The Dutch mortgage market is highly competitive, with financial institutions constantly seeking ways to optimise their pricing strategies and market share. However, there is limited publicly available academic research on the prediction of company-level mortgage application volumes and the calculation of price elasticity, due in large part to the confidential nature of company-specific data. The main research question of this thesis is: which forecasting model offers the highest prediction accuracy for predicting company-level mortgage application volumes one week ahead? An important second goal was analysing the responsiveness of company-level market share to relative pricing changes compared to competitors across different categories of mortgages.

To address this, a novel two-part forecasting framework is introduced that separates the total market volume of mortgage applications from company-specific market share predictions. A newly proposed hybrid model combining predictions of ARIMAX, XGBoost, and Random Forest was found to outperform existing methods and the benchmark model, achieving the highest prediction accuracy with a Root Mean Squared Error (RMSE) of 1.9E+07 for total mortgage volume predictions. Market share predictions utilized Double Machine Learning (DML), which yielded an MSE of 1.9E-03 and effectively quantified price elasticity across grouped mortgage loans with similar characteristics. Results showed significant elasticity differences: non-ported mortgages and longer fixed-rate period loans were highly elastic, while ported loans remained inelastic. This suggests that the relative price position plays a crucial role in predicting market share, though external factors such as announced future interest rate
increases of competitors also exert substantial influence.

For company-level volume predictions, a weighted average approach using market share and total mortgage application volume predictions was identified as the optimal model with a Ratio squared, defined as the squared division of the RMSE by the mean of the company volumes, value of 0.59 across all segments and a value of 0.05 for the aggregated company volume. This highlights the large variability for different types of mortgages. The model significantly outperformed the benchmark values and an ARIMAX model directly trained to predict the company, further underlining the importance of splitting the forecasting task into a two-part framework that integrates market-wide and company-specific dynamics.

The results confirm the importance of using multiple models to separate volume and market share predictions, offering financial institutions enhanced strategic decision-making capabilities. In conclusion, this thesis makes a novel contribution to the mortgage industry by providing a company-specific forecasting model and insights into price elasticity across mortgage types.