pith. sign in

arxiv: 2512.21804 · v1 · submitted 2025-12-25 · 💻 cs.CV · cs.AI· cs.CE

S&P 500 Stock's Movement Prediction using CNN

Rahul Gupta This is my paper

Pith reviewed 2026-05-16 18:59 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.CE
keywords S&P 500stock movement predictionCNNdeep learningmultivariate time seriesfinancial forecastingimage classification
0
0 comments X

The pith

A convolutional neural network predicts S&P 500 stock movements by treating raw multivariate data as image-like matrices.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper applies a convolutional neural network directly to raw historical stock data for S&P 500 companies to forecast price direction. It includes corporate actions such as splits and dividends without extra feature engineering and reshapes the multivariate sequences into matrices that the network processes like images. This format allows the same model to produce forecasts for any single stock, for sectors, or for an entire portfolio. The approach avoids traditional financial indicators and relies on the CNN to extract patterns from the unaltered market numbers. If the method works, it shows that image-classification tools can handle financial time series when the data is presented in matrix form.

Core claim

Convolutional Neural Network (CNN) is used on the multi-dimensional stock numbers taken from the market mimicking them as a vector of historical data matrices and the model achieves promising results. The predictions can be made stock by stock, i.e., a single stock, sector-wise or for the portfolio of stocks.

What carries the argument

Reformatting raw multivariate stock time series that include splits and dividends into matrix representations fed to a convolutional neural network for binary movement classification.

Load-bearing premise

Raw stock price data with splits and dividends can be converted directly into image-like matrices for a CNN without losing essential information or failing to generalize beyond the training window.

What would settle it

The claim would be refuted if the CNN's accuracy on held-out future data from a different market regime drops to random-guessing levels.

Figures

Figures reproduced from arXiv: 2512.21804 by Rahul Gupta.

Figure 2
Figure 2. Figure 2: Apple Inc. (AAPL) split-unadjusted price movement [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Cleaned data before normalization was applied [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Data-augmentation and sliding window The simple normalization technique was used to get all the data on the same scale to avoid a knock-on effect on your ability to learn. Ensuring standardized feature values by normalizing the input data implicitly and weights all features equally in their representation. The complete dataset used in training and testing of this paper has been loaded from an authorized ma… view at source ↗
Figure 9
Figure 9. Figure 9: Initial execution loss and accuracy MSFT 2-day (T+2) [PITH_FULL_IMAGE:figures/full_fig_p006_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Loss/Accuracy table MODEL RESULTS: Following page contains visual graph plots of losses and accuracies generated after training the proposed method/model and results look promising where JPM (JP Morgan) stock movement forecasting accuracy touches 91% correct results indicate state-of-the-arts prediction. The proposed method outperforms the baselines previously mentioned in this paper and several models in… view at source ↗
read the original abstract

This paper is about predicting the movement of stock consist of S&P 500 index. Historically there are many approaches have been tried using various methods to predict the stock movement and being used in the market currently for algorithm trading and alpha generating systems using traditional mathematical approaches [1, 2]. The success of artificial neural network recently created a lot of interest and paved the way to enable prediction using cutting-edge research in the machine learning and deep learning. Some of these papers have done a great job in implementing and explaining benefits of these new technologies. Although most these papers do not go into the complexity of the financial data and mostly utilize single dimension data, still most of these papers were successful in creating the ground for future research in this comparatively new phenomenon. In this paper, I am trying to use multivariate raw data including stock split/dividend events (as-is) present in real-world market data instead of engineered financial data. Convolution Neural Network (CNN), the best-known tool so far for image classification, is used on the multi-dimensional stock numbers taken from the market mimicking them as a vector of historical data matrices (read images) and the model achieves promising results. The predictions can be made stock by stock, i.e., a single stock, sector-wise or for the portfolio of stocks.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes applying a Convolutional Neural Network (CNN) to predict movements in S&P 500 stocks by framing multivariate raw market data—including splits and dividends as-is—as image-like historical matrices. It claims this yields promising results and supports predictions at the individual stock, sector, or portfolio level, contrasting with prior work that relies on engineered single-dimensional features.

