Paper: arXiv 2411.12748

Authors: Mabsur Fatin Bin Hossain, Lubna Zahan Lamia, Md Mahmudur Rahman, Md Mosaddek Khan

Abstract

Time series forecasting is a key tool in financial markets, helping to predict asset prices and guide investment decisions. In highly volatile markets, such as cryptocurrencies like Bitcoin (BTC) and Ethereum (ETH), forecasting becomes more difficult due to extreme price fluctuations driven by market sentiment, technological changes, and regulatory shifts. Traditionally, forecasting relied on statistical methods, but as markets became more complex, deep learning models like LSTM, Bi-LSTM, and the newer FinBERT-LSTM emerged to capture intricate patterns. Building upon recent advancements and addressing the volatility inherent in cryptocurrency markets, we propose a hybrid model that combines Bidirectional Long Short-Term Memory (Bi-LSTM) networks with FinBERT to enhance forecasting accuracy for these assets. This approach fills a key gap in forecasting volatile financial markets by blending advanced time series models with sentiment analysis, offering valuable insights for investors and analysts navigating unpredictable markets.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced deep learning architectures (Bi-LSTM, FinBERT transformers) requiring knowledge of neural networks and sequence modeling. It demonstrates strong empirical rigor by using specific datasets (Bitcoin/Ethereum news and price data), comparing against multiple benchmarks, and mentioning intra-day performance analysis, suggesting thorough backtesting readiness.

Research Flowchart

  flowchart TD
  A["Research Goal: Predict Crypto Volatility"] --> B["Data Collection & Preprocessing"]
  B --> C["Market Data & News Sentiment"]
  C --> D["Computational Process: Hybrid Model"]
  subgraph D ["FinBERT-BiLSTM Architecture"]
      D1["FinBERT: Extracts Sentiment Features"]
      D2["Bi-LSTM: Captures Temporal Dependencies"]
  end
  D1 --> D2
  D2 --> E["Forecast Price & Volatility"]
  E --> F["Key Findings & Outcomes"]
  F --> G["Improved Accuracy vs. Traditional Models"]
  F --> H["Enhanced Risk Management Strategies"]

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 hybrid model that combines FinBERT with a Bidirectional LSTM to forecast cryptocurrency prices.

Source sentence
i. Tailored for the volatile cryptocurrency market, we present a combined model of FinBERT and Bi-LSTM.

Assumes 3

  • If a decision is made to buy or sell, the transaction is assumed to occur at the same price as the closing price of the day d‑1. execution
    Source sentence
    If a decision is made to buy or sell, the transaction is assumed to occur at the same price as the closing price of the day d −1.
  • The required amount of Bitcoin or Ethereum for purchase will always be available on the market, with no liquidity constraints. market
    Source sentence
    It is assumed that the required amount of Bitcoin or Ethereum for purchase will always be available on the market, and transactions will not fail due to liquidity constraints or other technical issues.
  • FinBERT‑integrated models use the sentiment score from day d‑1 because the decision must be made one day in advance. model
    Source sentence
    The FinBERT‑integrated models use the sentiment score from day d −1, as the decision must be made one day in advance.

Data

  • Historical price data · Yahoo Finance market-prices public
    2023-01-01 to 2024-? (585 days) · Bitcoin (BTC) and Ethereum (ETH)
    Source sentence
    The historical price data for Bitcoin (BTC) and Ethereum (ETH) was obtained from Yahoo Finance6,
  • Transaction rate · Coin‑Remitter other public
    Bitcoin (BTC) and Ethereum (ETH)
    Source sentence
    we utilized a transaction rate of 0.0001 BTC for Bitcoin and 0.003 ETH for Ethereum, which includes the gas station optimization feature for Ethereum.
  • News and financial text collection · Coindesk, Binance and other crypto news platforms text-news public
    2023-01-01 to 2024-? (585 days) · Bitcoin (BTC) and Ethereum (ETH)
    Source sentence
    we collected news articles and financial texts from reputable cryptocurrency news platforms, including Coindesk8, Binance9 and other specialized websites

Recognised sources: Twitter / X Yahoo Finance

Costs

Modeled Costs are part of the reported headline results.

Transaction rates of 0.0001 BTC per trade for Bitcoin and 0.003 ETH per trade for Ethereum are applied.

Source sentence
we utilized a transaction rate of 0.0001 BTC for Bitcoin and 0.003 ETH for Ethereum, which includes the gas station optimization feature for Ethereum.

