Traffic Volatility Forecasting Using an Omnibus Family GARCH Modeling Framework

Area of Science:

• Transportation Science
• Econometrics
• Time Series Analysis

Background:

• Traffic volatility modeling is crucial for accurate short-term traffic flow forecasting.
• Existing Generalized Autoregressive Conditional Heteroscedastic (GARCH) models may not fully capture traffic volatility's asymmetric properties due to parameter estimation constraints.
• A lack of comprehensive model comparison in traffic forecasting contexts creates challenges in selecting appropriate volatility models.

Purpose of the Study:

• To propose a unified framework for developing various traffic volatility forecasting models, accommodating both symmetric and asymmetric properties.
• To enable flexible estimation of key parameters (Box-Cox transformation coefficient λ, shift factor b, rotation factor c) for enhanced model development.
• To evaluate and compare the performance of different GARCH-family models within the proposed framework for traffic forecasting.

Main Methods:

• Developed an omnibus traffic volatility forecasting framework allowing unified modeling of symmetric and asymmetric properties.
• Incorporated flexible estimation of three key parameters: Box-Cox transformation coefficient (λ), shift factor (b), and rotation factor (c).
• Evaluated models using extensive traffic speed datasets from urban and freeway segments in China and the USA, employing metrics like MAE, MAPE, VMAE, DA, KP, and ACL.

Main Results:

• The proposed framework effectively accommodates various GARCH-family models, including standard GARCH, TGARCH, NGARCH, NAGARCH, GJR-GARCH, and FGARCH.
• Experimental results demonstrated the framework's flexibility and effectiveness in developing traffic volatility forecasting models.
• Performance evaluation provided insights into selecting appropriate models for different traffic forecasting scenarios.

Conclusions:

• The omnibus framework offers a versatile approach to traffic volatility modeling, addressing limitations of previous methods.
• The study highlights the importance of considering asymmetric properties in traffic volatility.
• The findings guide the development and selection of superior traffic volatility forecasting models for practical applications.