Existing microeconomic models for simulating poverty heavily rely on static projection from statistical inference. When used for simulation these models tend to conceive farm households as passive victims and thereby underestimate their resilience and adaptive capacity. Farming systems research has much to contribute to the research on poverty by bringing in a detailed understanding of farm household decision-making, which directly relates to their adaptive capacity. This paper presents a novel methodology to simulate poverty dynamics using a farming systems approach. The methodology is based on mathematical programming of farm households but adds three innovations: First, poverty levels are quantified by including a three-step budgeting system, including a savings model, a Working-Leser model, and an Almost Ideal Demand System. Second, the model is extended with a disinvestment model to simulate farm household coping strategies to food insecurity. Third, multi-agent systems are used to tailor each mathematical program to a real-world household and so to capture the heterogeneity of opportunities and constraints at the farm level as well as to quantify the distributional effects of change. An empirical application to Uganda illustrates the methodology. The method opens exciting new prospects for applying farming systems research and multi-agent systems to poverty analysis and the ex ante assessment of alternative policy interventions.