits of Performance Fees in Mutual Funds (2022) — link
  36. Introduction to recommender systems handbook (2015)
  37. Toward the next generation of recommender systems: a survey of the state-of-the-art and possible extensions (2005)
  38. Recommender systems survey (2013)
  39. A comprehensive survey of neighborhood-based recommendation methods (2011)
  40. Deep neural networks for YouTube recommendations (2016)
  41. Factorization Meets the Neighborhood: A Multifaceted Collaborative Filtering Model (2008)
  42. Collaborative Filtering for Implicit Feedback Datasets (2008)
  43. Neural Recommender Systems: A Survey and Taxonomy (2019) — link
  44. Content-based Recommender Systems: State of the Art and Trends (2011) — link
  45. Fab: Content-Based, Collaborative Recommendation (1997) — link
  46. A Gentle Introduction to Graph Neural Networks (2021) — link
  47. Semi-Supervised Classification with Graph Convolutional Networks (2017) — link
  48. Graph Convolutional Neural Networks for Web-Scale Recommender Systems (2018) — link
  49. Ensemble learning for the early prediction of neonatal jaundice with genetic features (2021) — link
  50. Predicting Temporal Sets with Deep Neural Networks (2020) — link
  51. Graph Neural Networks for Recommender Systems: Challenges, Methods, and Directions (2021) — link
  52. Multi-Task Learning as Multi-Objective Optimization (2019) — link
  53. Embarrassingly Shallow Autoencoders for Sparse Data (2019) — link
  54. RecVAE: A New Variational Autoencoder for Top-N Recommendations with Implicit Feedback (2020) — link
  55. LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation (2020) — link
  56. Graph Convolutional Matrix Completion (2017) — link
  57. Improved Recurrent Neural Networks for Session-based Recommendations (2016) — link
  58. BERT4Rec: Sequential Recommendation with Bidirectional Encoder Representations from Transformer (2019) — link
  59. Modeling Personalized Item Frequency Information for Next-basket Recommendation (2020) — link
  60. Neural Collaborative Filtering (2017) — link
  61. Item-based top-N recommendation algorithms (2004) — link
  62. RecBole: Towards a Unified, Comprehensive and Efficient Framework for Recommendation Algorithms (2021)
  63. RecBole 2.0: Towards a More Up-to-Date Recommendation Library (2022)
  64. Recent Advances in RecBole: Extensions with more Practical Considerations (2022) — link
  65. Masked and Swapped Sequence Modeling for Next Novel Basket Recommendation in Grocery Shopping (2023) — link
  66. Personalized finance advisory through case-based recommender systems and diversification strategies (2015) — link
  67. A stock market portfolio recommender system based on association rule mining (2013) — link
  68. Torch Metrics (2024) — link

0 of 76 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.