Significance. If supported by concrete held-out performance metrics and proper temporal validation, the approach could demonstrate the viability of treating raw multivariate financial series as CNN inputs without heavy feature engineering, potentially advancing deep learning applications in quantitative finance. However, the current lack of any reported accuracy, baseline comparisons, or validation details prevents assessment of whether the central claim holds.

major comments (2)
  1. [Abstract] Abstract: The central claim that the CNN model 'achieves promising results' is unsupported by any quantitative evidence. No accuracy, precision, recall, F1-score, Sharpe ratio, or other performance scalars are reported, nor are baseline comparisons (e.g., to LSTM, ARIMA, or random walk) or details on the train/test temporal split provided, leaving the empirical assertion untestable.
  2. [Abstract] Abstract/Methods: The input representation is described only at a high level ('multi-dimensional stock numbers ... mimicking them as a vector of historical data matrices'). No tensor shape (e.g., channels × time × features), normalization scheme, handling of missing values, or confirmation that splits/dividends are used without lookahead bias is specified, which is load-bearing for reproducibility and validity of the CNN application.
minor comments (2)
  1. [Abstract] Abstract: Grammatical issues include 'stock consist of S&P 500 index' (should be 'stocks consisting of the S&P 500 index') and 'most these papers' (should be 'most of these papers').
  2. [Abstract] Abstract: The citation style for [1, 2] is incomplete; full references should be provided in the bibliography.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which identify key areas where the manuscript can be strengthened for clarity and reproducibility. We address each major comment below and confirm that revisions will be made to incorporate the requested details.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that the CNN model 'achieves promising results' is unsupported by any quantitative evidence. No accuracy, precision, recall, F1-score, Sharpe ratio, or other performance scalars are reported, nor are baseline comparisons (e.g., to LSTM, ARIMA, or random walk) or details on the train/test temporal split provided, leaving the empirical assertion untestable.

    Authors: We agree that the abstract's claim of 'promising results' requires quantitative support to be testable. In the revised manuscript we will update the abstract to report concrete held-out metrics (accuracy, F1-score) and include explicit baseline comparisons together with the temporal train/test split used. These additions will directly address the empirical gap noted by the referee. revision: yes

  2. Referee: [Abstract] Abstract/Methods: The input representation is described only at a high level ('multi-dimensional stock numbers ... mimicking them as a vector of historical data matrices'). No tensor shape (e.g., channels × time × features), normalization scheme, handling of missing values, or confirmation that splits/dividends are used without lookahead bias is specified, which is load-bearing for reproducibility and validity of the CNN application.

    Authors: We acknowledge that the current description of the input is high-level. The revised methods section will specify the precise tensor shape, the normalization scheme (fitted exclusively on training data), the procedure for missing values, and an explicit statement confirming that splits and dividends are incorporated chronologically with no lookahead bias. revision: yes

Circularity Check

0 steps flagged

No derivation chain or equations present; empirical claim has no load-bearing steps to reduce

full rationale

The manuscript describes treating raw multivariate stock data (including splits/dividends) as image-like matrices and applying a CNN, claiming promising results. No equations, derivations, fitted parameters, or mathematical steps are supplied in the abstract or described text. Citations [1,2] refer to prior traditional methods and do not overlap with the author or serve as self-citation load-bearing for any result. No self-definitional, fitted-input, or ansatz-smuggling patterns exist because there is no derivation chain at all. The central assertion is an empirical performance claim whose grounding is not visible, but this absence precludes any circularity finding rather than creating one.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities. The approach implicitly assumes standard CNN convolution and pooling operations plus the validity of treating financial time series as static image matrices.

pith-pipeline@v0.9.0 · 5521 in / 1104 out tokens · 20756 ms · 2026-05-16T18:59:16.241726+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

30 extracted references · 30 canonical work pages · 3 internal anchors

  1. [1]

    publicly traded companies market capitalization

    Introduction The Standard & Poor's 500 Index or S&P 500 as it called is a weighted index of the 500 largest U.S. publicly traded companies market capitalization. S&P 500 is one of the furthermost commonly quoted American indexes because it is representative o f the largest U.S. public corporations and it focuses on the large-cap sector of the U.S. market....

