1: \begin{abstract}

2:  This paper introduces \flexts,

3:  a new conditional density estimator for

4:  time series. \flexts \ is a flexible 

5:  nonparametric method,

6:  which can be based on an arbitrary

7:  regression method.

8:  It is shown that \flexts \ inherits

9:  the rate of convergence of 

10:  the chosen regression method.

11:  Hence, \flexts \ can adapt its

12:  convergence by employing 

13:  the regression method that 

14:  best fits the structure of data.

15:  From an empirical perspective,

16:  \flexts \ is compared to

17:  NNKCDE and GARCH in both 

18:  simulated and real data.

19:  \flexts\ is shown to generally obtain

20:  the best performance among

21:  the selected methods according to

22:  either the CDE loss or the pinball loss.

23:  \par\vskip\baselineskip\noindent

24:  \textbf{Keywords:} 

25:  nonparametric statistics,

26:  conditional density estimation,

27:  time series,

28:  machine learning \\ 

29:  \textbf{MSC:} 00-01; 99-00

30: \end{abstract}

31: