Paper: arXiv 2409.03762
Authors: Gabriel Rodrigues Palma, Mariusz Skoczeń, Phil Maguire
Abstract
The decisions traders make to buy or sell an asset depend on various analyses, with expertise required to identify patterns that can be exploited for profit. In this paper we identify novel features extracted from emergent and well-established financial markets using linear models and Gaussian Mixture Models (GMM) with the aim of finding profitable opportunities. We used approximately six months of data consisting of minute candles from the Bitcoin, Pepecoin, and Nasdaq markets to derive and compare the proposed novel features with commonly used ones. These features were extracted based on the previous 59 minutes for each market and used to identify predictions for the hour ahead. We explored the performance of various machine learning strategies, such as Random Forests (RF) and K-Nearest Neighbours (KNN) to classify market movements. A naive random approach to selecting trading decisions was used as a benchmark, with outcomes assumed to be equally likely. We used a temporal cross-validation approach using test sets of 40%, 30% and 20% of total hours to evaluate the learning algorithms’ performances. Our results showed that filtering the time series facilitates algorithms’ generalisation. The GMM filtering approach revealed that the KNN and RF algorithms produced higher average returns than the random algorithm.
Complexity vs Empirical Score
- Math Complexity: 6.0/10
- Empirical Rigor: 5.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced machine learning methods (GMM, RF, KNN) and complex feature engineering, constituting moderate mathematical density. It also includes significant empirical components such as real market data, temporal cross-validation, and backtesting-like performance metrics, though the data span is relatively short.
Research Flowchart
flowchart TD
A["Research Goal<br>Predict Financial Market<br>movements for profit"] --> B{"Data Acquisition"}
B --> C["Inputs: 6 months of minute candles<br>Bitcoin, Pepecoin, Nasdaq<br>59-minute windows for hour-ahead prediction"]
C --> D["Feature Extraction & Filtering"]
D --> E["Method A: Linear Models<br>Method B: GMM"]
E --> F["Classification Algorithms"]
F --> G["Random Forest (RF)"]
F --> H["K-Nearest Neighbours (KNN)"]
G --> I["Evaluation: Temporal Cross-Validation"]
H --> I
I --> J["Benchmark: Random Selection<br>(Naive trading)"]
J --> K["Key Findings/Outcomes"]
K --> L["GMM Filtering + KNN/RF<br>produced higher average returns<br>than random benchmark<br>Improved algorithm generalization"]
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 a new set of features based on linear model slopes, intercepts and peak curvature, and applies Gaussian Mixture Model filtering to improve machine‑learning trading performance.
Source sentence
In this paper, we propose a new set of features designed based on the slope and intercept of a linear model fitted to price peaks. The curvature of each peak is used as an explanatory variable
Assumes 4
- Trading decisions are based on a symmetric threshold algorithm using the 4% quantile of percentage‑change series to create a balanced dataset. market
Source sentence
We defined trading decisions for all markets by employing a simple symmetric threshold algorithm based on the 4% quantile of the Ωm. We selected the specified quantile to produce a
- Baseline naive algorithm assumes each of the three decisions (buy, sell, hold) is equally likely. model
Source sentence
Before employing any learning algorithm, we defined a baseline algorithm representing a naive prediction where a decision dm(t) for a market m on minute t is obtained, assuming that the three possible decisions are equally likely to be sampled.
- Features are computed using only past information up to minute 59 to predict minute 60, respecting temporal causality. data
Source sentence
computed for the entire time series as proposed by Parente et al. (2024), and their values at time t = 59 were used to predict t = 60, respecting the temporal component and not including any future information on feature calculation.
- The number of GMM components is selected by the Bayesian Information Criterion (BIC) for each market. model
Source sentence
We selected the number of GMM components (i.e. clusters) based on the BIC criterion, selecting the model with the lowest values of BIC for each market.
Data
- Bitcoin minute candles crypto-onchain public
Source sentence
... markets, with minute candles describing the open, close, high and low, using the OKX API.
- Pepecoin minute candles crypto-onchain public
Source sentence
... Pepecoin markets, with minute candles describing the open, close, high and low, using the OKX API.
- Nasdaq minute candles market-prices public
Source sentence
Also, we used the backtest‑market API to collect minute candles for the Nasdaq market in the same period.
