1: \begin{abstract}

2: We study the maximum likelihood estimator of the drift parameters of a

3: stochastic differential equation, with both drift and diffusion coefficients

4: constant on the positive and negative axis, yet discontinuous at zero. 

5: This threshold diffusion is called drifted Oscillating Brownian motion.  

6: 

7:  For this continuously observed diffusion, the maximum likelihood estimator coincide with a quasi-likelihood estimator with constant diffusion term.

8: We show that this estimator is the limit, as observations become dense in time, of the (quasi)-maximum likelihood estimator based on discrete observations.

9: 

10: In long time, the asymptotic behaviors of

11: the positive and negative occupation times rule the ones of the estimators.

12: 

13: Differently from

14: most known results in the literature, we do not restrict ourselves to the

15: ergodic framework: indeed, depending on the signs of the drift, the process may

16: be ergodic, transient or null recurrent. For each regime, we establish whether

17: or not the estimators are consistent; if they are, we prove the convergence in

18: long time of the properly rescaled difference of the estimators towards a normal or

19: mixed normal distribution. These theoretical results are backed by numerical

20: simulations.

21: \end{abstract}

22: