Off-the-Shelf Neural Network Architectures for Forex Time Series Prediction come at a Cost

Paper: arXiv 2405.10679

Authors: Theodoros Zafeiriou, Dimitris Kalles

Abstract

Our study focuses on comparing the performance and resource requirements between different Long Short-Term Memory (LSTM) neural network architectures and an ANN specialized architecture for forex market prediction. We analyze the execution time of the models as well as the resources consumed, such as memory and computational power. Our aim is to demonstrate that the specialized architecture not only achieves better results in forex market prediction but also executes using fewer resources and in a shorter time frame compared to LSTM architectures. This comparative analysis will provide significant insights into the suitability of these two types of architectures for time series prediction in the forex market environment.

Complexity vs Empirical Score

• Math Complexity: 2.0/10
• Empirical Rigor: 3.5/10
• Quadrant: Philosophers — conceptual discussion, limited math and data

Why this score: The paper provides a basic practical benchmark comparing standard LSTM architectures with a custom ANN using technical indicators on tick data. However, it lacks deep mathematical formulation, formal statistical validation, and detailed architectural specifics for the proprietary custom model, heavily relying on references to prior work.

Research Flowchart

  flowchart TD
    A[Goal: Compare Specialized ANN vs. LSTM<br>for Forex Time Series Prediction] --> B[Input Data: Forex Market Time Series]
    B --> C[Evaluate Architectures: Specialized ANN vs. LSTMs]
    C --> D[Measure Predictive Performance & Resource Usage<br>Execution Time, Memory, Compute]
    D --> E[Outcome: Specialized ANN achieves better predictions]
    D --> F[Outcome: Specialized ANN uses fewer resources & faster execution]

Cited References (12)

1. The use of technical analysis in the foreign exchange market (1992) — link
2. DeepAR: Probabilistic forecasting with autoregressive recurrent networks (2019) — link
3. Generative Adversarial Networks: An Overview (2018) — link
4. Benchmark Analysis of Representative Deep Neural Network Architectures (2018) — link
5. Predicting the Computational Cost of Deep Learning Models (2018) — link
6. Temporal Fusion Transformers for interpretable multi-horizon time series forecasting (2021) — link
7. Evolving a Deep Neural Network Training Time Estimator (2020) — link
8. Predicting Training Time Without Training (2020) — link

1 of 12 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.