New Zealand Activity Index (NZAC)
Nowcasting · Principal component analysis · Statistical inference
A real-time economic monitoring index developed to address the critical lack of timely economic data during the early stages of the COVID-19 pandemic, when standard indicators like GDP suffered from significant reporting lags. The index is derived by extracting a latent signal of economic activity from eight diverse monthly indicators, including consumer spending, electricity demand, and traffic volumes. Originally a Treasury-led tool for monitoring real-time economic activity, the index is now an established monthly indicator published by Statistics New Zealand.
View the indicator at Stats NZ and read the technical docs at New Zealand Treasury.
National Infrastructure Pipeline
Natural language processing · Large language models · Retrieval-augmented generation
This project delivered an LLM-based classifier to assign detailed labels to thousands of infrastructure projects based on raw text descriptions in the National Infrastructure Pipeline. The approach enabled the New Zealand Infrastructure Commission to move from a rudimentary legacy classification of 16 infrastructure sectors (e.g., Transport, Water, Energy), to a hierarchical taxonomy of more than 300 different types of infrastructure asset (e.g., Transport - Rail - Metro line network). The model's classification accuracy was benchmarked against human experts, achieving >90% accuracy across the 300+ possible labels. The result is a standardised, high-resolution view of New Zealand's National Infrastructure Pipeline, enabling detailed spending analysis, workforce modelling, and better policy decisions.
View the National Infrastructure Pipeline and infrastructure workforce modelling on the New Zealand Infrastructure Commission's insights platform (sign-up required).
Public Sector Discount Rates
Applied economics · Optimisation and decision theory · Intertemporal choice
An analysis of alternative economic frameworks for setting public sector discount rates produced for the New Zealand Treasury. This work helped shift the national policy debate, contributing to a gradual reduction in the New Zealand Government's central discount rate from 8% in 2015 to 2% in 2026. This downward adjustment significantly increases the weight we place on costs and benefits that occur in the future, which effectively upweights the interests of future generations in decisions we make now — particularly for decisions with long-term consequences like infrastructure and climate investment. The research was published in the Australian Economic Review (2018) and as a Treasury Working Paper (2017).
Read about the analysis in the Australian Economic Review and from the New Zealand Treasury.
Evolutionary Dynamics of Crime and Law Enforcement
Evolutionary game theory · Dynamical systems · Optimal control
A mathematical framework to capture the interaction between crime and law enforcement in large populations. The research identifies critical bifurcations (tipping points) in policy, demonstrating how slowly weakening enforcement can lead to a sudden, self-reinforcing collapse of social norms and a discontinuous spike in crime which can't easily be reversed. Published on arXiv.
Read the full paper on arXiv.
Modelling Cognitive Processes in Risky Decision-making
Eye-tracking · Gaussian mixture models · Poisson regression · Behavioral economics
The aim of this research was to gain insight into how individuals choose between risky decisions by using high-frequency eye-tracking data. This involved building a 3D Gaussian clustering model to classify 'fixations' — brief instants where a subject's eyes pause on a piece of information — and used Poisson regression to model the sequence of transitions between each fixation. The research provided early empirical evidence against the 'Priority Heuristic' and established the technical methodology used in a 2016 Journal of Behavioral Decision Making publication.
Read the final paper in the Journal of Behavioral Decision Making.
polymorphic