1: \begin{abstract}

2: Parametric estimation for diffusion processes is considered for high

3: frequency observations over a fixed time interval. The processes 

4: solve stochastic differential equations with an unknown parameter 

5: in the diffusion coefficient. We find easily verified conditions on approximate

6: martingale estimating functions under which estimators are consistent,

7: rate optimal, and efficient under high frequency (in-fill)

8: asymptotics. The asymptotic distributions of the estimators are shown to

9: be normal variance-mixtures, where the mixing distribution

10: generally depends on the full sample path of the diffusion process over 

11: the observation time interval. Utilising the concept of stable

12: convergence, we also obtain the more easily applicable

13: result that for a suitable data dependent normalisation, the estimators

14: converge in distribution to a standard normal distribution. The theory

15: is illustrated by a simulation study comparing an efficient and

16: a non-efficient estimating function for an ergodic and a non-ergodic model. \\ \\

17: {\bf Key words:} Approximate martingale estimating functions, 

18: discrete time sampling of diffusions, in-fill asymptotics, 

19: normal variance-mixtures, optimal rate, random Fisher information,

20: stable convergence, stochastic differential equation. \\ \\

21: {\bf Running title:} Efficient Estimation for High Frequency SDE Data. 

22: \end{abstract}

23: