Discussing the article: "Neural Networks in Trading: Spatio-Temporal Neural Network (STNN)"

New comment
 

Check out the new article: Neural Networks in Trading: Spatio-Temporal Neural Network (STNN).

In this article we will talk about using space-time transformations to effectively predict upcoming price movement. To improve the numerical prediction accuracy in STNN, a continuous attention mechanism is proposed that allows the model to better consider important aspects of the data.

To address the complexities of multivariate data, the Spatio-Temporal Information (STI) Transformation equation was developed based on the delay embedding theorem. The STI equation transforms the spatial information of multivariate variables into the temporal dynamics of the target variable. This effectively increases the sample size and mitigates the challenges posed by short-term data.

Transformer-based models, already familiar in handling data sequences, use the Self-Attention mechanism to analyze relationships between variables while disregarding their relative distances. These attention mechanisms capture global information and focus on the most relevant features, alleviating the curse of dimensionality.

In the study "Spatiotemporal Transformer Neural Network for Time-Series Forecasting", a Spatiotemporal Transformer Neural Network (STNN) was proposed to enable efficient multi-step forecasting of multivariate short-term time series. This approach leverages the advantages of the STI equation and the Transformer framework.

The authors highlight several key benefits of their proposed methods:

  1. STNN uses the STI equation to convert the spatial information of multivariate variables into the temporal evolution of the target variable, effectively increasing the sample size.
  2. A continuous attention mechanism is proposed to improve the accuracy of numerical prediction.
  3. The spatial Self-Attention structure in the STNN collects efficient spatial information from multivariate variables, while the temporal Self-Attention structure collect information about the temporal evolution. The Transformer structure combines spatial and temporal information.
  4. The STNN model can reconstruct the phase space of a dynamical system for time series forecasting.

Author: Dmitriy Gizlyk

 
Quantitative returns are too small to compare to human trading
New comment