In this section, we once more consider the stochastic recursion
\(W_{i+1} = \left[ V_{i} W_i + Y_i\right] ^+\), where
\(Y_i=B_i-A_i\). Again we impose the usual independence assumptions on the sequences
\(\{V_i\}_{i\in {\mathbb N}_0}\),
\(\{A_i\}_{i\in {\mathbb N}_0}\), and
\(\{B_i\}_{i\in {\mathbb N}_0}\). In addition, we assume that the
\(A_i\) are exp(
\(\lambda \)) distributed. The ‘multiplicative adjustments’
\(\{V_i\}_{i\in {\mathbb N}_0}\) are assumed to form a sequence of unit uniformly distributed random variables on [0, 1]. By Theorem
2, since
\({{\mathbb E}}(\log |V|)<0\), a steady-state distribution of
\(\{W_i\}_{i\in {\mathbb N}_0}\) always exists. We shall first study its transient distribution and then obtain the steady-state distribution.
Using the independence between
\(\{A_i\}_{i\in {\mathbb N}_0}\),
\(\{B_i\}_{i\in {\mathbb N}_0}\), and
\(\{V_i\}_{i\in {\mathbb N}_0}\) and the exponentiality of the
\(A_i\), we obtain
$$\begin{aligned} \Phi _{W_{i+1}}(s)&= \Phi _{V_iW_i}(s)\Phi _{B}(s)\frac{\lambda }{\lambda -s} + 1- {{\mathbb P}}(V_{i}W_{i}+B_i-A_i<0) \frac{\lambda }{\lambda -s}\nonumber \\&\quad -{{\mathbb P}}(V_{i}W_{i}+B_i-A_i\geqslant 0) \nonumber \\&= \Phi _{V_iW_i}(s)\Phi _{B}(s)\frac{\lambda }{\lambda -s} - p_{i+1} \frac{s}{\lambda -s}, \end{aligned}$$
(44)
where we set
\(p_i:= {{\mathbb P}}(W_{i}=0)\). Our goal is to write (
44) fully in terms of the functions
\(\Phi _{W_i}(s)\). To this end, performing the change of variable
\(v:=su\), we obtain
$$\begin{aligned} \Phi _{V_iW_i}(s) = \int _0^1 {{\mathbb E}}(\mathrm{e}^{-suW_i})\,\mathrm{d}u=\frac{1}{s} \int _0^s \Phi _{W_i}(v)\,\mathrm{d}v. \end{aligned}$$
(45)
By multiplying with
\(\lambda -s\), we thus obtain the following recursive integral equation.
Since the LST
\(\Phi _{W_0}(s)=\mathrm{e}^{-sw}\) of
\(W_0\) is known, Relation (
46) in principle allows us to recursively determine all the transforms
\(\Phi _{W_i}(\cdot )\),
\(i\in {{\mathbb {N}}}\). Observe that, when
\(s=\lambda \), the right-hand side should become zero; using (
45) we obtain
$$\begin{aligned} p_{i+1} = \frac{\Phi _{B}(\lambda )}{\lambda } \int _0^{\lambda } \Phi _{W_i}(v)\,\mathrm{d}v. \end{aligned}$$
(47)
This formula can easily be interpreted probabilistically, using the memoryless property of the exponential distribution for
\(A_i\):
$$\begin{aligned} {{\mathbb P}}(W_{i+1}=0) = P(A_i \geqslant B_i + V_iW_i) = {{\mathbb P}}(A_i \geqslant B_i) P(A_i \geqslant V_iW_i) = \Phi _{B}(\lambda ) \Phi _{V_iW_i}(\lambda ). \end{aligned}$$
It is not possible to obtain explicit expressions for
\(\Phi _{W_i}\). However, as so often, one can utilize the method of generating functions to turn the recursion (
46) into some sort of differential or integral equation. Therefore, we multiply Eq. (
46) by
\(r^{i+1}\) and sum over
i to obtain
$$\begin{aligned} U_{W}(r,s)= \frac{\lambda r\,\Phi _{B}(s)}{s(\lambda -s)} \int _0^{s} U_{W}(r,v)\,\mathrm{d}v+\kappa (s), \end{aligned}$$
(48)
where, to simplify the notation,
$$\begin{aligned} \kappa (s)=\Phi _{W_0}(s)-\frac{s}{\lambda -s}\left( U_{W}(r,\infty )-p_0\right) . \end{aligned}$$
(49)
With
\(I(s)=\int _0^{s}U_{W}(r,v)\,\mathrm{d}v\), we obtain the linear first-order differential equation
