Technical Indicator Networks (TINs): An Interpretable Neural Architecture Modernizing Classic al Technical Analysis for Adaptive Algorithmic Trading
This work addresses the need for more adaptive and interpretable AI systems in financial trading by bridging traditional technical analysis with neural networks.
The paper tackles the problem of modernizing classical technical indicators for algorithmic trading by showing they can be reconstructed as neural networks with fixed weights, and introduces Technical Indicator Networks (TINs) as a general architecture that upgrades these indicators to handle multi-dimensional inputs like price, volume, and sentiment.
This work proposes that a vast majority of classical technical indicators in financial analysis are, in essence, special cases of neural networks with fixed and interpretable weights. It is shown that nearly all such indicators, such as moving averages, momentum-based oscillators, volatility bands, and other commonly used technical constructs, can be reconstructed topologically as modular neural network components. Technical Indicator Networks (TINs) are introduced as a general neural architecture that replicates and structurally upgrades traditional indicators by supporting n-dimensional inputs such as price, volume, sentiment, and order book data. By encoding domain-specific knowledge into neural structures, TINs modernize the foundational logic of technical analysis and propel algorithmic trading into a new era, bridging the legacy of proven indicators with the potential of contemporary AI systems.