Paper: arXiv 2506.21246

Authors: Giorgos Demosthenous, Chryssis Georgiou, Eliada Polydorou

Abstract

This study investigates the impact of data source diversity on the performance of cryptocurrency forecasting models by integrating various data categories, including technical indicators, on-chain metrics, sentiment and interest metrics, traditional market indices, and macroeconomic indicators. We introduce the Crypto100 index, representing the top 100 cryptocurrencies by market capitalization, and propose a novel feature reduction algorithm to identify the most impactful and resilient features from diverse data sources. Our comprehensive experiments demonstrate that data source diversity significantly enhances the predictive performance of forecasting models across different time horizons. Key findings include the paramount importance of on-chain metrics for both short-term and long-term predictions, the growing relevance of traditional market indices and macroeconomic indicators for longer-term forecasts, and substantial improvements in model accuracy when diverse data sources are utilized. These insights help demystify the short-term and long-term driving factors of the cryptocurrency market and lay the groundwork for developing more accurate and resilient forecasting models.

Complexity vs Empirical Score

  • Math Complexity: 5.0/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs standard statistical and machine learning methods (like feature reduction algorithms and regression models) without heavy mathematical derivations, but demonstrates strong empirical rigor through a comprehensive dataset, backtesting on the Crypto100 index, and public artifact availability.

Research Flowchart

  flowchart TD
  A["Research Goal: Assessing Importance of Data Source Diversity"] --> B
  subgraph B ["Methodology"]
      B1["Crypto100 Index: Top 100 Cryptos by MCap"]
      B2["Feature Reduction Algorithm"]
      B3["Multi-source Integration"]
  end
  B --> C
  subgraph C ["Data Sources"]
      C1["Technical Indicators"]
      C2["On-chain Metrics"]
      C3["Sentiment & Interest"]
      C4["Traditional Market Indices"]
      C5["Macroeconomic Indicators"]
  end
  C --> D["Computational Analysis across Time Horizons"]
  D --> E
  subgraph E ["Key Findings"]
      E1["On-chain metrics are paramount for both short & long-term forecasts"]
      E2["Traditional & Macro indicators gain relevance for long-term forecasts"]
      E3["Data diversity significantly improves model accuracy & resilience"]
  end

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 introduces the Crypto100 index and a novel feature reduction algorithm to select the most impactful and resilient features from diverse data sources.

Source sentence
We introduce the Crypto100 index, representing the top 100 cryptocurrencies by market capitalization, and propose a novel feature reduction algorithm to identify the most impactful and resilient features from diverse data sources.

Assumes 3

  • Bitcoin’s on-chain metrics are representative of the broader cryptocurrency market. data
    Source sentence
    it was assumed and later validated (Subsection 4.1) that Bitcoin’s on-chain metrics could also affect the rest of the market, and were therefore included in the dataset as market representatives.
  • USDT data is excluded due to its bad reputation and questionable activity, which could mislead the study. data
    Source sentence
    Although USDT has a much higher marketcap we chose not to include it due to it’s bad reputation and questionable activity [10], which might negatively influence this study leading to misleading conclusions.
  • Technical indicators constructed from BTC historical market information are highly correlated with and influential on the broader cryptocurrency market. model
    Source sentence
    Technical indicators were constructed using only BTC historical market information, as we found them to be highly correlated with and influential on the broader cryptocurrency market.

Data

  • Crypto100 index alternative public
    top 100 cryptocurrencies by market capitalization
    Source sentence
    We introduce the Crypto100 index, representing the top 100 cryptocurrencies by market capitalization
  • On‑chain and market data (daily) · Coinmetrics.io, CoinGecko.com, European Central Bank (ECB) crypto-onchain public
    Source sentence
    For each data category, daily data were collected from sources such as Coinmetrics.io [7], CoinGecko.com [1], the European Central Bank (ECB) [3], and others [14, 24].
  • USDC on‑chain metrics · Coinmetrics.io crypto-onchain public
    USDC stablecoin
    Source sentence
    To investigate the importance of stablecoins, on-chain information from the second largest stablecoin by market capitalization, USDC, were also gathered.

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • Data source diversity significantly enhances forecasting model performance across time horizons. performance
    Source sentence
    Our comprehensive experiments demonstrate that data source diversity significantly enhances the predictive performance of forecasting models across different time horizons.
  • On‑chain metrics are the most important data category for both short‑term and long‑term predictions. predictive-accuracy
    Source sentence
    Key findings include the paramount importance of on-chain metrics for both short-term and long-term predictions
  • Random Forest model overall MSE improvement averaged 455.67% (2017) and 426.67% (2019) when using diverse data sources. performance
    Source sentence
    Overall, the performance of the RF forecasting model improved on average by 455.67% and 426.67% in 2017 and 2019 sets, respectively.
  • Sentiment and Interest Metrics achieved the largest category‑wise improvement (up to 1118.16%). performance
    Source sentence
    ... benefited the most from data source diversity with the improvement level reaching all the way up to 1118.16%.

Checked by others

Supported by

Code

https://github.com/gdemos01/FAB-2024 — the authors' own repository

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (25)

  1. A unified approach to interpreting model predictions (2017)
  2. Some simple economics of stablecoins (2022)
  3. Eleven quick tips for data cleaning and feature engineering (2022)
  4. Grid Search (2024)
  5. Origins of the financial market crisis of 2008 (2009)
  6. Bitcoin: A Peer-to-Peer Electronic Cash System — link
  7. Cryptocurrency Categories (2024)
  8. Crypto Market Sizing 2023 (2024) — link
and 17 more
  1. Cryptocurrency trading: a comprehensive survey (2022)
  2. Convolution on neural networks for high-frequency trend prediction of cryptocurrency exchange rates using technical indicators (2020)
  3. On technical trading and social media indicators for cryptocurrency price classification through deep learning (2022)
  4. Predicting Ethereum prices with machine learning based on Blockchain information (2021)
  5. Forecasting and trading cryptocurrencies with machine learning under changing market conditions (2021)
  6. Predicting Cryptocurrencies Market Phases through On-Chain Data Long-Term Forecasting (2023)
  7. CoinMetrics: Crypto Intelligence for the Future of Finance (2024)
  8. Harmonised Index of Consumer Prices (HICP) (2024)
  9. Measuring Economic Policy Uncertainty (2024)
  10. LunarCrush: Social Intelligence for Crypto (2024)
  11. How Is the S&P 500 Index Calculated? (2024)
  12. On the efficiency and its drivers in the cryptocurrency market: the case of Bitcoin and Ethereum (2024)
  13. Past, present, and future of the application of machine learning in cryptocurrency research (2022)
  14. Portfolio optimization in the era of digital financialization using cryptocurrencies (2020)
  15. Vanguard Short-Term Bond ETF (BSV) Profile (2024)
  16. Invesco DB US Dollar Index Bullish Fund (UUP) Profile (2024)
  17. Is Bitcoin really untethered? (2020)

0 of 25 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.