Paper: arXiv 2504.19623

Authors: Giovanni Ballarin, Jacopo Capra, Petros Dellaportas

Abstract

Stock return prediction is a problem that has received much attention in the finance literature. In recent years, sophisticated machine learning methods have been shown to perform significantly better than ‘‘classical’’ prediction techniques. One downside of these approaches is that they are often very expensive to implement, for both training and inference, because of their high complexity. We propose a return prediction framework for intraday returns at multiple horizons based on Echo State Network (ESN) models, wherein a large portion of parameters are drawn at random and never trained. We show that this approach enjoys the benefits of recurrent neural network expressivity, inherently efficient implementation, and strong forecasting performance.

Complexity vs Empirical Score

  • Math Complexity: 4.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper employs relatively accessible machine learning techniques (ESNs with regularized linear regression) and focuses on practical implementation with specific datasets and backtesting metrics, making it more applied than theoretical. The mathematical complexity is moderate due to the reservoir computing framework, but the emphasis is on empirical validation and computational efficiency for trading contexts.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Forecast Intraday Stock Returns<br>Multiple Horizons"] --> B["Methodology<br>Multihorizon Echo State Network"]
  B --> C["Data Input<br>Intraday Equity Data"]
  C --> D["Computational Process<br>Reservoir Computing & ESN Training"]
  D --> E["Key Finding 1<br>Superior Predictive Accuracy<br>vs Classical Methods"]
  D --> F["Key Finding 2<br>High Efficiency<br>Minimal Training Parameters"]
  D --> G["Key Finding 3<br>Scalability & Expressivity<br>Deep RNN Performance"]

Cited References (108)