Finds 4

  • FinBERT‑Bi‑LSTM achieves the highest 30‑day future prediction accuracy of 98.25% for Bitcoin, slightly surpassing FinBERT‑LSTM (98.14%). performance
    Source sentence
    FinBERT‑Bi‑LSTM model achieves the highest accuracy of 98.25%, slightly surpass‑ing the FinBERT‑LSTM’s 98.14% showing that our hypothesis mentioned in Section 4 works for future multi‑day forecasting.
  • FinBERT‑Bi‑LSTM reaches 98.21% intra‑day and 98.07% one‑day‑ahead prediction accuracy on Bitcoin, delivering a remarkable profit from the one‑day‑ahead strategy. predictive-accuracy
    Source sentence
    FinBERT‑Bi‑LSTM model achieves an accuracy of 98.21% for intra‑day predictions and 98.07% for one‑day‑ahead predictions on Bitcoin (BTC), resulting in a remarkable profit from the one‑day‑ahead strategy.
  • On Ethereum, the model attains 97.5% intra‑day and 97.36% one‑day‑ahead prediction accuracy, generating a significant profit. predictive-accuracy
    Source sentence
    On Ethereum (ETH), it achieves an accuracy of 97.5% for intra‑day predictions and 97.36% for one‑day‑ahead predictions, generating a significant profit.
  • Maximum Data Training yields 98.3% long‑term accuracy for Bitcoin, while Validation‑Enhanced Training gives 95.25% for Ethereum. performance
    Source sentence
    Maximum Data Training approach yields an impressive accuracy of 98.3% for BTC, while the Validation‑Enhanced Training approach performs better for ETH with an accuracy of 95.25%.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 25 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.

Cited References (40)

  1. Predicting Stock Prices with FinBERT-LSTM: Integrating News Sentiment Analysis (2024)
  2. FinBERT-LSTM: Deep Learning-based Stock Price Prediction Using News Sentiment Analysis (2022)
  3. FinBERT: A Pre-trained Financial Language Representation Model for Financial Text Mining (2020) — link
  4. Speculative bubbles in Bitcoin markets? An empirical investigation into the fundamental value of Bitcoin (2015) — link
  5. Exploring the dynamic relationships between cryptocurrencies and other financial assets (2018) — link
  6. Bitcoin: Medium of exchange or speculative assets? (2018) — link
  7. The inefficiency of Bitcoin (2016) — link
  8. The economics of Bitcoin and similar private digital currencies (2015) — link
and 32 more
  1. GARCH modelling of cryptocurrencies (2017) — link
  2. On the return-volatility relationship in the Bitcoin market around the price crash of 2013 (2017) — link
  3. Deep learning with long short-term memory networks for financial market predictions (2018) — link
  4. Price Movement Prediction of Cryptocurrencies Using Sentiment Analysis and Machine Learning (2019) — link
  5. Sentiment Analysis of News for Effective Cryptocurrency Price Prediction (2019) — link
  6. Cryptocurrencies as a financial asset: A systematic analysis (2018) — link
  7. Spillovers between Bitcoin and other assets during bear and bull markets (2018) — link
  8. Volatility estimation for Bitcoin: A comparison of GARCH models (2017) — link
  9. A multi-channel CNN-BiLSTM model for stock price forecasting using a deep learning method (2020) — link
  10. FinBERT: Financial Sentiment Analysis with Pre-trained Language Models (2019)
  11. Challenge and Opportunity: Deep Learning-Based Stock Price Prediction by Using Bi-Directional LSTM Model (2023) — link
  12. Time Series Analysis: Forecasting and Control (1994)
  13. Economic, legal and financial perspectives on cryptocurrencies: A review on cryptocurrency growth, opportunities and future prospects (2020) — link
  14. Bitcoin: A Peer-to-Peer Electronic Cash System (2008)
  15. Transaction-based classification and detection approach for Ethereum smart contract (2021) — link
  16. Factors Influencing Cryptocurrency Prices: Evidence from Bitcoin, Ethereum, Dash, Litecoin, and Monero (2018)
  17. Bitcoin price prediction using Deep Learning Algorithm (2019) — link
  18. Forecasting Cryptocurrency Prices Using LSTM, GRU, and Bi-Directional LSTM: A Deep Learning Approach (2023) — link
  19. Using sentiment analysis to predict interday Bitcoin price movements (2018) — link
  20. Comparing automated text classification methods (2019) — link
  21. Framewise phoneme classification with bidirectional LSTM networks (2005)
  22. A dual-stage attention-based recurrent neural network for time series prediction (2017)
  23. Long short-term memory (1997) — link
  24. Bidirectional recurrent neural networks (1997) — link
  25. LARNN: Linear Attention Recurrent Neural Network (2018)
  26. Financial sentiment analysis using FinBERT with application in predicting stock movement — → 2306.02136
  27. Hybrid LSTM and GRU for Cryptocurrency Price Forecasting Based on Social Network Sentiment Analysis Using FinBERT (2023) — link
  28. Stock market volatility: a systematic review (2024) — link
  29. Forecasting gold price fluctuations using improved multilayer perceptron neural network and whale optimization algorithm (2019) — link
  30. Predicting COVID-19 cases using bidirectional LSTM on multivariate time series (2021) — link
  31. BERT for Stock Market Sentiment Analysis (2019)
  32. Effect of Leader’s Voice on Financial Market: An Empirical Deep Learning Expedition on NASDAQ, NSE, and Beyond (2024)

1 of 40 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.