UNIVARIATE AND MULTIVARIATE TIME SERIES REGRESSION MODELING USING AUTOREGRESSIVE DISTRIBUTED LAG AND VECTOR AUTOREGRESSIVE WITH EXOGENOUS VARIABLE

• Main Author: Sherin, Anastasia
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/47746
• Institution: Institut Teknologi Bandung

In our daily life, many events have a correlation with themselves in the past. More complex, some events have a causal relationship, therefore an event is not only influenced by itself in the past, but is also influenced by other events. This concept is known as time series regression. There are two time series regression models to be used, namely the Autoregressive Distributed Lag (ADL) for univariate response and the Vector Autoregressive with Exogenous Variable (VARX) for multivariate responses. On the other hand, the need for data based decision making is increasing. Data needs to be processed quickly using the right model. Therefore, this study aims to find the characteristics of data that fit to be modeled with the ADL and VARX, to help people who will process time series data, so they can determine the right model quickly. The goodness of model is measured by the model’s predictive ability to determine the predicted value of several lag times ahead. In this study, there are comparisons the results of time series data predictions with multivariate responses using the ADL and VARX model. To answer this curiosity, five case studies were conducted. The results of the case study show that the ADL model is compatible with stationary level and first order data, fluctuating data, and data with outliers at one or two points. But the ADL model is not having a good capabilities if outliers on the data are scattered at various points. Meanwhile, the VARX model matches stationary level and first order stationary data, data that fluctuates and has an upward trend. Finally, it was concluded that for multivariate responses data, if the trends in the data are not drastic, then the VARX and ADL model have the same predictive ability. But for data with a drastic upward trend, the VARX model has better predictive capabilities. Therefore, for time series data with multivariate responses, it will be more effective if directly modeling with the VARX model.