    work page

  2. [2]

    In addition to the data noise, due to its nature of convolving CNN intuitionally would be better of handling prediction of a stock dividend and split events

    due to its noise, and volatile features make CNN natural choice over LSTM. In addition to the data noise, due to its nature of convolving CNN intuitionally would be better of handling prediction of a stock dividend and split events. The input to our algorithm is a stock’s raw historical time-series numbers (OPEN, HIGH, LOW, CLOSE, VOLUME, ADJ_OPEN, ADJ_HI...

    work page

  3. [3]

    Related Works Persio and Honchar [8] laid the framework for comparing different neural network models for stock prediction , concerning the forecast of their trend movements up or down, in their paper Artificial Neural Networks architectures for stock price prediction: comparisons and applications. Like our data set, the S&P 500 historical time series, pr...

    work page

  4. [4]

    An open-high- low-close-volume (or OHLCV in short) with its ad justed counterpart is used for training and testing this model

    Dataset and Features The dataset we are using in this paper and model is raw data and not the engineered financial data. An open-high- low-close-volume (or OHLCV in short) with its ad justed counterpart is used for training and testing this model. The OHLCV is used to illustrate movements in the price of a financial instrument over time along with volume ...

    work page

  5. [5]

    This work proposes to harness CNN networks to efficiently learn complex non - linear models directly from the raw data, for th e prediction of the financial market trends

    Methods Deep-Learning (DL) [18] approaches that relate to computational algorithms using artificial neural networks with many layers allows to directly analyze raw data measurements without having to encode the measurements in task -specific representations. This work proposes to harness CNN networks to efficiently learn complex non - linear models direct...

    work page

  6. [6]

    Experiments BASE MODEL ACCURACIES: The base model [ 23] accuracy after training the model for around an hour with different sets of data and multiple stocks the highest accuracy achieved was 69%. Additionally, to provide further perspective, fo llowing image demonstrates the prediction accuracies from some of the Deep Learning model s implemented in previ...

    work page

  7. [7]

    We can also utilize the existing model for predicti on of movement of S&P 500 index once we train the model on 500 constituents historical data

    Conclusion The results achieved by the model proposed in this paper seems promising for forecasting single stock movement as well as sector-wise progression. We can also utilize the existing model for predicti on of movement of S&P 500 index once we train the model on 500 constituents historical data. There can be many ideas for future extensions or even ...

    work page

  8. [8]

    An empirical comparison of machine learning models for time series forecasting,

    Ahmed et al., "An empirical comparison of machine learning models for time series forecasting," Econometric Reviews, 2010

    work page 2010

  9. [9]

    Machine learning strategies for time series forecasting,

    Bontempi, "Machine learning strategies for time series forecasting," in Business Intelligence, Springer,, 2013, pp. 62-77

    work page 2013

  10. [10]

    Standard & Poor's 500 Index - S&P 500,

    "Standard & Poor's 500 Index - S&P 500," [Online]. Available: https://www.investopedia.com/terms/s/sp500.asp

    work page

  11. [11]

    Forecasting Economics and Financial Time Series: ARIMA vs. LSTM

    S. Siami-Namini and A. S. Namin, "Forecasting Economics and Financial Time Series: ARIMA vs. LSTM," 16 Mar 2018. [Online]. Available: https://arxiv.org/abs/1803.06386

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  12. [12]

    Notes on LSTMs for Time Series Prediction in Finance,

    L. Schuermann, "Notes on LSTMs for Time Series Prediction in Finance," Big Theta, 6 Apr 2017. [Online]. Available: http://bigtheta.io/2017/04/06/notes -on-lstms-in-finance.html

    work page 2017

  13. [13]

    A Spatial Mapping Algorithm with Applications in Deep Learning-Based Structure Classification

    Corcoran et al., "A Spatial Mapping Algorithm with Applications in Deep Learning-Based Structure Classification," 22 Feb 2018. [Online]. Available: https://arxiv.org/pdf/1802.02532.pdf

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  14. [14]

    A novel text mining approach to financial time series forecasting,

    Wang et al., "A novel text mining approach to financial time series forecasting," 15 Apr 2012. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0925231211007302

    work page 2012

  15. [15]

