A Proof of Lemma 1
The proof is similar to that of the Eq. (5) in Lim and Shin [12] and that of the Eq. (17) in He and Pagès [8]:
For any fixed \(T > 0\), we can define an equivalent martingale probability measure \(\widetilde{{\mathbb {P}}}^T\) as
$$\begin{aligned} \widetilde{\mathbb {P}}^T(A) = {\mathbb {E}}[e^{-\frac{1}{2}\theta ^2T-\theta B_T}{\mathbf {1}}_A], ~~\text{ for } \text{ any }~~ A \in {\mathcal {F}}_T. \end{aligned}$$
(45)
Then, by Girsanov theorem, the process \({\widetilde{B}}^T_t \triangleq B_t + \theta t\) is a standard Brownian motion under \(\widetilde{\mathbb {P}}^T\) for \(t \in [0, T]\). By Itô’s formula, we have
$$\begin{aligned} d(e^{-rt}X_t) = e^{-rt}\left( -c_t+Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}+Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) dt+e^{-rt}\pi _t\sigma d{\widetilde{B}}^T_t. \end{aligned}$$
(46)
Since \(\varvec{\delta }\) is of bounded variation and \(e^{-rt}X_t\) evolving according to (46) has continuous sample paths, we can apply integration by parts to get
$$\begin{aligned}&e^{-rt}X_t\delta _t +\int _0^te^{-rs}\left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) \delta _sds - \int _0^te^{-rs}X_s\,d\delta _s\\&\quad = x+\int _0^te^{-rs}\pi _s\sigma \delta _s d{\widetilde{B}}^T_s. \end{aligned}$$
Since \(X_t \ge 0\) a.s. for \(t \ge 0\) and \(\varvec{\delta }\) is a decreasing process, \(\int _0^te^{-rs}X_s\,d\delta _s \le 0\). Therefore, the left-hand side of the above equation is a continuous local martingale bounded from below and hence a supermartingale under \(\widetilde{\mathbb {P}}^T\) for \(t \in [0, T]\) so that
$$\begin{aligned} \widetilde{\mathbb {E}}^T\left[ e^{-rT}X_T\delta _T+\int _0^Te^{-rt}\left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _t dt - \int _0^Te^{-rt}X_t\,d\delta _t\right] \le x, \end{aligned}$$
where \(\widetilde{{\mathbb {E}}}^T\) denotes the expectation operator under the probability measure \(\widetilde{{\mathbb {P}}}^T\). Since \(\int _0^te^{-rs}X_s\,d\delta _s \le 0\), we have
$$\begin{aligned} \widetilde{{\mathbb {E}}}^T\left[ e^{-rT}X_T\delta _T+\int _0^Te^{-rt}\left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _t dt \right] \le x. \end{aligned}$$
(47)
Note that
$$\begin{aligned}&\widetilde{{\mathbb {E}}}^T\left[ e^{-rT}X_T\delta _T+\int _0^Te^{-rt}\left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _t dt \right] \nonumber \\&\quad ={\mathbb {E}}\left[ e^{-\frac{1}{2}\theta ^2T-\theta B_T}e^{-rT}X_T\delta _T+e^{-\frac{1}{2}\theta ^2T-\theta B_T}\int _0^Te^{-rt}\left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _tdt\right] \nonumber \\&\quad ={\mathbb {E}}\left[ H_TX_T\delta _T+\int _0^Te^{-rt}\left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _te^{-\frac{1}{2}\theta ^2T-\theta B_T}dt\right] \nonumber \\&\quad ={\mathbb {E}}[H_TX_T\delta _T]+\int _0^T{\mathbb {E}}\left[ e^{-rt}\left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _te^{-\frac{1}{2}\theta ^2T-\theta B_T}\right] dt\nonumber \\&\quad ={\mathbb {E}}[H_TX_T\delta _T]+\int _0^T{\mathbb {E}}\left[ {\mathbb {E}}\left[ \left. e^{-rt}\left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _te^{-\frac{1}{2}\theta ^2T-\theta B_T}\right| {\mathcal {F}}_t\right] \right] dt\nonumber \\&\quad ={\mathbb {E}}[H_TX_T\delta _T]+\int _0^T{\mathbb {E}}\left[ \left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _tH_t{\mathbb {E}}\left[ \left. e^{-\frac{1}{2}\theta ^2(T-t)-\theta (B_T-B_t)}\right| {\mathcal {F}}_t\right] \right] dt\nonumber \\&\quad ={\mathbb {E}}[H_TX_T\delta _T]+\int _0^T{\mathbb {E}}\left[ \left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _tH_t{\mathbb {E}}\left[ e^{-\frac{1}{2}\theta ^2(T-t)-\theta (B_T-B_t)}\right] \right] dt\nonumber \\&\quad ={\mathbb {E}}[H_TX_T\delta _T]+\int _0^T{\mathbb {E}}\left[ \left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _tH_t\right] dt\nonumber \\&\quad ={\mathbb {E}}\left[ H_TX_T\delta _T+\int _0^T\left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _tH_tdt\right] , \end{aligned}$$
(48)
where the first equality comes from (45), the second from (8), the third from Fubini theorem, the forth from the tower property, the fifth from the fact that \(\left( c_t-Y_0{\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _tH_t\) is \({\mathcal {F}}_t\)-measurable, the sixth from the independent increment of a Brownian motion, the seventh from the fact that \({\mathbb {E}}\left[ e^{-\frac{1}{2}\theta ^2(T-t)-\theta (B_T-B_t)}\right] = 1\), the last again from Fubini theorem. Combining (47) and (48), we have
$$\begin{aligned} {\mathbb {E}}\left[ H_TX_T\delta _T+\int _0^T\left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _tH_tdt\right] \le x. \end{aligned}$$
Since \(H_TX_T\delta _T \ge 0\) by (2) and (8), we have
$$\begin{aligned} {\mathbb {E}}\left[ \int _0^T\left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _tH_tdt\right] \le x. \end{aligned}$$
(49)
Since (49) holds for any \(T \ge 0\), we get (11). \(\square \)
B Proof of Lemma 2
The proof is similar to that of Lemma 1 in He and Pagès [8]. Let
$$\begin{aligned} \varGamma _t \triangleq \sup _{T \in [t, \infty )}{\mathbb {E}}\left[ \left. \int _0^T\left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds\right| {\mathcal {F}}_t\right] \quad \text{ for }~~t\ge 0. \end{aligned}$$
(50)
Then,
$$\begin{aligned} -\left( Y_0+Y_1\right) {\mathbb {E}}\left[ \left. \int _0^\infty H_sds\right| {\mathcal {F}}_t\right] \le \varGamma _t \le {\mathbb {E}}\left[ \left. \int _0^{\infty }c_sH_sds\right| {\mathcal {F}}_t\right] . \end{aligned}$$
Taking the expectation on each term of the above inequalities, we have, by the tower property,
$$\begin{aligned} -\left( Y_0+Y_1\right) {\mathbb {E}}\left[ \int _0^\infty H_sds\right] \le {\mathbb {E}}\left[ \varGamma _t\right] \le {\mathbb {E}}\left[ \int _0^{\infty }c_sH_sds\right] . \end{aligned}$$
(51)
However,
$$\begin{aligned} -\left( Y_0+Y_1\right) {\mathbb {E}}\left[ \int _0^\infty H_sds\right] = -\left( Y_0+Y_1\right) \int _0^\infty {\mathbb {E}}\left[ H_s\right] ds = -\frac{Y_0+Y_1}{r}, \end{aligned}$$
(52)
and
$$\begin{aligned} {\mathbb {E}}\left[ \int _0^{\infty }c_sH_sds\right]= & {} {\mathbb {E}}\left[ \int _0^\infty \left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds\right] \nonumber \\&+ \,{\mathbb {E}}\left[ \int _0^\infty \left( Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}+Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds\right] \nonumber \\\le & {} {\mathbb {E}}\left[ \int _0^\infty \left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds\right] \nonumber \\&+\left( Y_0+Y_1\right) {\mathbb {E}}\left[ \int _0^\infty H_sds\right] \nonumber \\\le & {} x + \frac{Y_0+Y_1}{r}, \end{aligned}$$
(53)
where the last inequality (53) comes from (13) and (52). By (51), (52), and (53), \(\varGamma _t\) is integrable for every \(t \ge 0\), and
$$\begin{aligned} \varGamma _0 = \sup _{T \in [0, \infty )}{\mathbb {E}}\left[ \int _0^T\left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds\right] \le x, \end{aligned}$$
(54)
where the inequality holds since \((\varvec{\varTheta }, {\mathbf {c}})\) satisfies the budget constraint (12). Note that
$$\begin{aligned} \varGamma _t \ge \int _{0}^{t}\left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds. \end{aligned}$$
(55)
Consider any fixed t and \(t^{\prime }\) with \(t < t^{\prime }\). For any given \(\epsilon > 0\), there exists \(T^{\prime } \ge t^{\prime }\) such that
$$\begin{aligned} \varGamma _{t^{\prime }} - {\mathbb {E}}\left[ \left. \int _{0}^{T^{\prime }}\left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds\right| {\mathcal {F}}_{t^{\prime }}\right] < \epsilon . \end{aligned}$$
(56)
Thus, we have
$$\begin{aligned} {\mathbb {E}}[\varGamma _{t^{\prime }} | {\mathcal {F}}_t]= & {} {\mathbb {E}}\left[ \left. \varGamma _{t^{\prime }} - {\mathbb {E}}\left[ \left. \int _{0}^{T^{\prime }}\left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds\right| {\mathcal {F}}_{t^{\prime }}\right] \right| {\mathcal {F}}_t\right] \nonumber \\&+\, {\mathbb {E}}\left[ \left. {\mathbb {E}}\left[ \left. \int _{0}^{T^{\prime }}\left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds\right| {\mathcal {F}}_{t^{\prime }}\right] \right| {\mathcal {F}}_t\right] \end{aligned}$$
(57)
$$\begin{aligned}\le & {} \epsilon + {\mathbb {E}}\left[ \left. \int _{0}^{T^{\prime }}\left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds\right| {\mathcal {F}}_t\right] \nonumber \\\le & {} \epsilon + \varGamma _t, \end{aligned}$$
(58)
where the first inequality holds by (56) and the tower property, the second since \(T^{\prime } \ge t^{\prime } > t\). Since the inequality (58) holds holds for arbitrary \(\epsilon > 0\), we have \({\mathbb {E}}[\varGamma _{t^{\prime }} | {\mathcal {F}}_t] \le \varGamma _t\). That is, \((\varGamma _t)_{t\ge 0}\) is a supermartingale under \({\mathbb {P}}\) so that the Doob-Meyer Decomposition Theorem allows us to write
$$\begin{aligned} \varGamma _t = \varGamma _0 + M_t - K_t, \end{aligned}$$
(59)
where \((M_t)_{t\ge 0}\) is a martingale under \({\mathbb {P}}\) with \(M_0 = 0\) and \((K_t)_{t\ge 0}\) is an increasing process with \(K_0 = 0\). By the martingale representation theorem, we can find an \({\mathcal {F}}_t\)-adapted process \((\psi _t)_{t\ge 0}\) with \(\int _0^t\psi ^2_tds<\infty ,~\text{ a.s. }\), such that
