1: \begin{abstract}

2: In this article, the problem of semi-parametric inference on the

3: parameters of a multidimensional L\'evy process $ L_{t} $ with

4: independent components based on the low-frequency observations of the

5: corresponding time-changed L\'evy process $ L_{\mathcal{T}(t)}$,

6: where $ \mathcal{T} $ is a nonnegative, nondecreasing real-valued

7: process independent of $ L_{t}$, is studied. We show that this

8: problem is closely related to the problem of composite function

9: estimation that has recently gotten much attention in statistical

10: literature. Under suitable identifiability conditions, we propose a

11: consistent estimate for the L\'evy density of $ L_{t} $ and derive

12: the uniform as well as the pointwise convergence rates of the

13: estimate proposed. Moreover, we prove that the rates obtained are

14: optimal in a minimax sense over suitable classes of time-changed L\'evy

15: models. Finally, we present a simulation study showing the performance

16: of our estimation algorithm in the case of time-changed Normal Inverse

17: Gaussian (NIG) L\'evy processes.

18: \end{abstract}