    Artificial Neural Networks architectures for stock price prediction: Comparisons and applications,

    L. D. Persio and O. Honchar, "Artificial Neural Networks architectures for stock price prediction: Comparisons and applications," International Journal of Circuits, vol. 10, no. Systems and Signal Processing, p. 403–413, 2016

    work page 2016

  16. [16]

    Financial Time Series Prediction Using Deep Learning,

    A. Navon and Y. Keller, "Financial Time Series Prediction Using Deep Learning," Electrical Engineering and Systems Science, vol. Signal Processing, 2017

    work page 2017

  17. [17]

    Predicting Time Series with Space-Time Convolutional and Recurrent Neural Networks,

    Groß et al., "Predicting Time Series with Space-Time Convolutional and Recurrent Neural Networks," in ESANN 2017, Bruges (Belgium), 2017

    work page 2017

  18. [18]

    Conditional Time Series Forecasting with Convolutional Neural Networks

    Borovykh et al., "Conditional Time Series Forecasting with Convolutional Neural Networks," Statistics > Machine Learning, 14 Mar 2017. [Online]. Available: https://arxiv.org/abs/1703.04691

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  19. [19]

    Stock Market Prediction with Deep Learning: A Character-based Neural Language Model for Event-based 9 Trading,

    L. d. S. Pinheiro and M. Dras, "Stock Market Prediction with Deep Learning: A Character-based Neural Language Model for Event-based 9 Trading," Macquarie University, 2017

    work page 2017

  20. [20]

    Stock price prediction using LSTM, RNN and CNN- sliding window model,

    Selvin et al., "Stock price prediction using LSTM, RNN and CNN- sliding window model," in ResearchGate, 2017

    work page 2017

  21. [21]

    Siripurapu, Convolutional Networks for Stock Trading, Stanford, 2015

    A. Siripurapu, Convolutional Networks for Stock Trading, Stanford, 2015

    work page 2015

  22. [22]

    Bińkowski et al., Autoregressive Convolutional Neural Networks for Asynchronous Time Series, Computer Science > Learning, 2017

    work page 2017

  23. [23]

    A Deep Neural-Network Based Stock Trading System Based on Evolutionary Optimized Technical Analysis Parameters,

    O. B. Sezer, M. Ozbayoglu and E. Dogdu, "A Deep Neural-Network Based Stock Trading System Based on Evolutionary Optimized Technical Analysis Parameters," Procedia Computer Science, vol. 114, pp. 473-480, 2017

    work page 2017

  24. [24]

    Stock Dividends And Stock Splits,

    Investopedia, "Stock Dividends And Stock Splits," Investopedia, [Online]. Available: https://www.investopedia.com/walkthrough/corporate- finance/5/dividends/stock-dividends-splits.aspx

    work page

  25. [25]

    Deep learning,

    Y. LeCun, Y. Bengio and G. Hinton, "Deep learning," 2015

    work page 2015

  26. [26]

    Classification-based financial markets prediction using deep neural networks,

    M. F. Dixon, D. Klabjan and J. Hoon, "Classification-based financial markets prediction using deep neural networks," 2016

    work page 2016

  27. [27]

    Neural Network for Sentiment Analysis,

    L. Lenc and T. Hercig, "Neural Network for Sentiment Analysis," pp. 48-55, 2016

    work page 2016

  28. [28]

    Loss Functions,

    "Loss Functions," Readthedocs.io, 2017. [Online]. Available: http://ml- cheatsheet.readthedocs.io/en/latest/loss_functions.html

    work page 2017

  29. [29]

    Karpathy, "CS231n," Stanford University, 01 04 2018

    A. Karpathy, "CS231n," Stanford University, 01 04 2018. [Online]. Available: http://cs231n.github.io/

    work page 2018

  30. [30]

    Convolutional Neural Stock Market Technical Analyser,

    P. Xu, "Convolutional Neural Stock Market Technical Analyser," GitHub, 3 6 2017. [Online]. Available: https://github.com/philipxjm/Convolutional-Neural-Stock-Market- Technical-Analyser/blob/master/README.md

    work page 2017