$$\begin{aligned} M_t = \int _0^t\psi _sdB_s,~\quad \text{ for }~ t\ge 0. \end{aligned}$$
(60)
Consider the portfolio process \(\varvec{\pi }\) given by
$$\begin{aligned} \pi _t = \frac{\psi _t}{\sigma H_t} + \frac{\theta }{\sigma }X_t,\quad t \ge 0, \end{aligned}$$
where \(X_t\) is the agent’s wealth at time t. With the policy \((\varvec{\varTheta }, {\mathbf {c}}, \varvec{\pi })\), the stochastic differential equation (10) becomes
$$\begin{aligned} d(H_tX_t) = \left[ -c_t+Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}+Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right] H_tdt+\psi _tdB_t. \end{aligned}$$
Thus,
$$\begin{aligned} H_tX_t= & {} x - \int _{0}^{t}\left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds + \int _0^t\psi _sdB_s\\= & {} x - \int _{0}^{t}\left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds + M_t\\= & {} x -\varGamma _0 + \varGamma _t - \int _{0}^{t}\left( c_s-Y_0 {\mathbf {1}}_{\{\varTheta _s=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _s=A_1\}}\right) H_sds + K_t\\\ge & {} 0, \end{aligned}$$
where the second equality comes from (60), the third from (59), and the inequality from (54), (55), and the fact that \(K_t \ge 0\). Thus, \(X_t\) satisfies (2). \(\Box \)
C Proof of Lemma 3
For any triple \((\varvec{\varTheta }, {\mathbf {c}}, \varvec{\pi })\in {\mathcal {A}}(x)\), \(\varvec{\delta }\in {\mathcal {D}}\), and \(\lambda > 0\), we have
$$\begin{aligned} J(x;\varvec{\varTheta }, \mathbf {c}, \varvec{\pi })&=\mathbb {E}\left[ \int _0^\infty e^{-\rho t}\left( L_0^{\gamma _1-\gamma }\frac{c_t^{1-\gamma _1}}{1-\gamma _1}\mathbf {1}_{ \{\varTheta _t=A_0\}}+L_1^{\gamma _1-\gamma }\frac{c_t^{1-\gamma _1}}{1-\gamma _1}\mathbf {1}_{\{ \varTheta _t=A_1\}}\right) dt\right] \nonumber \\&\le \mathbb {E}\left[ \int _0^\infty e^{-\rho t}\left\{ \left( \widetilde{u}_0(z_t)+Y_0z_t\right) \mathbf {1}_{\{\varTheta _t=A_0\}}+ \left( \widetilde{u}_1(z_t)+Y_1z_t\right) \mathbf {1}_{\{ \varTheta _t=A_1\}}\right\} dt\right] \nonumber \\&\quad + \lambda \mathbb {E}\left[ \int _0^\infty \left( c_t-Y_0 \mathbf {1}_{\{\varTheta _t=A_0\}}-Y_1 \mathbf {1}_{\{\varTheta _t=A_1\}}\right) \delta _tH_tdt\right] \end{aligned}$$
(61)
$$\begin{aligned}&\le \mathbb {E}\left[ \int _0^\infty e^{-\rho t}\left\{ \left( \widetilde{u}_0(z_t)+Y_0z_t\right) \mathbf {1}_{\{ \varTheta _t=A_0\}}+\left( \widetilde{u}_1(z_t)+Y_1z_t\right) \mathbf {1}_{\{ \varTheta _t=A_1\}}\right\} dt\right] \nonumber \\&\quad + \lambda x \end{aligned}$$
(62)
$$\begin{aligned}&\le \mathbb {E}\left[ \int _0^\infty e^{-\rho t}\left\{ \left( \widetilde{u}_0(z_t)+Y_0z_t\right) \mathbf {1}_{\{0<z_t \le \bar{z}\}}+\left( \widetilde{u}_1(z_t)+Y_1z_t\right) \mathbf {1}_{\{z_t> \bar{z}\}}\right\} dt\right] \nonumber \\&\quad +\lambda x \nonumber \\&= \varPhi (\varvec{\delta }, \lambda ), \end{aligned}$$
(63)
where the first inequality comes from (5), the second inequality from Remark 3, and the last inequality from Remark 2. Thus, the inequality (16) holds for any triple \((\varvec{\varTheta }, {\mathbf {c}}, \varvec{\pi })\in {\mathcal {A}}(x)\), \(\varvec{\delta }\in {\mathcal {D}}\), and \(\lambda > 0\). By (6), the inequality (61) holds with equality if and only if
$$\begin{aligned} c_t=\left\{ \begin{array}{ll} f_0(z_t) = L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}z_t^{-\frac{1}{\gamma _1}}, &{}\quad \text{ for } \varTheta _t=A_0,\\ f_1(z_t) = L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}z_t^{-\frac{1}{\gamma _1}}, &{}\quad \text{ for } \varTheta _t=A_1. \end{array}\right. \end{aligned}$$
The inequality (62) holds with equality if and only if
$$\begin{aligned} {\mathbb {E}}\left[ \int _0^\infty \left( c_t-Y_0 {\mathbf {1}}_{\{\varTheta _t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta _t=A_1\}}\right) \delta _tH_tdt\right] = x. \end{aligned}$$
The inequality (63) holds with equality if and only if
$$\begin{aligned} \varTheta _t=\left\{ \begin{array}{ll} A_0, &{}\quad \text{ if } 0<z_t\le {\bar{z}},\\ A_1, &{}\quad \text{ if } z_t>{\bar{z}}. \end{array}\right. \end{aligned}$$
Thus, we have shown the second statement of the lemma. For the proof of the last statement, note that, for any triple \((\varvec{\varTheta }, {\mathbf {c}}, \varvec{\pi })\in {\mathcal {A}}(x)\), we have
$$\begin{aligned} J(x;\varvec{\varTheta }, {\mathbf {c}}, \varvec{\pi }) \le \varPhi (\varvec{\delta }^*, \lambda ^*) = J(x;\varvec{\varTheta }^*, {\mathbf {c}}^*, \varvec{\pi }^*), \end{aligned}$$
