Paper: arXiv 2504.12771
Authors: Yu Zhang, Zelin Wu, Claudio Tessone
Abstract
Cryptocurrencies are digital tokens built on blockchain technology, with thousands actively traded on centralized exchanges (CEXs). Unlike stocks, which are backed by real businesses, cryptocurrencies are recognized as a distinct class of assets by researchers. How do investors treat this new category of asset in trading? Are they similar to stocks as an investment tool for investors? We answer these questions by investigating cryptocurrencies’ and stocks’ price time series which can reflect investors’ attitudes towards the targeted assets. Concretely, we use different machine learning models to classify cryptocurrencies’ and stocks’ price time series in the same period and get an extremely high accuracy rate, which reflects that cryptocurrency investors behave differently in trading from stock investors. We then extract features from these price time series to explain the price pattern difference, including mean, variance, maximum, minimum, kurtosis, skewness, and first to third-order autocorrelation, etc., and then use machine learning methods including logistic regression (LR), random forest (RF), support vector machine (SVM), etc. for classification. The classification results show that these extracted features can help to explain the price time series pattern difference between cryptocurrencies and stocks.
Complexity vs Empirical Score
- Math Complexity: 3.0/10
- Empirical Rigor: 7.5/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper relies on standard machine learning implementations (MLP, CNN, RNN, etc.) with minimal theoretical derivations, placing it lower in math complexity, but it demonstrates high empirical rigor through detailed data preprocessing, model training, and explicit backtesting-ready procedures on real financial datasets.
Research Flowchart
flowchart TD
A["Research Goal: Compare Price Patterns<br>Cryptocurrencies vs. Stocks"] --> B["Data Collection"]
B --> C["Feature Extraction<br>Mean, Variance, Kurtosis, Autocorrelation"]
C --> D["Machine Learning Classification"]
D --> E{"Model Training & Testing"}
E -- High Accuracy --> F["Key Finding 1:<br>Distinct Price Patterns"]
E -- Feature Importance --> G["Key Finding 2:<br>Features Explain Differences"]
F --> H["Conclusion:<br>Crypto & Stocks are distinct asset classes"]
G --> H
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 various machine learning models to classify cryptocurrency and stock price time series and to extract statistical features to explain their differences.
Source sentence
In this paper, we will mainly use different machine learning methods to classify cryptocurrencies’ and stocks’ price time series and then extract features to explain their differences.
Assumes 5
- Cryptocurrency price points are limited to NY stock market hours to align with stock data. market
Source sentence
To align their price time series, we select cryptocurrency price points exclusively from 9:30 AM to 4:00 PM (New York time) on working days, matching stock market hours.
- Missing values in stock price series are filled with neighboring same‑day values. data
Source sentence
The original stock price time series data have some missing values we filled them with neighboring values from the same day.
- All price series are sampled at the minute level and contain 98,532 points after processing. data
Source sentence
Each time series consists of 98,532 time points (391 minutes per day × 252 days) at the minute level after selection and filling.
- Classification performance is evaluated on both balanced and unbalanced data samples. model
Source sentence
Because the experiments are also run with the configuration of two kinds of different data temporal segmentation and two kinds of data features, then there are four sets of sub‑experiments.
Show 1 more
- Four basic price features (open, high, low, close) are used as inputs for classification. data
Source sentence
For each asset, four key price features (open, high, low, and close) are collected.
Data
- Cryptocurrency minute‑level price series market-prices public
Source sentence
131 popular cryptocurrency price time series data were collected from Binance with USDT as a unit.
- Stock minute‑level price series market-prices commercial
Source sentence
735 stocks’ price time series from the NASDAQ and NYSE exchanges were collected from Bloomberg.
Costs
Not applicable
Finds 4
- All models achieve accuracy rates above 92% across most configurations, with weekly data yielding the highest performance. predictive-accuracy
Source sentence
All models’ accuracy rates with different configurations achieve > 92% in most data feed settings (“balanced” versus “unbalanced”, “weekly” versus “daily”, “four features” versus “one feature”), except one setting with daily unbalanced data.
- When using weekly slices, the smallest observed accuracy and F1 scores are around 98% and 93%, respectively. predictive-accuracy
Source sentence
The classification accuracy rate and F1 score of all models are extremely high when weekly slices of price sample data are used. The smallest accuracy rate and F1 score are
- Classifying return‑rate time series also yields exceptionally high accuracy, confirming that return series retain distinguishing characteristics. predictive-accuracy
Source sentence
Classification of return rate time series using nine machine learning methods still yields exceptionally high performance, indicating that these series retain the key distinguishing characteristics between cryptocurrency and stock price time series.
- Extracted statistical features (mean, variance, kurtosis, skewness, autocorrelations, etc.) partially explain the price‑pattern differences but are insufficient alone. statistical
Source sentence
The results indicate that these features partially explain the pattern differences: while sufficient for some models, they fall short for others, suggesting that additional critical features remain to be identified.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 28 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.