Forecasting New Zealand Stock Returns Through Intermarket Analysis Using Global Vector Autoregressive Model and Machine Learning Methods

Abstract

This study applies intermarket analysis to forecast the New Zealand NZX 50 stock returns using a hybrid framework that combines econometric models with machine learning (ML) algorithms. Daily return data from 3 January 2001 to 31 December 2024 are employed to examine the interconnectedness between the New Zealand equity market and major global financial markets. This study follows a three-stage methodology: ML-based feature selection using LASSO, ridge, and elastic net regressions; econometric validation through a Global Vector Autoregressive (GVAR) model; and forecasting implementation using Support Vector Regression (SVR), Random Forest (RF), Long Short-Term Memory, and Artificial Neural Networks. Feature selection consistently identifies the Australian ASX 200, Japanese Nikkei 225, U.S. S&P 500, and U.S. 10-year Treasury yield as the most influential predictors, with Australia exerting the strongest impact. GVAR results reveal significant short-term spillover effects, but no long-term co-integrating relationships, indicating independent market trends. The U.S. market emerges as the dominant transmitter of shocks, while New Zealand acts as a net receiver. Forecasting results show RF and SVR outperform alternative models, with optimal performance achieved using a fifth-lag structure. The findings support short-term, spillover-based investment strategies and contribute to a transparent, replicable framework for forecasting equity markets in small open economies.