Likelihood-free inference for time-series simulation models

Agent-based models often lack a tractable likelihood function, precluding classical likelihood-based statistical inference and parameter estimation. As a result, likelihood-free approaches have emerged in recent decades as a means to performing statistical inference for such models. An example is approximate Bayesian computation, in which the pertinence of parameter settings is assessed by some meaningful notion of distance between the simulated and observed data. Agent-based models present a particular challenge in this respect: they often generate time-series data, which can be high-dimensional and complex in structure. In this talk, we will discuss recently developed approaches to performing Bayesian inference for intractable time-series simulation models. We will first discuss the problem of likelihood-free inference for time-series simulation models, before discussing some existing approaches that are popular in the literature. We will then discuss recent work on the use of path signatures as a means to deriving approximate Bayesian posteriors for time-series simulators (preprint available at, and discuss the properties of path signatures that make them appealing tools in this context. We will then present experimental results on their use in approximate Bayesian computation and, if time allows, we will discuss ongoing work on parameter inference for economic agent-based models.