Real-time GDP nowcasting in New Zealand : an ensemble machine learning approach : a thesis presented for the degree of Master of Philosophy, School of Natural and Computational Sciences, Massey University, New Zealand
Year: 2019
Type: dissertation
Abstract: Gross Domestic Product (GDP) measures the monetary value of all final goods and services \nthat are produced in a region during a period of time. For most countries, GDP is released \na limited number of times a year and often with a lag. Understanding the current economic \nsituation, instead of figures quarters ago, is of vital importance for both policy and private \nentrepreneurs. It is crucial to create a live GDP predictor that could Nowcast current GDP \ngrowth rate in the period of government statistics release delay. \nThe Econometric approach for GDP Nowcasting has dominated the forecasting area for \nmany years. However, most of the traditional econometric models could only incorporate \na small handful of variables with a linear model structure, which could not meet the \nrequirement of the “big data” era for a better model prediction ability with a large amount \nof unbalanced variables. With the improvement of computation ability and the increment \nof high frequency variables, data-driven approaches like Machine Learning Methods have \nbeen applied into Nowcasting area. It does not only show a stronger forecasting ability in \nhandling large number of predictors but also present a superior robustness for non-linear \ndata structure. In this research, an Ensemble Method constructed by several Machine \nLearning Methods have been generated to provide more timely available GDP figures in \nthe period of government statistics release delay. \nHaving integrated an input dataset with data from multiple data sources such as public \nstatistical websites, Reserve Bank of NZ and Stats NZ, our cooperators New Zealand \nTransport Agency (NZTA) and PayMark, this study is conducted by first applying different \nMachine Learning methods such as Lightgbm, Xgboost, Support Vector Machine, K- \nNearest Neighbors, Ridge Regression, Lasso, Adaboost models. Then these algorithms \nare combined to generate an Ensemble Model with the assistance of an averaging method, \nwhich weights each model individually based on its historical prediction accuracy. The \nresult of the final Ensemble Model is compared with the most commonly used benchmark \nARIMA model and Random Walk model in terms of Mean Square Error (MSE) and \nMedian Absolute Error(MAE) value. Statistical tests, such as Friedman Test and Wilcoxon \nSigned-Rank Test, are employed to check the significance of model superiority. \nThe results indicate that the Ensemble Model significantly outperforms individual \nMachine Learning algorithm and Random Walk model in forecasting accuracy. When \ncompared with the ARIMA model, it shows slightly better prediction ability with more \nfore-sights especially in a fluctuating environment.
Author: Jinjin Fan
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W3102280279