Paper: arXiv 2506.03153
Authors: Junzhe Jiang, Chang Yang, Xinrun Wang, Bo Li
Abstract
Stock market indices serve as fundamental market measurement that quantify systematic market dynamics. However, accurate index price prediction remains challenging, primarily because existing approaches treat indices as isolated time series and frame the prediction as a simple regression task. These methods fail to capture indices’ inherent nature as aggregations of constituent stocks with complex, time-varying interdependencies. To address these limitations, we propose Cubic, a novel end-to-end framework that explicitly models the adaptive fusion of constituent stocks for index price prediction. Our main contributions are threefold. i) Fusion in the latent space: we introduce the fusion mechanism over the latent embedding of the stocks to extract the information from the vast number of stocks. ii) Binary encoding classification: since regression tasks are challenging due to continuous value estimation, we reformulate the regression into the classification task, where the target value is converted to binary and we optimize the prediction of the value of each digit with cross-entropy loss. iii) Confidence-guided prediction and trading: we introduce the regularization loss to address market prediction uncertainty for the index prediction and design the rule-based trading policies based on the confidence. Extensive experiments across multiple stock markets and indices demonstrate that Cubic consistently outperforms state-of-the-art baselines in stock index prediction tasks, achieving superior performance on both forecasting accuracy metrics and downstream trading profitability.
Complexity vs Empirical Score
- Math Complexity: 6.5/10
- Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs only partially modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)
Why this score: The paper introduces advanced deep learning concepts like latent space fusion and binary encoding classification with specific loss functions, indicating moderate-to-high mathematical complexity. It also reports extensive experiments across multiple markets and backtested trading profitability, demonstrating strong empirical rigor.
Research Flowchart
flowchart TD
subgraph Problem & Goal
A["Research Question<br>Why Regression?"]
B["Challenge: Index aggregation<br>ignores constituent stocks"]
end
subgraph Methodology
C["Cubic Framework<br>Adaptive Fusion Model"]
D["Binary Encoding Classification<br>Replace regression with<br>binary digit classification"]
E["Confidence-guided Trading<br>Regularization + Rule-based policy"]
end
subgraph Outcomes
F["Superior Forecast Accuracy<br>vs SOTA baselines"]
G["Enhanced Trading Profitability<br>via confidence-weighted decisions"]
end
A --> B
B --> C
C --> D
D --> E
E --> F
E --> G
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 CUBIC, an end‑to‑end framework that fuses constituent stocks in latent space, uses binary‑encoding classification, and adds confidence‑guided prediction and trading.
Source sentence
we propose CUBIC, a novel end-to-end framework that explicitly models the adaptive fusion of constituent stocks for index price prediction.
Assumes 3
- A flat transaction cost of 0.1% per trade is applied to all trading results. execution
Source sentence
We consider a transaction cost of 0.1% for each trade.
- The binary encoding of target values provides a robust alternative to continuous regression. model
Source sentence
since regression tasks are challenging due to continuous value estimation, we reformulate the regression into the classification task, where the target value is converted to binary and we optimize the prediction of the value of each digit with cross-entropy loss.
- Confidence outputs from the classifier can be used as natural confidence measures for trading decisions. model
Source sentence
we introduce confidence-guided prediction and trading that leverages classification-based probability outputs as natural confidence measures and incorporates specialized regularization loss to enhance prediction reliability and guide trading decisions.
Data
USA stock market index · equities · daily
Yahoo Finance
Costs
Unclear - sources disagree
Source sentence (the authors' words)
We consider a transaction cost of 0.1% for each trade.
Finds 5
Sharpe 0.92 · annual return 9.2%
- Binary encoding with feature fusion (BN+FS) achieves the highest annualized return of 8.6% and ICLR of 0.450 on the US stock market. performance
Source sentence
BN+FS achieving the highest annualized return of 8.6% and ICLR of 0.450.
- Adding mean confidence (BF+Mean) raises MLP’s ICLR from 0.284 to 0.367 and LSTM’s Sharpe Ratio from 0.560 to 0.733. predictive-accuracy
Source sentence
Adding mean confidence (BF+Mean) improves performance significantly, with MLP’s ICLR increasing from 0.284 to 0.367 and LSTM’s SR from 0.560 to 0.733.
- Combining confidence mechanisms with decision making (BF+Trend+DM) yields the highest Sharpe Ratio of 1.324 for MLP, up from 0.916. performance
Source sentence
BF+Trend+DM achieves the highest SR of 1.324 with MLP, up from 0.916.
- Binary encoding improves direction prediction, with IC for Transformer models increasing from 0.021 to 0.029 and ICLR from 0.234 to 0.310. predictive-accuracy
Source sentence
For Transformer models, BN enhances IC from 0.021 to 0.029 and ICLR from 0.234 to 0.310.
