[Paper Review] C2P2: A Collective Cryptocurrency Up/Down Price Prediction Engine

Chongyang Bai, Tommy E. White+3 more|arXiv (Cornell University)|Jun 3, 2019

Blockchain Technology Applications and Security21 references4 citations

TL;DR

C2P2 proposes a collective cryptocurrency price prediction model that jointly predicts up/down movements for 21 major cryptocurrencies by leveraging inter-cryptocurrency similarities and iterative collective classification. It outperforms prior methods by 5.1–83% in AUC, with statistically significant improvements, especially through similarity-based feature integration.

ABSTRACT

We study the problem of predicting whether the price of the 21 most popular cryptocurrencies (according to coinmarketcap.com) will go up or down on day d, using data up to day d-1. Our C2P2 algorithm is the first algorithm to consider the fact that the price of a cryptocurrency c might depend not only on historical prices, sentiments, global stock indices, but also on the prices and predicted prices of other cryptocurrencies. C2P2 therefore does not predict cryptocurrency prices one coin at a time --- rather it uses similarity metrics in conjunction with collective classification to compare multiple cryptocurrency features to jointly predict the cryptocurrency prices for all 21 coins considered. We show that our C2P2 algorithm beats out a recent competing 2017 paper by margins varying from 5.1-83% and another Bitcoin-specific prediction paper from 2018 by 16%. In both cases, C2P2 is the winner on all cryptocurrencies considered. Moreover, we experimentally show that the use of similarity metrics within our C2P2 algorithm leads to a direct improvement for 20 out of 21 cryptocurrencies ranging from 0.4% to 17.8%. Without the similarity component, C2P2 still beats competitors on 20 out of 21 cryptocurrencies considered. We show that all these results are statistically significant via a Student's t-test with p<1e-5. Check our demo at https://www.cs.dartmouth.edu/dsail/demos/c2p2

Motivation & Objective

• To address the limitations of univariate cryptocurrency price prediction by modeling interdependencies among multiple cryptocurrencies.
• To improve up/down price movement prediction accuracy by incorporating cross-coin similarities and collective classification.
• To evaluate the impact of diverse feature types—including Reddit sentiment, stock indices, and time-lagged features—on predictive performance.
• To demonstrate that collective modeling with iterative refinement and similarity features significantly enhances prediction robustness and accuracy.
• To provide a statistically validated, real-world applicable framework for forecasting daily price direction across major cryptocurrencies.

Proposed method

• C2P2 uses a collective classification framework that jointly predicts price direction (up/down) for all 21 cryptocurrencies simultaneously, rather than independently.
• It computes five types of similarity metrics between feature vectors of all cryptocurrency pairs to model interdependence and shared dynamics.
• The algorithm employs an iterative prediction process: initial tentative predictions are refined in subsequent rounds using predictions from other coins as input features.
• Time-lagged features are incorporated to capture temporal dependencies in price, sentiment, and market indicators.
• A combination of feature selection, optimal time lags, and classification algorithms (e.g., XGBoost, Random Forest) is empirically tuned for each coin.
• Reddit sentiment data and global financial indicators (e.g., stock indices, gold, oil, bonds) are used as input features to reflect market sentiment and macroeconomic influence.

Experimental results

Research questions

• RQ1 Can modeling interdependencies among multiple cryptocurrencies improve the accuracy of up/down price movement predictions compared to univariate models?
• RQ2 To what extent do similarity metrics between cryptocurrency feature vectors enhance predictive performance in collective classification?
• RQ3 How does iterative refinement using tentative predictions from other coins affect final prediction accuracy?
• RQ4 What is the relative contribution of different feature types—such as sentiment, market indices, and price history—to prediction performance?
• RQ5 Does the inclusion of similarity features lead to statistically significant improvements across diverse cryptocurrencies?

Key findings

• C2P2 achieves a 5.1–83% improvement in AUC over a recent 2017 competitive baseline across all 21 cryptocurrencies, with all gains being statistically significant (p < 10⁻⁵).
• C2P2 outperforms a 2018 Bitcoin-specific prediction model by 16% in AUC, again with all results being statistically significant.
• The inclusion of similarity features improves AUC for 20 out of 21 cryptocurrencies by 0.4% to 17.8%, with all improvements being statistically significant.
• Even without similarity features, C2P2 still outperforms competitors on 20 out of 21 cryptocurrencies, demonstrating robustness.
• An ablation study confirms that similarity features are a key driver of performance, as their removal causes a significant drop in prediction accuracy.
• The rolling window evaluation strategy, used to simulate real-world deployment, confirms the model’s strong generalization and temporal stability.