  1. Theory of Random Forests (2026) — link
  2. From Many Models, One: Macroeconomic Forecasting with Reservoir Ensembles (2025) — link
  3. Echo State Networks for Bitcoin Time Series Prediction (2025) — link
  4. High-Dimensional Learning in Finance — → 2506.03780
  5. Synthetic data for portfolios: a throw of the dice will never abolish chance — → 2501.03993
  6. Markowitz portfolios under transaction costs (2025) — link
  7. Time-Causal VAE: Robust Financial Time Series Generator — → 2411.02947
  8. Can GANs Learn the Stylized Facts of Financial Time Series? — → 2410.09850
and 100 more
  1. Enhancing mean–variance portfolio optimization through GANs-based anomaly detection (2024) — link
  2. Approximation Bounds for Recurrent Neural Networks with Application to Regression (2024) — link
  3. How noise affects memory in linear recurrent networks (2024) — link
  4. Factor-GAN: Enhancing stock price prediction and factor investment with Generative Adversarial Networks (2024) — link
  5. Stocks price prediction based on optimized echo state network by sparrow search algorithm (2024) — link
  6. Fin-GAN: forecasting and classifying financial time series via generative adversarial networks (2024) — link
  7. A Systematic Review of Echo State Networks From Design to Application (2024) — link
  8. Randomized signature methods in optimal portfolio selection — → 2312.16448
  9. A GANs-Based Approach for Stock Price Anomaly Detection and Investment Risk Management (2023) — link
  10. On Correlated Stock Market Time Series Generation (2023) — link
  11. Large Models for Time Series and Spatio-Temporal Data: A Survey and Outlook (2023) — link
  12. Complexity in Factor Pricing Models (2023) — link
  13. Generative AI for End-to-End Limit Order Book Modelling: A Token-Level Autoregressive Generative Model of Message Flow Using a Deep State Space Network — → 2309.00638
  14. Mathematical Analysis of Machine Learning Algorithms (2023) — link
  15. Sampling weights of deep neural networks (2023) — link
  16. Inverse Approximation Theory for Nonlinear Recurrent Neural Networks (2023) — link
  17. Reservoir computing for macroeconomic forecasting with mixed-frequency data (2022) — link
  18. How and When are High-Frequency Stock Returns Predictable? (2022) — link
  19. Missing values handling for machine learning portfolios (2022) — link
  20. The Virtue of Complexity in Return Prediction (2022) — link
  21. Rare Gems: Finding Lottery Tickets at Initialization (2022) — link
  22. Machine learning techniques and data for stock market forecasting: A literature review (2022) — link
  23. Novel hybrid model based on echo state neural network applied to the prediction of stock price return volatility (2021) — link
  24. Deep learning-based exchange rate prediction during the COVID-19 pandemic (2021) — link
  25. ZerO Initialization: Initializing Neural Networks with only Zeros and Ones (2021) — link
  26. Bayesian neural networks for stock price forecasting before and during COVID-19 pandemic (2021) — link
  27. Sanity Checks for Lottery Tickets: Does Your Winning Ticket Really Win the Jackpot? (2021) — link
  28. Random feature neural networks learn Black-Scholes type PDEs without curse of dimensionality (2021) — link
  29. Tabular Data: Deep Learning is Not All You Need (2021) — link
  30. From Black Wednesday to Brexit: Macroeconomic shocks and correlations of equity returns in France, Germany, Italy, Spain, and the United Kingdom (2021) — link
  31. Machine Learning Advances for Time Series Forecasting (2020) — link
  32. Predicting systemic financial crises with recurrent neural networks (2020) — link
  33. Incorporating overnight and intraday returns into multivariate GARCH volatility models (2020) — link
  34. Forecasting energy consumption and wind power generation using deep echo state network (2020) — link
  35. Time series prediction using deep echo state networks (2020) — link
  36. Approximation Bounds for Random Neural Networks and Reservoir Systems (2020) — link
  37. Proving the Lottery Ticket Hypothesis: Pruning is All You Need (2020) — link
  38. Backpropagation algorithms and Reservoir Computing in Recurrent Neural Networks for the forecasting of complex spatiotemporal dynamics (2019) — link
  39. Risk bounds for reservoir computing (2019) — link
  40. Optuna: A Next-generation Hyperparameter Optimization Framework (2019) — link
  41. An ensemble of LSTM neural networks for high‐frequency stock market classification (2019) — link
  42. Stock price prediction based on deep neural networks (2019) — link
  43. Financial Data Forecasting Using Optimized Echo State Network (2018) — link
  44. Recent Advances in Physical Reservoir Computing: A Review (2018) — link
  45. Reservoir Computing Universality With Stochastic Inputs (2018) — link
  46. Echo state networks are universal (2018) — link
  47. Universal features of price formation in financial markets: perspectives from deep learning (2018) — link
  48. The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks (2018) — link
  49. Dynamic portfolio management with views at multiple horizons (2016) — link
  50. A random forest guided tour (2015) — link
  51. Time Series and Panel Data Econometrics (2015) — link
  52. Exchange rate forecasting using echo state networks for trading strategies (2014) — link
  53. Dynamic Trading with Predictable Returns and Transaction Costs (2013) — link
  54. Introduction to Risk Parity and Budgeting (2013) — link
  55. Stock Prediction Based on Phase Space Reconstruction and Echo State Networks (2013) — link
  56. Application of echo state networks in short-term electric load forecasting (2012) — link
  57. Intelligent stock trading system based on improved technical analysis and Echo State Network (2011) — link
  58. Scikit-learn: Machine Learning in Python (2011) — link
  59. Signal Weighting (2010) — link
  60. Reservoir Computing approach to Great Lakes water level forecasting (2010) — link
  61. Reservoir computing approaches to recurrent neural network training (2009) — link
  62. Short-term stock price prediction based on echo state networks (2009) — link
  63. Weighted Sums of Random Kitchen Sinks: Replacing minimization with randomization in learning (2008)
  64. The Tortoise and the Hare: Portfolio Dynamics for Active Managers (2008) — link
  65. Uniform approximation of functions with random bases (2008) — link
  66. Statistical arbitrage in the US equities market (2008) — link
  67. Information Horizon, Portfolio Turnover, and Optimal Alpha Models (2007) — link
  68. Dynamic Portfolio Analysis (2007) — link
  69. A Hybrid Neurogenetic Approach for Stock Forecasting (2007) — link
  70. Universal Approximation using Incremental Constructive Feedforward Networks with Random Hidden Nodes (2006) — link
  71. Nonlinear Time Series: Nonparametric and Parametric Methods (2005) — link
  72. A well-conditioned estimator for large-dimensional covariance matrices (2004) — link
  73. Non‐linear Time Series Models in Empirical Finance (2003) — link
  74. Neural network forecasts of Canadian stock returns using accounting ratios (2003) — link
  75. Nonparametric Statistics for Stochastic Processes: Estimation and Prediction (2000) — link
  76. Neural Network Learning: Theoretical Foundations (1999) — link
  77. FORECASTING WITH ARTIFICIAL NEURAL NETWORKS: THE STATE OF THE ART (1998) — link
  78. Testing the equality of prediction mean squared errors (1997) — link
  79. THE ECONOMETRICS OF FINANCIAL MARKETS (1996) — link
  80. Multi-Task Learning for Stock Selection (1996)
  81. Comparing Predictive Accuracy (1995) — link
  82. Stock performance modeling using neural networks: A comparative study with regression models (1994) — link
  83. Feedforward neural networks with random weights (1992) — link
  84. Portfolio Selection (1971) — link
  85. Intraday Market Return Predictability Culled from the Factor Zoo (2023) — link
  86. Why do tree-based models still outperform deep learning on typical tabular data? (2022) — link
  87. Missing Financial Data (2022) — link
  88. AI and Compute: How Much Longer Can Computing Power Drive Artificial Intelligence Progress? (2022)
  89. THE RIGHT TOOL FOR THE JOB (2017)
  90. A Practical Guide to Applying Echo State Networks (2012) — link
  91. The Model Confidence Set ∗ (2009)
  92. Nonparametric Econometrics: Theory and Practice (2006)
  93. Random Forests (2001) — link
  94. The Elements of Statistical Learning: Data Mining, Inference, and Prediction (2001) — link
  95. The’’echo state’‘approach to analysing and training recurrent neural networks (2001)
  96. Comparative study of stock trend 40 (1998)
  97. Comparative study of stock trend prediction using time delay, recurrent and probabilistic neural networks (1998)
  98. Multivariable Functional Interpolation and Adaptive Networks (1988)
  99. International Journal of Uncertainty, Fuzziness and Knowledge-based Systems F the Vanishing Gradient Problem during Learning Recurrent Neural Nets and Problem Solutions
  100. more flexible output layers (e.g., a fully trainable shallow network) could further improve performance, and whether any gains obtained would be worthwhile given the increase in training complexity

6 of 108 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.