Original title: Forecasting Economically Significant Bitcoin Moves: A Multi-Scale TCN with Profit-Optimized Thresholds
Parsa Yousefnezhad, Gholamreza Mansourfar, Mohammadreza Feizi Derakhshi · July 2026
Bitcoin prices swing wildly, making it tough for investors to protect their money. Researchers wanted to know if they could predict whether Bitcoin's price will climb more than 5% over the next 7 days. They gathered market activity, public sentiment, and on-chain records tracking blockchain transactions from February 2018 to December 2025. Then, they built a specialized deep learning system. It scans information across different time windows, using layered filters to spot patterns over 1 to 4 days without getting overwhelmed. Instead of just aiming for typical accuracy, the system tunes its trading decisions specifically to maximize financial gain. The researchers discovered that their design earned a profit of 1.703 and beat 5 standard competing models, offering a practical way to guide investment choices.
Imagine you're deciding whether to bring an umbrella on a walk. You check the weather forecast once and commit. But the forecast is only right half the time, and you're more concerned about getting wet than being accurate. You could just guess, but you'd rather optimize for staying dry. In this paper, the model predicts Bitcoin's price moves, but instead of just being accurate, it's trained to maximize profit, like choosing to bring the umbrella only when it's likely to rain and the cost of getting wet is high.
Bitcoin's wild swings make investing risky. This model helps investors and financial institutions forecast big price moves, so they can hedge and optimize portfolios. Better predictions mean smarter decisions and potentially higher profits.
The abstract does not specify the dataset size, the exact features used, or the computational cost. It also does not address whether the model works in real-time or for other cryptocurrencies. The profit metric is not defined in detail, and the model's performance may not generalize beyond the tested period.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Forecasting Economically Significant Bitcoin Moves: A Multi-Scale TCN with Profit-Optimized Thresholds — Parsa Yousefnezhad, Gholamreza Mansourfar, Mohammadreza Feizi Derakhshi
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.