Online Program Home
My Program

Abstract Details

Activity Number: 519 - SPEED: Methodological Advances in Time Series: BandE Speed Session, Part 2
Type: Contributed
Date/Time: Wednesday, July 31, 2019 : 10:30 AM to 11:15 AM
Sponsor: Business and Economic Statistics Section
Abstract #307895
Title: Testing for Unit Roots Using Artificial Neural Networks
Author(s): Rukman Ekanayake* and V A Samaranayake
Companies: and Missouri University of Science and Technology
Keywords: ANN; Time Series; nonstationary; Dickey-Fuller Tests; Machine Learning
Abstract:

Since the seminal paper by David Dickey and Wayne Fuller in 1979, there has been a continued interest in developing tests to detect unit roots in the ARMA formulation of empirical time series. Both asymptotic distribution-based as well as bootstrap-based tests have been developed with each method exhibiting both strengths and weaknesses. The use of artificial neural networks (ANNs) for forecasting empirical time series has also grown over the last quarter century, but there has been no serious attempt to develop an ANN-based methodology for unit root testing. Results of an initial attempt, to establish the proof of concept that an ANNs can be trained to detect the presence of a unit root in time series, is presented in this paper. Comparison with the Augmented Dickey-Fuller (ADF) test via Monte-Carlo simulations show the ANN outperforming the ADF for all parameter combinations studied, except for some exceptions for small sample sizes. Overall, results show promise in the use of ANNs to test for unit roots, but several issues such as the control of Type I error, optimal number of input nodes, hidden nodes, and hidden layers, have to be resolved prior to recommending this methodology as a viable alternative to existing test for unit roots.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program