Forecasting cryptocurrency volatility: a novel framework based on the evolving multiscale graph neural network
A novel EMGNN-based framework is proposed for cryptocurrency volatility forecasting, leveraging multiscale dynamics, graph structures, and robust evaluation criteria.
What it examines
The study introduces an EMGNN-based framework for forecasting cryptocurrency volatility by modeling evolving, multiscale interactions between crypto and traditional financial markets. It compares this graph neural network approach to traditional econometric, machine learning, and deep learning models, aiming to improve prediction accuracy and interpretability in volatile, dynamic financial environments.
What it concludes
The study concludes that the EMGNN-based framework outperforms alternative models in forecasting cryptocurrency volatility, yielding lower errors and improved economic performance. It effectively captures dynamic multiscale interactions and offers better interpretability, suggesting applications in investment risk management, market analysis, and policy-making.
Evidence objects
The pioneering evolving multiscale graph neural network (EMGNN) framework effectively integrates dynamic, scale-specific interactions between cryptocurrency markets and traditional financial assets, yielding lower $MSFE$, $MAFE$ errors and higher $CER$ returns.
key_findings bullet 1 · key_findings · validation V0
Surprisingly, recent findings reveal that cryptocurrencies are intricately connected with stocks, bonds, and macroeconomic uncertainties, especially during crises like COVID-19, thereby challenging conventional views and expanding insights into financial networks.
key_findings bullet 2 · key_findings · validation V0
Using bias--variance decomposition, sensitivity analysis, and multi-layer graph convolution techniques, EMGNN outperforms econometric, machine learning, and deep learning models, though it requires heavy computation and considers only high capitalization cryptocurrencies.
key_findings bullet 3 · key_findings · validation V0
An innovative framework employing an evolving multiscale graph neural network forecasts cryptocurrency volatility by modeling dynamic intersections between digital and traditional markets. Capturing complex multiscale interactions with interpretable graph structures, this approach offers fresh insights versus econometric and machine learning models, establishing originality, novelty, and remarkable impact for financial analysis.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
- … market forecasting mainly relied on traditional econometric models. These models have … models, including traditional econometric, classic ML, and well-known DL models. …
Source row: 893 · abstract type: snippet