$$\begin{aligned} I'(s) = \frac{\lambda r\Phi _{B}(s)}{s(\lambda -s)}I(s)+\kappa (s). \end{aligned}$$
(50)
As follows by standard techniques, this inhomogeneous differential equation is solved by
$$\begin{aligned} I(s)=\mathrm{exp}\left( \lambda r \int _c^s\frac{\Phi _{B}(t)}{t(\lambda -t)}\,\mathrm{d}t \right) \left( \theta (c)+\int _c^s \kappa (u)\, \mathrm{exp}\left( -\lambda r\int _c^u\frac{\Phi _{B}(t)}{t(\lambda -t)}\,\mathrm{d}t \right) \,\mathrm{d}u\right) , \end{aligned}$$
(51)
where necessarily
\(\theta (c)=I(c)\) and we assume for the time being that
\(c\in (s,\lambda )\) if
\(s<\lambda \) and
\(c\in (\lambda ,s)\) if
\(\lambda <s\). Since
\(\Phi _{B}\) is bounded and bounded away from zero, we have as
\(s\downarrow \lambda \) (and likewise if
\(s\uparrow \lambda \) in the case where
\(s<\lambda \)),
$$\begin{aligned} \mathrm{exp}\left( \lambda r \int _c^s\frac{\Phi _{B}(t)}{t(\lambda -t)}\,\mathrm{d}t \right) \rightarrow \infty . \end{aligned}$$
As a consequence, to make sure
\(U_{W}(r,\lambda )\) remains bounded,
$$\begin{aligned} \theta (c)=-\int _c^\lambda \kappa (u)\, \mathrm{exp}\left( -\lambda r\int _c^u\frac{\Phi _{B}(t)}{t(\lambda -t)}\,\mathrm{d}t \right) \,\mathrm{d}u, \end{aligned}$$
and therefore
$$\begin{aligned} I(s)&=\mathrm{exp}\left( \lambda r \int _c^s\frac{\Phi _{B}(t)}{t(\lambda -t)}\,\mathrm{d}t \right) \int _\lambda ^s \kappa (u)\, \mathrm{exp}\left( -\lambda r\int _c^u\frac{\Phi _{B}(t)}{t(\lambda -t)}\,\mathrm{d}t \right) \,\mathrm{d}u \nonumber \\&=\int _\lambda ^s \kappa (u)\, \mathrm{exp}\left( \lambda r\int _u^s\frac{\Phi _{B}(t)}{t(\lambda -t)}\,\mathrm{d}t \right) \,\mathrm{d}u. \end{aligned}$$
(52)
It is to be noted that the integrand in (
52) tends to
\(\infty \) as
\(s\rightarrow \lambda \), which follows from the finiteness of
\(I(\lambda )\). Inserting (
52) into (
50), and recalling that
\(U_{W}(r,s)=I'(s)\), yields the following expression for the generating function
\(U_{W}\):
$$\begin{aligned} U_{W}(r,s)=\kappa (s)+ \frac{\lambda r\Phi _{B}(s)}{s(\lambda -s)}\int _\lambda ^s \kappa (u)\, \mathrm{exp}\left( \lambda r\int _u^s\frac{\Phi _{B}(t)}{t(\lambda -t)}\,\mathrm{d}t \right) \,\mathrm{d}u. \end{aligned}$$
(53)
Plugging (
49) into (
53) we obtain
$$\begin{aligned} U_{W}(r,s)&=\kappa (s)+ \frac{\lambda r\Phi _{B}(s)}{s(\lambda -s)}\int _\lambda ^s \left( \Phi _{W_0}(u)-\frac{u\left( U_{W}(r,\infty )-p_0\right) }{\lambda -u}\right) \,\\&\quad \mathrm{exp}\left( \lambda r\int _u^s\frac{\Phi _{B}(t)}{t(\lambda -t)}\,\mathrm{d}t \right) \,\mathrm{d}u. \end{aligned}$$
It remains to determine
\(U_{W}(r,\infty )\). Keeping in mind that
\(U_{W}(r,s)-\kappa (s)\rightarrow 0\) by (
49) we obtain, after some rearrangements,
$$\begin{aligned} U_{W}(r,\infty )=p_0-\frac{\int _\lambda ^\infty \Phi _{W_0}(u) \mathrm{exp}\left( -\lambda r\int _u^\infty \frac{\Phi _{B}(t)}{t(t-\lambda )}\,\mathrm{d}t \right) \,\mathrm{d}u}{\int _\lambda ^\infty \frac{u}{u-\lambda }\, \mathrm{exp}\left( -\lambda r\int _u^\infty \frac{\Phi _{B}(t)}{t(t-\lambda )}\,\mathrm{d}t \right) \,\mathrm{d}u}. \end{aligned}$$
We summarize our findings in the following theorem.