Do exponentially distributed holding times imply Markovian?

42 Views Asked by At

I'm trying to understand the significance of the memoryless properties for Markov processes and for the exponential distribution. Both this answer and this answer have been very helpful. However, I am still not sure if this is correct:

For a stochastic process $(X(t))_{t\geq 0}$ with discrete state space, do exponentially distributed holding times $T_i \sim \exp(\lambda)$ imply that the process is a Markov process?

I would argue that the answer is yes as at any given time point $T$ the probability for no jump to occur until $T+s$ is

\begin{align} P(X_{T+s}=x_T|X_t,t\leq T) &= P(T_{i}>T+s|X_T=x_T) \\ &= P(T_{i}>T+s|T_{i}>T) \\ &= P(T_{i}>s) \\ &= e^{- \lambda s} \end{align}

(where $T_{i}$ is the holding time of the current state). So this probability only depends on the increment, which would mean not only is the process Markovian, but also homogeneous. Is that correct?