Paper: arXiv 2412.19932
Authors: Kamil Ł. Szydłowski, Jarosław A. Chudziak
Abstract
This paper investigates the application of Transformer-based neural networks to stock price forecasting, with a special focus on the intersection of machine learning techniques and financial market analysis. The evolution of Transformer models, from their inception to their adaptation for time series analysis in financial contexts, is reviewed and discussed. Central to our study is the exploration of the Hidformer model, which is currently recognized for its promising performance in time series prediction. The primary aim of this paper is to determine whether Hidformer will also prove itself in the task of stock price prediction. This slightly modified model serves as the framework for our experiments, integrating the principles of technical analysis with advanced machine learning concepts to enhance stock price prediction accuracy. We conduct an evaluation of the Hidformer model’s performance, using a set of criteria to determine its efficacy. Our findings offer additional insights into the practical application of Transformer architectures in financial time series forecasting, highlighting their potential to improve algorithmic trading strategies, including human decision making.
Complexity vs Empirical Score
- Math Complexity: 4.0/10
- Empirical Rigor: 6.0/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper focuses on applying a known Transformer architecture (Hidformer) to financial data with standard deep learning practices like MSE loss, showing moderate empirical rigor through evaluation, but lacks deep mathematical derivations, keeping math complexity low.
Research Flowchart
flowchart TD
A["Research Goal:<br>Assess Hidformer for<br>Stock Price Forecasting"] --> B["Data Preparation<br>Financial Time Series<br>Technical Indicators"]
B --> C["Model Architecture<br>Modified Hidformer<br>Transformer-based"]
C --> D["Experimental Process<br>Training & Validation"]
D --> E{"Performance Evaluation"}
E --> F["Key Finding 1: Positive<br>Effective for Stock Prediction"]
E --> G["Key Finding 2: Impact<br>Potential for Algorithmic Trading"]
E --> H["Key Finding 3: Synthesis<br>ML + Technical Analysis"]
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 using the Hidformer transformer-style neural network to predict future stock prices based on past quotations, extending it to forecast multiple future quotes.
Source sentence
Below, we present an idea on how to use the Hidformer artificial neural network to predict future stock prices
Assumes 5
- Technical analysis premises: market discounts everything, prices move in trends, and history repeats itself. market
Source sentence
The justification for using technical analysis as a tool for market analysts is based on three premises [10]. First, the market discounts everything, meaning that all factors affecting market prices are reflected in those prices. Second, prices move in trends, indicating that an existing trend will continue until it re
- Daily valuation frequency is used to balance trend capture and data complexity. data
Source sentence
For our study, we assume a set of parameters based on daily valuations, which provide a balance between capturing market trends and managing data complexity [11].
- Prediction uses the previous T_x prices and volumes to forecast the close price on the next T_y trading days. model
Source sentence
In each prediction, we use the previous Tx prices and volumes to predict the close price on the next Ty trading days, utilizin…
- Trading strategy assumes buying one unit if predicted close > current close, otherwise short one unit, with returns calculated as log price change times sign of prediction difference. execution
Source sentence
If the predicted close price ˆyt+1 is larger than the latest observed close price yt, we long one position index (buy one unit of the index); otherwise, we short one position index (sell one unit of the index). Thus, the return at time t + 1 is expressed as follows (yt+1 is the true next close price): Rt+1 = ln yt+1 /
Show 1 more
- Normalization of price data to [0,1] range using min‑max scaling across all dimensions. data
Source sentence
where ˆxt is the normalized price at time t and xmax (xmin) is the maximum (the minimum) value in the entire sample regarding all dimensions of the prices.
Data
equities · daily
- 6 stock indices from Consumer Goods sector market-prices public
Source sentence
As a historical price dataset we selected the 43-year price movements from 12/12/1980 to 29/12/2023 of 6 stock indices (see Table 1) which are part of the Consumer Goods sector. We used Yahoo Finance as the data source.
Costs
Not modeled
Finds 1
- Hidformer achieved the lowest MAE (0.159) among compared models, with comparable MSE and MAPE, and statistically significant improvements (p‑values < 0.01). performance
Source sentence
Hidformer 0.159 (0.001) 0.040 (0.001) 66 (1) The preliminary results of the experiments demonstrate that Hidformer out‑performs in many cases other classic methods significantly (P-values lower than 0.01) or achieves at least similar results both from the perspective of absolute errors (MAE and MSE) and relative error
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 37 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
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.