RESEARCH · GLOBAL
Temporal convolutional nets with profit-optimized thresholds forecast economically significant Bitcoin moves
Paper proposes multi-scale temporal convolutional networks (TCNs) tuned to detect Bitcoin price swings that clear transaction costs. Backtests show profit-optimized threshold selection beats fixed-level alerts.
WHY IT MATTERS
TCNs offer efficient alternative to LSTMs for crypto trading signals; profit-centered thresholds reduce false-positive alerts. Crypto desks and fintech custodians may adopt for volatility forecasting.
Source: arXiv q-fin · 2026-08-28