1: \begin{abstract}

2:   Approximate Bayesian Computation (ABC) is a popular computational

3:   method for likelihood-free Bayesian inference.  The term

4:   ``likelihood-free'' refers to problems where the likelihood is

5:   intractable to compute or estimate directly, but where it is

6:   possible to generate simulated data~$X$ relatively easily given a

7:   candidate set of parameters~$\theta$ simulated from a prior

8:   distribution.  Parameters which generate simulated data within some

9:   tolerance~$\delta$ of the observed data~$x^\ast$ are regarded as

10:   plausible, and a collection of such $\theta$ is used to estimate the

11:   posterior distribution $\theta\,|\,X\!=\!x^\ast$.  Suitable choice

12:   of~$\delta$ is vital for ABC methods to return good approximations

13:   to~$\theta$ in reasonable computational time.

14: 

15:   While ABC methods are widely used in practice, particularly in

16:   population genetics, rigorous study of the mathematical properties

17:   of ABC estimators lags behind practical developments of the method.

18:   We prove that ABC estimates converge to the exact solution under

19:   very weak assumptions and, under slightly stronger assumptions,

20:   quantify the rate of this convergence.  In particular, we show that

21:   the bias of the ABC estimate is asymptotically proportional

22:   to~$\delta^2$ as $\delta \downto0$.  At the same time, the

23:   computational cost for generating one ABC sample increases like

24:   $\delta^{-q}$ where $q$ is the dimension of the observations.  Rates

25:   of convergence are obtained by optimally balancing the mean squared

26:   error against the computational cost.  Our results can be used to

27:   guide the choice of the tolerance parameter~$\delta$.

28: \end{abstract}

29: