1: \begin{abstract} 

2: We present a local density estimator based on first order statistics. To estimate the density at a point, $x$,

3: the original sample is

4: divided into subsets and the average minimum sample distance to $x$ over all

5: such subsets is used to define the density estimate at $x$. The tuning 

6: parameter is thus the number of subsets instead of the typical bandwidth of kernel or 

7: histogram-based density estimators. 

8: The proposed method is similar to nearest-neighbor density estimators but it provides

9: smoother estimates.

10: We derive the asymptotic distribution of

11: this minimum sample distance statistic to study globally optimal

12: values for the number and size of the subsets.

13: Simulations are used to illustrate and compare the

14: convergence properties of the estimator.

15: The results show that

16: the method provides good estimates of a wide variety of

17: densities without changes of the tuning parameter, and that it

18: offers competitive convergence performance.

19: 

20: \end{abstract}

21: