Statistics Seminar: Peng -- Data science honour presentation

The statistics seminar this week will be a session for Data Science honour
presentation.  The seminar/presentation will last 30 minutes.  

A comparison of hybrid ARIMA-GARCH and ARIMA-ANN on time series forecasting
Speaker: Yunlin Peng 
Location: AGR 829 
Zoom link: https://uni-sydney.zoom.us/j/82618524792

Due to a broader range of applications in time series applications in different domains,
especially in finance, accurate forecasting is crucial in decision-making.
Autoregressive Integrated Moving Average (ARIMA) model and Generalized Autoregressive
Conditional Heteroskedasticity (GARCH) model are popular statistical time series models
for modelling the linearity and volatility in time series. In recent years, Artificial
Neural Networks (ANNs) have been implemented in time series forecasting due to its
ability of capturing nonlinearity in data.  Despite regularly high accuracy of these
three models, complex real-world data is less likely to contain pure linear or
nonlinear patterns, which significantly restricts the performance of these models.
Hence, hybrid models combining linear and nonlinear models by utilizing each model’s
advantages have been investigated. In this project, performances of individual models,
ARIMA and ANN, and hybrid models, ARIMA-GARCH, additive ARIMA-ANN, and multiplicative
ARIMA-ANN, will be compared on various datasets with different characteristics.  The
empirical results indicate that hybrid models based on ARIMA-ANN outperform other
models, and the multiplicative ARIMA-ANN enhances the forecast accuracy on
mixed-linearity data the most.