1: \begin{abstract}

2: \noindent In this paper, we consider subgeometric ergodicity of univariate

3: nonlinear autoregressions with autoregressive conditional heteroskedasticity

4: (ARCH). The notion of subgeometric ergodicity was introduced in the

5: Markov chain literature in 1980s and it means that the transition

6: probability measures converge to the stationary measure at a rate

7: slower than geometric; this rate is also closely related to the convergence

8: rate of $\beta$-mixing coefficients. While the existing literature

9: on subgeometrically ergodic autoregressions assumes a homoskedastic

10: error term, this paper provides an extension to the case of conditionally

11: heteroskedastic ARCH-type errors, considerably widening the scope

12: of potential applications. Specifically, we consider suitably defined

13: higher-order nonlinear autoregressions with possibly nonlinear ARCH

14: errors and show that they are, under appropriate conditions, subgeometrically

15: ergodic at a polynomial rate. An empirical example using energy sector

16: volatility index data illustrates the use of subgeometrically ergodic

17: AR\textendash ARCH models.

18: 

19: \bigskip{}

20: \bigskip{}

21: 

22: \noindent\textbf{JEL classification:} C22.

23: 

24: \bigskip{}

25: 

26: \noindent\textbf{MSC2020 classification}s: 60J05, 37A25.

27: 

28: \bigskip{}

29: 

30: \noindent\textbf{Keywords:} Nonlinear autoregressive model, autoregressive

31: conditional heteroskedasticity, ARCH, subgeometric ergodicity, polynomial

32: ergodicity, Markov chain, $\beta$-mixing.

33: \end{abstract}

34: