Exploring the drivers and prediction of fund flow in the Norwegian Bond Fund Market: Deploying classic statistics and machine learning methods

Abstract

This paper aims to analyze the drivers and predict flow of Norwegian bond funds one month in advance using machine learning models and macro variables. The study investigates a sample of 79 bond funds and a subset of medium credit risk funds. We utilize XGBoost feature importance to investigate the factors that influence net flow of Norwegian bond funds. Additionally, through the use of OLS regression, we address our hypothesis concerning the significant impact of performance and macro variables on net flow of the Norwegian bond fund market. Our findings reveal only a significant negative linear relationship between change EUR/NOK and Norwegian bond funds. To harness the capabilities of machine learning models, partial dependency plots are also examined in search for non-linear relationships. XGBoost reveals non-linear relationship among predicted net flow and changes in VIX, the models (XGBoost and MLP) show varying impacts of change EUR/NOK, and contradiction patterns in lagged net flow. Due to poor accuracy scores across prediction models, we are unable to achieve effective models for predicting bond fund flow one month ahead using top selected features.