where the inequality holds since \((\varvec{\varTheta }, {\mathbf {c}}, \varvec{\pi })\in {\mathcal {A}}(x)\), \(\varvec{\delta }\in {\mathcal {D}}\), and \(\lambda > 0\) are arbitrary in the inequality(16). Thus, the last statement of the lemma holds. \(\Box \)
D Proof of Theorem 1
The proof is similar to that of Lemma 1 in He and Pagès [8]: Fix any \(T > 0\) and let \(\varvec{\delta }^{\epsilon } \triangleq (\delta ^{\epsilon }_t)_{t\ge 0}\) with \(\delta ^{\epsilon }_t = (\lambda ^*\delta ^*_t + \epsilon {\mathbf {1}}_{[0,T)}(t))/(\lambda ^* + \epsilon )\) for \(t\ge 0\), where \(\epsilon > 0\). Then, \(\varvec{\delta }^{\epsilon } \in {\mathcal {D}}\). Let \(\lambda ^{\epsilon } \triangleq \lambda ^* + \epsilon \) and let \(z^{\epsilon }_t \triangleq z_t^{\varvec{\delta }^{\epsilon }, \lambda ^{\epsilon }}\) for \(t\ge 0\). Then, \(z^{\epsilon }_t = z^*_t + \epsilon e^{\rho t}H_t \) for \(0 \le t < T\) and \(z^{\epsilon }_t = z^*_t\) for \(t \ge T\). Define
$$\begin{aligned} {\mathcal {N}}^{\epsilon } \triangleq {\mathbb {E}}\left[ \int _0^\infty e^{-\rho t}\left\{ \left( {\widetilde{u}}_0(z^{*}_t)+Y_0z^{*}_t\right) { \mathbf {1}}_{\{0<z^{\epsilon }_t\le {\bar{z}}\}}+\left( { \widetilde{u}}_1(z^{*}_t)+Y_1z^{*}_t\right) {\mathbf {1}}_{\{z^{\epsilon }_t>{ \bar{z}}\}}\right\} dt\right] +\lambda ^{*}x. \end{aligned}$$
Since \({\mathcal {N}}^{\epsilon } \le \varPhi (\varvec{\delta }^{*}, \lambda ^*)\) by Remark 2 and \(\varPhi (\varvec{\delta }^{*}, \lambda ^*) \le \varPhi (\varvec{\delta }^{\epsilon }, \lambda ^{\epsilon })\), we have
$$\begin{aligned} \frac{\varPhi (\varvec{\delta }^{\epsilon }, \lambda ^{\epsilon }) - {\mathcal {N}}^{\epsilon }}{\epsilon } \ge 0. \end{aligned}$$
That is,
$$\begin{aligned} 0\le & {} {\mathbb {E}}\Bigg [\int _0^T e^{-\rho t}\Bigg \{\Big (\frac{{\widetilde{u}}_0(z^*_t+\epsilon e^{\rho t}H_t)-{\widetilde{u}}_0(z^*_t)}{\epsilon }+Y_0e^{\rho t}H_t\Big ){\mathbf {1}}_{\{0<z^{*}_t+\epsilon e^{\rho t}H_t\le {\bar{z}}\}}\nonumber \\&+\,\Big (\frac{{\widetilde{u}}_1(z^*_t+\epsilon e^{\rho t}H_t)-{\widetilde{u}}_1(z^*_t)}{\epsilon }+Y_1e^{\rho t}H_t\Big ){\mathbf {1}}_{\{z^{*}_t+\epsilon e^{\rho t}H_t>{\bar{z}}\}}\Bigg \}dt\Bigg ]+ x. \end{aligned}$$
(64)
Note that
$$\begin{aligned} 0 \le \int _0^T Y_0H_t{\mathbf {1}}_{\{0<z^{*}_t+\epsilon e^{\rho t}H_t\le {\bar{z}}\}}\,dt \le Y_0\int _0^T H_t\,dt. \end{aligned}$$
By using Fubini Theorem, we get
$$\begin{aligned} {\mathbb {E}}\Bigg [Y_0\int _0^T H_t\,dt\Bigg ] = Y_0\int _0^T {\mathbb {E}}[H_t]\,dt = Y_0\int _0^T e^{-rt}\,dt = \frac{Y_0}{r}(1-e^{-rT}) < \infty . \end{aligned}$$
By the dominated convergence theorem, we have
$$\begin{aligned} \lim _{\epsilon \downarrow 0}{\mathbb {E}}\Bigg [\int _0^T Y_0H_t{\mathbf {1}}_{\{0<z^{*}_t+\epsilon e^{\rho t}H_t\le {\bar{z}}\}}\,dt\Bigg ] = {\mathbb {E}}\Bigg [\lim _{\epsilon \downarrow 0}\int _0^T Y_0H_t{\mathbf {1}}_{\{0<z^{*}_t+\epsilon e^{\rho t}H_t\le {\bar{z}}\}}\,dt\Bigg ]. \end{aligned}$$
Since \(Y_0H_t{\mathbf {1}}_{\{0<z^{*}_t+\epsilon e^{\rho t}H_t\le {\bar{z}}\}} \le Y_0H_t\) and \(\int _0^T Y_0H_t\,dt < \infty \) a. s., the limit and the integration in the second term of the above equation can be interchanged by the dominated convergence theorem so that we have
$$\begin{aligned} \lim _{\epsilon \downarrow 0}{\mathbb {E}}\Bigg [\int _0^T Y_0H_t{\mathbf {1}}_{\{0<z^{*}_t+\epsilon e^{\rho t}H_t\le {\bar{z}}\}}\,dt\Bigg ] = {\mathbb {E}}\Bigg [\int _0^T Y_0H_t{\mathbf {1}}_{\{0<z^{*}_t\le {\bar{z}}\}}\,dt\Bigg ]. \end{aligned}$$
(65)
Similarly, we have
$$\begin{aligned} \lim _{\epsilon \downarrow 0}{\mathbb {E}}\Bigg [\int _0^T Y_1H_t{\mathbf {1}}_{\{z^{*}_t+\epsilon e^{\rho t}H_t> {\bar{z}}\}}\,dt\Bigg ] = {\mathbb {E}}\Bigg [\int _0^T Y_1H_t{\mathbf {1}}_{\{z^{*}_t > {\bar{z}}\}}\,dt\Bigg ]. \end{aligned}$$
(66)
By using (65) and (66), and by applying Fatou’s Lemma which is possible since \({\widetilde{u}}_i(z^*_t+\epsilon e^{\rho t}H_t)-{\widetilde{u}}_i(z^*_t) \le 0\) for \(i = 0, 1\), in (64), we get
$$\begin{aligned} 0\le & {} {\mathbb {E}}\Bigg [\int _0^T e^{-\rho t}\Bigg \{\Big (e^{\rho t}H_t{\widetilde{u}}_0^{\prime }(z^*_t)+Y_0e^{\rho t}H_t\Big ){\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}\\&+\Big (e^{\rho t}H_t{\widetilde{u}}_1^{\prime }(z^*_t)+Y_1e^{\rho t}H_t\Big ){\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\Bigg \}dt\Bigg ]+ x\\= & {} -{\mathbb {E}}\left[ \int _0^T\left( c^*_t-Y_0 {\mathbf {1}}_{\{\varTheta ^*_t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta ^*_t=A_1\}}\right) H_tdt\right] + x, \end{aligned}$$
where the equality comes from (7) and (19). Therefore, we have
$$\begin{aligned} {\mathbb {E}}\left[ \int _0^T\left( c^*_t-Y_0 {\mathbf {1}}_{\{\varTheta ^*_t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta ^*_t=A_1\}}\right) H_tdt\right] \le x. \end{aligned}$$