Costs
Not modeled
Finds 5
- KNN and Random Forests trained on the proposed features achieved higher average returns than the random baseline. performance
Source sentence
The GMM filtering approach revealed that the KNN and RF algorithms produced higher average returns than the random algorithm.
- Random Forests achieved the highest accumulated percentage change (APC) for Bitcoin when using the proposed features. performance
Source sentence
The highest APC obtained was with the Random Forests algorithm based on the proposed features.
- KNN achieved the highest APC for Nasdaq using the proposed features, with no clear improvement from standardisation. performance
Source sentence
The highest APC obtained was with the KNN algorithm, which also used the proposed features.
- KNN trained with the proposed features produced the highest APC for the Pepecoin market. performance
Source sentence
The KNN algorithm trained with the proposed features produced the highest APC for the Pepecoin market.
- Filtering the time series with GMM improves average performance over the naive equal‑probability algorithm. performance
Source sentence
Our results revealed the potential of the proposed features, showing that filtering the time series enhances average performance over a naive algorithm that assigns equal probability for each trading decision.
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.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (39)
- A profitable trading algorithm for cryptocurrencies using a Neural Network model (2024)
- Short-term bitcoin market prediction via machine learning (2021) — link
- Modeling traders’ behavior with deep learning and machine learning methods: evidence from BIST 100 index (2020)
- Cryptocurrency price forecasting – A comparative analysis of ensemble learning and deep learning methods (2024) — link
- Evaluating machine learning classification for financial trading: An empirical approach (2016) — link
- Importance of machine learning in making investment decision in stock market (2021)
- Interpretable trading pattern designed for machine learning applications (2023) — link
- A review of machine learning experiments in equity investment decision-making: why most published research findings do not live up to their promise in real life (2021)
and 31 more
- Clustering Analysis of Stock Volume and Price Relationship based on Gaussian Mixture Model (2014)
- Gaussian mixture models for time series modelling, forecasting, and interpolation (2013)
- Preprocessing large datasets using Gaussian mixture modelling to improve prediction accuracy of truck productivity at mine sites (2022)
- Using deep neural networks coupled with principal component analysis for ore production forecasting at open-pit mines (2024)
- Understanding learning from EEG data: Combining machine learning and feature engineering based on hidden Markov models and mixed models (2023) — link
- Pattern-based prediction of population outbreaks (2023)
- Improving stock trading decisions based on pattern recognition using machine learning technology (2021)
- Decision-making for financial trading: A fusion approach of machine learning and portfolio selection (2019)
- Cryptocurrency trading: a comprehensive survey (2022)
- A hybrid decision support system for adaptive trading strategies: Combining a rule-based expert system with a deep reinforcement learning strategy (2024)
- A hybrid stock trading framework integrating technical analysis with machine learning techniques (2016)
- A novel deep reinforcement learning based automated stock trading system using cascaded lstm networks — → 2212.02721
- Deep reinforcement learning for financial trading using multi-modal features (2024)
- Evaluating machine learning classification for financial trading: An empirical approach (2016)
- Forecasting and trading cryptocurrencies with machine learning under changing market conditions (2021)
- Trading patterns of institutional investors: applications of machine learning (2024)
- A new method for stock price prediction based on MRFs and SSVM (2017)
- Financial Market Prediction Using Deep Neural Networks with Hardware Acceleration (2022)
- Optimization of mixture models on time series networks encoded by visibility graphs: an analysis of the US electricity market (2023)
- Three-State Opinion Dynamics for Financial Markets on Complex Networks (2023)
- Time Series Forecast With Stock’s Price Candlestick Patterns And Sequence Similarities (2024)
- Ensemble Learning with an Adversarial Hypergraph Model and a Convolutional Neural Network to Forecast Stock Price Variations. (2024)
- A general framework for portfolio construction based on generative models of asset returns — → 2312.03294
- Dynamic clustering of financial assets (2014)
- Deep learning models for price forecasting of financial time series: A review of recent advancements: 2020–2022 — → 2305.04811
- Special issue on feature engineering editorial (2021)
- Vest: Automatic feature engineering for forecasting (2021)
- Predicting stock returns of Tehran exchange using LSTM neural network and feature engineering technique (2021)
- Deep learning-based feature engineering for stock price movement prediction (2019)
- Imaging feature-based clustering of financial time series (2023)
- A novel feature engineering approach for high-frequency financial data (2023)
3 of 39 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.