Economic simulations with many heterogeneous agents repeatedly access and retain large volumes of state and time-series data. In the Oxford economic model maintained by the Institute for New Economic Thinking at the University of Oxford, complete histories can remain in active memory and be repeatedly reconstructed even when only selected periods are required. This creates avoidable data movement, growing memory requirements, and additional runtime.

Relevant concepts from agent-based modelling, memory organisation, data locality and memory-aware programming were reviewed. The Oxford model was analysed to identify memory structures and access patterns. A smaller Oxford-inspired prototype was then developed to test four strategies: selective historical access, bounded rolling histories, incremental archival, and compact numeric representations.

The experiments show that the largest gains come from reducing the amount of historical data involved in active computation. Bounded rolling histories produced the strongest overall improvement, reducing retained history by 98.86% and achieving a 9.33× speedup relative to the original baseline at 240 simulated months. Selective historical access also reduced unnecessary data handling, while incremental archival preserved complete historical output without restoring unbounded in-memory histories. Mixed precision provided an additional reduction in memory use while keeping numerical differences within the defined tolerances. For the client, the thesis provides a practical optimisation sequence for future work on the Oxford model: access only required history, retain only active dependencies, archive older output separately, and then compact the remaining data. As the experiments used a simplified prototype, the measured gains demonstrate potential rather than guaranteed production performance. Overall, the findings show that memory-aware programming can substantially improve the scalability of large economic agent-based simulations.