(67)
Since this holds for any \(T \ge 0\), the job and consumption policies \((\varvec{\varTheta }^*, {\mathbf {c}}^*)\) satisfy the budget constraint (12). Hence, by Lemma 2, there exists a portfolio process \(\varvec{\pi }^*\) such that
$$\begin{aligned} (\varvec{\varTheta }^*, {\mathbf {c}}^*, \varvec{\pi }^*)\in {\mathcal {A}}(x). \end{aligned}$$
(68)
Let \(\lambda ^{\epsilon } \triangleq (1+\epsilon )\lambda ^*\) where \(-1 < \epsilon \ne 0\). Let \({\bar{z}}^{\epsilon }_t \triangleq z_t^{\varvec{\delta }^*, \lambda ^{\epsilon }} = (1+\epsilon )z_t^*\) for \(t\ge 0\), and let
$$\begin{aligned} \bar{{\mathcal {N}}}^{\epsilon } \triangleq {\mathbb {E}}\left[ \int _0^\infty e^{-\rho t}\left\{ \left( {\widetilde{u}}_0(z^{*}_t)+Y_0z^{*}_t\right) {\mathbf {1}}_{ \{0<{\bar{z}}^{\epsilon }_t\le {\bar{z}}\}}+\left( {\widetilde{u}}_1(z^{*}_t)+Y_1z^{*}_t \right) {\mathbf {1}}_{\{{\bar{z}}^{\epsilon }_t>{\bar{z}}\}}\right\} dt\right] + \lambda ^{*}x. \end{aligned}$$
Since \(\bar{{\mathcal {N}}}^{\epsilon } \le \varPhi (\varvec{\delta }^{*}, \lambda ^*)\) by Remark 2 and \(\varPhi (\varvec{\delta }^{*}, \lambda ^*) \le \varPhi (\varvec{\delta }^{*}, \lambda ^{\epsilon })\), we have
$$\begin{aligned} \limsup _{\epsilon \downarrow 0}\frac{\varPhi (\varvec{\delta }^{*}, \lambda ^{\epsilon }) - \bar{{\mathcal {N}}}^{\epsilon }}{\epsilon } \ge 0 ~~\text{ and }~~ \liminf _{\epsilon \uparrow 0}\frac{\varPhi (\varvec{\delta }^{*}, \lambda ^{\epsilon }) - \bar{{\mathcal {N}}}^{\epsilon }}{\epsilon } \le 0, \end{aligned}$$
(69)
where
$$\begin{aligned}&\frac{\varPhi (\varvec{\delta }^{*}, \lambda ^{\epsilon }) - \bar{{\mathcal {N}}}^{\epsilon }}{\epsilon }\\&\quad ={\mathbb {E}}\Bigg [\int _0^{\infty } e^{-\rho t}\Bigg \{\Big (\frac{{\widetilde{u}}_0(z^*_t+\epsilon \lambda ^*\delta ^*_te^{\rho t}H_t)-{\widetilde{u}}_0(z^*_t)}{\epsilon }+Y_0\lambda ^*\delta ^*_te^{\rho t}H_t\Big ){\mathbf {1}}_{\{0<{\bar{z}}^{\epsilon }_t\le {\bar{z}}\}}\\&\qquad +\,\Big (\frac{{\widetilde{u}}_1(z^*_t+\epsilon \lambda ^*\delta ^*_te^{\rho t}H_t)-{\widetilde{u}}_1(z^*_t)}{\epsilon }+Y_1\lambda ^*\delta ^*_te^{\rho t}H_t\Big ){\mathbf {1}}_{\{{\bar{z}}^{\epsilon }_t>{\bar{z}}\}}\Bigg \}dt\Bigg ]+ \lambda ^*x. \end{aligned}$$
Note that \(0 < \lambda ^*\delta ^*_t \le \lambda ^*\) for all \(t \ge 0\). Similarly to the way of deriving (67), by applying Fatou’s Lemma to (69), we get
$$\begin{aligned} {\mathbb {E}}\left[ \int _0^{\infty }\left( c^*_t-Y_0 {\mathbf {1}}_{\{\varTheta ^*_t=A_0\}}-Y_1 {\mathbf {1}}_{\{\varTheta ^*_t=A_1\}}\right) \delta ^*_tH_tdt\right] = x. \end{aligned}$$
(70)
By (68), (19), (70), and Lemma 3, the theorem holds. \(\Box \)
E Proof of Lemma 4
The inequality (21) implies \(D_2>0.\) Define a function F(z) for \(z > 0\) as
$$\begin{aligned} F(z)\triangleq (n_+-n_-)n_-D_2z^{n_--1}-\frac{n_+\gamma _1+1-\gamma _1}{ \gamma _1K_1}L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}z^{-\frac{1}{\gamma _1}}+(n_+-1)\frac{Y_1}{r}. \end{aligned}$$
(71)
Then, \(F^{\prime }(z)>0\) and \(\displaystyle {\lim _{z\uparrow \infty }F(z)=(n_+-1){Y_1}/{r}>0}.\) We have
$$\begin{aligned} F({\bar{z}})&=(n_+-n_-)n_-D_2{\bar{z}}^{n_--1}-\frac{n_+\gamma _1+1-\gamma _1}{ \gamma _1K_1}L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}{\bar{z}}^{-\frac{1}{\gamma _1}}+(n_+-1) \frac{Y_1}{r}\\&=\frac{n_-(n_+\gamma _1+1-\gamma _1)}{\gamma _1K_1}\frac{L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}}{L_0^{ \frac{\gamma _1-\gamma }{\gamma _1}}-L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}}(Y_1-Y_0)\\&\quad -\frac{(n_+\gamma _1+1- \gamma _1)(n_-\gamma _1+1-\gamma _1)}{\gamma _1^2K_1}\frac{L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}}{L_0^{ \frac{\gamma _1-\gamma }{\gamma _1}}-L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}}(Y_1-Y_0)\\&\quad -\frac{(1-n_-)(1-n_+)}{r}(Y_1-Y_0)+\frac{(n_+-1)Y_0}{r}\\&=-\frac{(1-\gamma _1)(n_+\gamma _1+1-\gamma _1)}{\gamma _1^2K_1}\frac{L_0^{\frac{\gamma _1- \gamma }{\gamma _1}}}{L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}-L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}}(Y_1-Y_0)+ \frac{(n_+-1)Y_0}{r}\\&<0, \end{aligned}$$
where the second equality is obtained from plugging \(D_2\) in (25) and \({\bar{z}}\) in Remark 2 into the given equation, the third equality from the equality (22) in Remark 6, and the inequality from the condition (24). Thus \(F(\cdot )=0\) has a unique solution \({\tilde{z}} > {\bar{z}}\). The function v is continuous at \(z = {\tilde{z}}\) since it is defined to be \(v(z) = v({\tilde{z}})\) for \(z\ge {\tilde{z}}\).