- Fusion in latent space and binary encoding together boost model performance across architectures, e.g., MLP IC from 0.018 to 0.024 and Sharpe ratio from 0.584 to 0.855. performance
Source sentence
For MLP, introducing BN improves the IC from 0.018 to 0.024 and the Sharpe ratio from 0.584 to 0.855.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 36 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 (66)
- Attention is all you need (2017)
- Forecasting international equity market volatility: A new approach (2022)
- Efficient integration of multi-order dynamics and internal dynamics in stock movement prediction (2023)
- Modeling the momentum spillover effect for stock prediction via attribute-driven graph attention networks (2021)
- MASTER: Market-Guided Stock Transformer for Stock Price Forecasting (2024)
- StockMixer: A Simple Yet Strong MLP-Based Architecture for Stock Price Forecasting (2024)
- Stock market index prediction using deep Transformer model (2022)
- Application of LSTM, GRU and ICA for stock price prediction (2019)
and 58 more
- StockNet: Predicting Stock Price Movement Using Attention-Based Recurrent Networks (2017) — link
- Transformer-based attention network for stock movement prediction (2022)
- Forecasting stock index price using the CEEMDAN-LSTM model (2021)
- Predicting Stock Market Index Using Machine Learning Techniques (2015)
- Attention Is All You Need (2017)
- Predicting direction of stock price index movement using artificial neural networks and support vector machines: The sample of the Istanbul Stock Exchange (2011)
- Long short-term memory (2012)
- Stock market index prediction based on reservoir computing models (2021)
- On index investing (2022)
- Deep learning for time series forecasting: a survey (2021)
- High-performance stock index trading via neural networks and trees (2020)
- Deep learning for stock market prediction (2018)
- Deep temporal learning for financial market prediction (2020)
- Temporal dependency in stock market prediction (2023)
- Qlib: An AI-oriented quantitative investment platform (2020) — link
- Deep learning with long short-term memory networks for financial market predictions (2018)
- Forecasting of stock prices using multi layer perceptron (2013)
- Stock market prediction using machine learning techniques: a decade survey on methodologies, recent developments, and future directions (2021)
- Stop regressing: Training value functions via classification for scalable deep rl (2024) — link
- MambaStock: Selective state space model for stock prediction — → 2402.18959
- Mamba: Linear-time sequence modeling with selective state spaces (2023) — link
- Mamba meets financial markets: A graph-mamba approach for stock price prediction — → 2410.03707
- A hybrid ARIMA-LSTM model for time series forecasting of stock market indices (2023)
- Stock market forecasting using ARIMA-based ensemble methods: A case study of the Chinese stock market (2021)
- Deep learning in stock market prediction: A systematic review (2022)
- Transformer-based deep learning models for financial time series forecasting: A comprehensive survey (2023)
- Hybrid Transformer-LSTM Architecture for Financial Time Series Prediction (2023)
- Deep LSTM Networks for Long-term Financial Market Prediction (2023)
- Temporal Attention Enhanced MLP for Financial Time Series Analysis (2023)
- Discrete Classification Framework for Financial Time Series Prediction (2023)
- Discrete representation strategies for foreign exchange prediction (2018)
- Classification-based Market Volatility Prediction Using Deep Learning (2023)
- Categorical Deep Learning for Multi-Asset Portfolio Optimization (2023)
- Performance of deep learning in prediction of stock market volatility. (2019)
- Discrete Representation Learning for Financial Market Prediction (2023)
- Adaptive Classification Frameworks for Financial Time Series Analysis (2023)
- Multi-modal Fusion of Market Information Using Deep Learning Architectures (2023)
- Adaptive Neural Architecture Selection for Financial Market Prediction (2023)
- Hierarchical Deep Learning Architecture for Financial Time Series Analysis (2023)
- Stock market prediction with multi-source heterogeneous information: A deep learning approach (2023)
- Enhanced stock price prediction with optimized ensemble modeling using multi-source heterogeneous data: Integrating LSTM attention mechanism and multidimensional gray model (2024)
- Attention-based deep learning framework for stock market prediction using multi-source information (2023)
- A graph-based approach to multi-source heterogeneous information fusion in stock market (2022)
- A hybrid deep learning framework for stock market prediction combining technical and sentiment analysis (2023)
- An ensemble of a boosted hybrid of deep learning models and technical analysis for forecasting stock prices (2022)
- Stock market prediction via multi-source multiple instance learning (2018)
- Hybridization of ARIMA with learning models for forecasting of stock market time series (2024)
- A Self-Attention-Based Stock Prediction Method Using Long Short-Term Memory Network Architecture (2023)
- Adaptive Long-Short Pattern Transformer for Stock Investment Selection. (2022)
- A hybrid approach of wavelet transform, ARIMA and LSTM model for the share price index futures forecasting (2024)
- Hierarchical Multi-Scale Gaussian Transformer for Stock Movement Prediction. (2020)
- Relation-aware transformer for portfolio policy learning (2021)
- MAPS: Multi-agent reinforcement learning-based portfolio management system (2020) — link
- Series decomposition Transformer with period-correlation for stock market index prediction (2024)
- Long Short-term Memory (1997)
- Understanding the difficulty of training deep feedforward neural networks (2010)
- Delving deep into rectifiers: Surpassing human-level performance on imagenet classification (2015)
- Regression as classification: Influence of task formulation on neural network features (2023)
2 of 66 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.