Long-range dependence in a Cox process directed by a Markov renewal process (Q933893)

Summary: A Cox process \(N_{\text{Cox}}\) directed by a stationary random measure \(\xi\) has second moment \(\operatorname{var} N_{\text{Cox}}(0,t]= E(\xi(0,t])+ \operatorname{var}\xi(0,t]\), where by stationarity \(E(\xi(0,t])= (\text{const.})t= E(N_{\text{Cox}}(0,t])\), so long-range dependence (LRD) properties of \(N_{\text{Cox}}\) coincide with LRD properties of the random measure \(\xi\). When \(\xi(A)= \int_A\nu_{J(u)}\,du\) is determined by a density that depends on rate parameters \(\nu_i\) \((i\in\mathbb X)\) and the current state \(J(\cdot)\) of an \(\mathbb X\)-valued stationary irreducible Markov renewal process (MRP) for some countable state space \(\mathbb X\) (so \(J(t)\) is a stationary semi-Markov process on \(\mathbb X\)), the random measure is LRD if and only if each (and then by irreducibility, every) generic return time \(Y_{jj}\) \((j\in \mathbb X)\) of the process for entries to state \(j\) has infinite second moment, for which a necessary and sufficient condition when \(\mathbb X\) is finite is that at least one generic holding time \(X_j\) in state \(j\), with distribution function (DF) \(H_j\), say, has infinite second moment (a simple example shows that this condition is not necessary when \(\mathbb X\) is countably infinite). Then, \(N_{\text{Cox}}\) has the same Hurst index as the MRP \(N_{\text{Cox}}\) that counts the jumps of \(J(\cdot)\), while as \(t\to\infty\), for finite \(\mathbb X\), \(\operatorname{var} N_{\text{Cox}}(0,t]\sim 2\lambda^2 \int_0^t {\mathcal G}(u)\,du\), \(\operatorname{var} N_{\text{Cox}}(0,t]\sim 2\int_0^t \sum_{i\in\mathbb X}(\nu_i-\bar\nu)^2 \varpi_i{\mathcal H}_i(t)\,du\), where \(\bar\nu= \sum_i\varpi_i\nu_i= E[\xi(0,1]]\), \(\varpi_j= \text{Pr}\{J(t)=j\}\), \(1/\lambda= \sum_j \check p_j\mu_j\), \(\mu_j= E(X_j)\), \(\{\check p_j\}\) is the stationary distribution for the embedded jump process of the MRP, \({\mathcal H}_j(t)= \mu_i^{-1} \int_0^\infty \min(u,t) [1-H_j(u)]\,du\), and \({\mathcal G}(t)\sim \int_0^t \min(u,t) [1-G_{jj}(u)]\,du/m_{jj}\sim \sum_i \varpi_i{\mathcal H}_i(t)\) where \(G_{jj}\) is the DF and \(m_{jj}\) the mean of the generic return time \(Y_{jj}\) of the MRP between successive entries to the state \(j\). These two variances are of similar order for \(t\to\infty\) only when each \({\mathcal H}_i(t)/{\mathcal G}(t)\) converges to some \([0,\infty]\)-valued constant, say, \(\gamma_i\), for \(t\to\infty\).