The second order ordinary differential equation of v, \(-\rho v(z) + {\mathcal {L}}v(z)+\left( {\widetilde{u}}_0(z)+Y_0z\right) =0\) for \(0<z<{\bar{z}}\), has two homogeneous solutions \(z^{n_+}\) and \(z^{n_-}\), and a particular solution \(L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{\gamma _1}{(1-\gamma _1)K_1}z^{-\frac{1-\gamma _1}{\gamma _1}}+\frac{Y_0}{r}z\), so that the general solution of it is
$$\begin{aligned} const_1\cdot z^{n_+} + const_2z^{n_-} + L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{\gamma _1}{(1-\gamma _1)K_1}z^{-\frac{1-\gamma _1}{\gamma _1}}+ \frac{Y_0}{r}z, \end{aligned}$$
where \(const_1\) and \(const_2\) are arbitrary constants. Therefore, v satisfies the first one in the Bellman equations (30). Here, we are conjecturing that the term \(const_2z^{n_-}\) is discarded due to the rapid growth of \(z^{n_-}\) as \(z\downarrow 0\). Similarly, v satisfies also the second one in the Bellman equations. Thus, the function v satisfies the Bellman equations (30).
By calculation, it can be shown that v is continuously differentiable at \(z = {\bar{z}}\) and twice continuously differentiable at \(z = {\tilde{z}}\). Since \(n_+ > 1\), we have
$$\begin{aligned} \lim _{z\downarrow 0}v^{\prime }(z) = \lim _{z\downarrow 0}\left[ C_1n_+z^{n_+-1}-L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}z^{-\frac{1}{\gamma _1}}+ \frac{Y_0}{r}\right] = -\infty . \end{aligned}$$
Since v(z) is twice continuously differentiable at \(z={\tilde{z}}\) and constant for \(z\ge {\tilde{z}}\), we have
$$\begin{aligned} v^{\prime }(z) = v^{\prime \prime }(z) = 0 ~~\text{ for } z\ge {\tilde{z}}. \end{aligned}$$
(72)
Now we show that v is strictly decreasing and strictly convex on \((0, {\tilde{z}}]\). To do it, it suffice to show that \(v^{\prime \prime }(z) > 0\) for \( z \in (0, {\bar{z}})\cup ({\bar{z}}, {\tilde{z}})\), since v is smoothly pasted at \({\bar{z}}\) and \(v^{\prime }({\tilde{z}}) = 0\).
(1) Firstly we will show that \(v^{\prime \prime }(z) > 0\) for \(0< z < {\bar{z}}\). We see that, for \(0<z<{\bar{z}}\),
$$\begin{aligned} v^{\prime }(z)&=n_+C_1z^{n_+-1}-L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}z^{-\frac{1}{\gamma _1}}+\frac{Y_0}{r},\\ v^{\prime \prime }(z)&=z^{-\frac{1}{\gamma _1}-1}\left[ n_+(n_+-1)C_1z^{n_++\frac{1}{\gamma _1}-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}\right] . \end{aligned}$$
Let
$$\begin{aligned} h_1(z):=n_+(n_+-1)C_1z^{n_++\frac{1}{\gamma _1}-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}. \end{aligned}$$
If \(C_1\ge 0\), then \(h_1(z)>0\) and hence \(v^{\prime \prime }(z) > 0\) for \(0< z < {\bar{z}}\). If \(C_1<0\), then \(h_1(z)\) is decreasing, so it is enough to show that
$$\begin{aligned} h_1({\bar{z}})>0. \end{aligned}$$
From the coefficients \(D_2\) in (25), \(D_1\) in (27), and \(C_1\) in (28), we see that
$$\begin{aligned}&\displaystyle D_1=-\frac{n_-}{n_+}\frac{\frac{n_+\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_+}{r}}{(n_+-n_-){ \bar{z}}^{n_--1}}(Y_1-Y_0){\tilde{z}}^{n_--n_+}+\frac{L_1^{\frac{\gamma _1- \gamma }{\gamma _1}}}{n_+K_1}{\tilde{z}}^{1-n_+-\frac{1}{\gamma _1}}- \frac{Y_1}{n_+r}{\tilde{z}}^{1-n_+},\\&\displaystyle C_1=D_1+\frac{\frac{n_-\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_-}{r}}{(n_+-n_-){ \bar{z}}^{n_+-1}}(Y_1-Y_0). \end{aligned}$$
Since \({\bar{z}}<{\tilde{z}}\), we have the following inequalities:
$$\begin{aligned} \left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-n_-}< \left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}<\left( \frac{{\bar{z}}}{{ \tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}<\left( \frac{{\bar{z}}}{{ \tilde{z}}}\right) ^{n_+-1}<1. \end{aligned}$$
(73)
Then \(h_1({\bar{z}})\) is given by
$$\begin{aligned} h_1({\bar{z}})&=-n_-(n_+-1)\frac{\frac{n_+\gamma _1+1-\gamma _1}{\gamma _1K_1}+ \frac{1-n_+}{r}}{(n_+-n_-)}(Y_1-Y_0){{\bar{z}}^{\frac{1}{\gamma _1}}}{ \left( \frac{{\bar{z}}}{\tilde{z}}\right) ^{n_+-n_-}}\\&\quad +(n_+-1)\frac{L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}}{K_1}{\left( \frac{{ \bar{z}}}{\tilde{z}}\right) ^{n_++\frac{1}{\gamma _1}-1}}-(n_+-1)\frac{Y_1}{r}{{ \bar{z}}^{\frac{1}{\gamma _1}}}{\left( \frac{{\bar{z}}}{\tilde{z}}\right) ^{n_+-1}}\\&\quad +n_+(n_+-1)\frac{\frac{n_-\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_-}{r}}{(n_+-n_-)}( Y_1-Y_0){{\bar{z}}^{\frac{1}{\gamma _1}}}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}} \frac{1}{\gamma _1 K_1}\\&=\frac{n_-(1-n_-)}{n_+-n_-}\left[ \frac{n_+\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_+}{r} \right] (Y_1-Y_0) {{\tilde{z}}^{\frac{1}{\gamma _1}}}{\left( \frac{{\bar{z}}}{ \tilde{z}}\right) ^{n_++\frac{1}{\gamma _1}-n_-}}\\&\quad -\frac{1}{\gamma _1 K_1}L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}{\left( \frac{{\bar{z}}}{\tilde{z}} \right) ^{n_++\frac{1}{\gamma _1}-1}}+n_+(n_+-1)\frac{\frac{n_-\gamma _1+1-\gamma _1}{ \gamma _1K_1}+\frac{1-n_-}{r}}{(n_+-n_-)}(Y_1-Y_0){{\bar{z}}^{\frac{1}{ \gamma _1}}}\\&\quad +L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}, \end{aligned}$$
where the second equality is from \(F({\tilde{z}})=0\) in (71).
If \(\gamma >1\), then \(\gamma _1>1\) and
$$\begin{aligned} n_+(n_+-1)\frac{\frac{n_-\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_-}{r}}{(n_+-n_-)}\frac{\gamma _1}{\gamma _1-1}>\frac{1}{\gamma _1 K_1}, \end{aligned}$$
(74)
which follows from (22), (20), and the fact that \(n_+>1\). If \(\gamma >1\), then \(\gamma>\gamma _1>1\) and \(L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}<L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\), and hence \(h_1({\bar{z}}) > 0\) by
$$\begin{aligned} h_1({\bar{z}})&=\frac{n_-(1-n_-)}{n_+-n_-}\left[ \frac{n_+\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_+}{r}\right] (Y_1-Y_0) {{\tilde{z}}^{\frac{1}{\gamma _1}}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-n_-}}\\&\quad -\frac{1}{\gamma _1 K_1}L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}}+n_+(n_+-1)\frac{\frac{n_-\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_-}{r}}{(n_+-n_-)}(Y_1-Y_0){{\bar{z}}^{\frac{1}{\gamma _1}}}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}\\&=-\frac{1-\gamma _1}{\gamma _1^2K_1}(Y_1-Y_0){{\bar{z}}^{\frac{1}{\gamma _1}}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}}-\frac{1}{\gamma _1 K_1}L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}}\\&\quad +n_+(n_+-1)\frac{\frac{n_-\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_-}{r}}{(n_+-n_-)}\frac{\gamma _1}{\gamma _1-1}\left( L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}-L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\right) {\left( 1-\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}\right) }+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}\\&{>}-\frac{1}{\gamma _1K_1}\left( L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}-L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\right) {\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}}-\frac{1}{\gamma _1 K_1}L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}}\\&\quad +\frac{1}{\gamma _1K_1}\left( L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}-L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\right) {\left( 1-\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}\right) }+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}\\&=\frac{1}{\gamma _1K_1}\left[ L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( 1-{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}}-\left( 1-\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}\right) \right) \right. \\&\quad \left. +L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( {\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}}-{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}}+\left( 1-\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}\right) \right) \right] \\&=\frac{1}{\gamma _1K_1}L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( 1-\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}\right) \\&>0, \end{aligned}$$
where the second equality follows from the equality (22), the first inequality from (74), and the last inequality from (73).
If \(0<\gamma <1\), then \(0<\gamma<\gamma _1<1\) and \(L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}>L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\), and hence \(h_1({\bar{z}}) > 0\) by
$$\begin{aligned} h_1({\bar{z}})&=\frac{n_-(1-n_-)}{n_+-n_-}\left[ \frac{n_+\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_+}{r}\right] (Y_1-Y_0) {{\tilde{z}}^{\frac{1}{\gamma _1}}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-n_-}}\\&\quad -\frac{1}{\gamma _1 K_1}L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}}+n_+(n_+-1)\frac{\frac{n_-\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_-}{r}}{(n_+-n_-)}(Y_1-Y_0){{\bar{z}}^{\frac{1}{\gamma _1}}}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}\\&>\frac{n_-(1-n_-)}{n_+-n_-}\left[ \frac{n_+\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_+}{r}\right] (Y_1-Y_0) {{\tilde{z}}^{\frac{1}{\gamma _1}}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-n_-}}\\&\quad -\frac{1}{\gamma _1 K_1}L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}}\\&\quad +n_+(n_+-1)\frac{\frac{n_-\gamma _1+1-\gamma _1}{\gamma _1K_1}+\frac{1-n_-}{r}}{(n_+-n_-)}(Y_1-Y_0){{\tilde{z}}^{\frac{1}{\gamma _1}}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-n_-}}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}\\&=-\frac{1-\gamma _1}{\gamma _1^2K_1}(Y_1-Y_0){{\bar{z}}^{\frac{1}{\gamma _1}}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}}-\frac{1}{\gamma _1 K_1}L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}\\&=-\frac{1}{\gamma _1K_1}\left( L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}-L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\right) {\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}}-\frac{1}{\gamma _1 K_1}L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}\\&=\frac{1}{\gamma _1K_1}\left[ L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( 1-{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}}\right) +L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( {\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_+-n_-}}-{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}}\right) \right] \\&>\frac{1}{\gamma _1K_1}\left[ L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( 1-{\left( \frac{{\bar{z}}}{{\tilde{z}}}\right) ^{n_++\frac{1}{\gamma _1}-1}}\right) \right] \\&>0, \end{aligned}$$
where the first inequality is from (73) for the fourth term in left-hand side, the second equality from (22), and the second and the last inequalities from (73).
(2) Secondly we show that \(v^{\prime \prime }(z) > 0\) for \({\bar{z}}< z < {\tilde{z}}\). We see that
$$\begin{aligned} v'(z)&=n_+D_1z^{n_+-1}+n_-D_2z^{n_--1}-L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}z^{-\frac{1}{\gamma _1}}+\frac{Y_1}{r},\\ v''(z)&=z^{-\frac{1}{\gamma _1}-1}\left[ n_+(n_+-1)D_1z^{n_++\frac{1}{\gamma _1}-1}+n_-(n_--1)D_2z^{n_-+\frac{1}{\gamma _1}-1}+L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}\right] \end{aligned}$$
Let
$$\begin{aligned} h_2(z)&:=n_+(n_+-1)D_1z^{n_++\frac{1}{\gamma _1}-1}+n_-(n_--1)D_2z^{n_-+\frac{1}{\gamma _1}-1}+L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}\\&=z^{n_-+\frac{1}{\gamma _1}-1}\left[ n_+(n_+-1)D_1z^{n_+-n_-}+n_-(n_--1)D_2\right] +L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}. \end{aligned}$$
Since \(D_1<0,~D_2>0\), and \(n_-+\frac{1}{\gamma _1}-1<0\), \(h_2(z)\) is strictly decreasing, so that
$$\begin{aligned} h_2(z) > h_2({\tilde{z}}) = 0, ~~\text{ for }~~{\bar{z}}< z < {\tilde{z}}, \end{aligned}$$
where the equality holds since \(v^{\prime \prime }({\tilde{z}}) = 0\). Thus \(v''(z)> 0\) for \({\bar{z}}<z<{\tilde{z}}\). \(\Box \)
F Proof of Theorem 2
Let
$$\begin{aligned} z^*_t \triangleq {\mathcal {Z}}(X_t), \quad t \ge 0, \end{aligned}$$
(75)
and let
$$\begin{aligned} \lambda ^* \triangleq z^*_0 = {\mathcal {Z}}(x). \end{aligned}$$
(76)
Since \(x = X_0 \ge 0\), \(\lambda ^*\) in (76) is well defined and \(0 < \lambda ^* \le {\tilde{z}}\). By (33) and (75), we have
$$\begin{aligned} X_t\ge {\bar{x}} \Leftrightarrow 0<z^*_t\le {\bar{z}} ~~\text{ and }~~ 0\le X_t<{\bar{x}} \Leftrightarrow z^*_t > {\bar{z}}. \end{aligned}$$
With the strategy \((\varvec{\varTheta }^*,\varvec{c}^*,\varvec{\pi }^*)\) in the theorem, the wealth process (1) satisfies
$$\begin{aligned} dX_t = \left[ rX_t+\theta ^2z^*_tv^{\prime \prime }(z^*_t)-c^*_t+Y_0 {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}+Y_1 {\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\right] dt+\theta z^*_tv^{\prime \prime }(z^*_t)dB_t. \end{aligned}$$
If \(X_t = 0\) for \(\exists t \ge 0\), then \(z^*_t = {\mathcal {Z}}(0) = {\tilde{z}}\) so that
$$\begin{aligned} dX_t = [-f_1({\tilde{z}}) + Y_1]dt. \end{aligned}$$
(77)
However, by using (21) and the fact that \({\tilde{z}}\) a solution to (26) and \(D_2 > 0\) as stated in Remark 7, we can show
$$\begin{aligned} -f_1({\tilde{z}}) + Y_1 > 0. \end{aligned}$$
(78)
Therefore, \(X_t = 0\) is an upward reflection boundary so that \(X_t \ge 0\) for all \(t\ge 0\) with the strategy \((\varvec{\varTheta }^*, {\mathbf {c}}^*, \varvec{\pi }^*)\). Thus, the relation (75) is well defined, \(0 < z^*_t \le {\tilde{z}}\), and
$$\begin{aligned} (\varvec{\varTheta }^*, {\mathbf {c}}^*, \varvec{\pi }^*)\in {\mathcal {A}}(x). \end{aligned}$$
(79)
Since the Bellman equations (30) hold for all \(z \in (0, {\bar{z}}) \cup ({\bar{z}}, {\tilde{z}})\), we have, by differentiation with respect to z,
$$\begin{aligned} \left\{ \begin{array}{ll} -rv^{\prime }(z) + (\rho -r+\theta ^2)zv^{\prime \prime }(z)+ \frac{1}{2}\theta ^2z^2v^{\prime \prime \prime }(z) -f_0(z)+Y_0 =0,&{}\quad \text{ for } 0<z<{\bar{z}},\\ -rv^{\prime }(z) + (\rho -r+\theta ^2)zv^{\prime \prime }(z)+ \frac{1}{2}\theta ^2z^2v^{\prime \prime \prime }(z) -f_1(z)+Y_1=0,&{}\quad \text{ for } {\bar{z}}<z<{\tilde{z}}, \end{array}\right. \end{aligned}$$
which, by using (31), can be rewritten as
$$\begin{aligned} \left\{ \begin{array}{ll} r{\mathcal {X}}(z) -(\rho -r+\theta ^2)z{\mathcal {X}}^{\prime }(z)- \frac{1}{2}\theta ^2z^2{\mathcal {X}}^{\prime \prime }(z) -f_0(z)+Y_0 =0,&{}\quad \text{ for } 0<z<{\bar{z}},\\ r{\mathcal {X}}(z) -(\rho -r+\theta ^2)z{\mathcal {X}}^{\prime }(z)- \frac{1}{2}\theta ^2z^2{\mathcal {X}}^{\prime \prime }(z) -f_1(z)+Y_1=0,&{}\quad \text{ for } {\bar{z}}<z<{\tilde{z}}. \end{array}\right. \nonumber \\ \end{aligned}$$
(80)
Let \(\delta ^*_t \triangleq e^{-\rho t}H_t^{-1}z^*_t/\lambda ^*\) for \(t\ge 0\). Thus, \(\delta ^*_t > 0\) for \(t\ge 0\) with \(\delta ^*_0 = 1\), and
$$\begin{aligned} z^*_t = \lambda ^*\delta ^*_t e^{\rho t}H_t. \end{aligned}$$
(81)
Suppose \(0< z^*_{t_0} = \lambda ^*\delta ^*_{t_0} e^{\rho t_0}H_{t_0} < {\tilde{z}}\) (equivalently \(X_{t_0} > 0\)) at \(\exists t_0 \ge 0\). Then, the process \((z^*_t)_{t\ge t_0}\), before \(z^*_{t}\) touching \({\tilde{z}}\), evolves according to
$$\begin{aligned} dz^*_t= & {} d{\mathcal {Z}}(X_t)\\= & {} {\mathcal {Z}}^{\prime }(X_t)\left\{ \left[ rX_t+\theta ^2z^*_tv^{\prime \prime }(z^*_t)-c^*_t+Y_0 {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}+Y_1 {\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\right] dt+\theta z^*_tv^{\prime \prime }(z^*_t)dB_t\right\} \\&+\, \frac{1}{2}{\mathcal {Z}}^{\prime \prime }(X_t)(\theta z^*_tv^{\prime \prime }(z^*_t))^2 dt\\= & {} {\mathcal {Z}}^{\prime }(X_t)\left\{ \left[ r{\mathcal {X}}(z^*_t)-\theta ^2z^*_t{\mathcal {X}}^{\prime }(z^*_t)-c^*_t+Y_0 {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}+Y_1 {\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\right] dt-\theta z^*_t{\mathcal {X}}^{\prime }(z^*_t)dB_t\right\} \\&+\, \frac{1}{2}{\mathcal {Z}}^{\prime \prime }(X_t)(\theta z^*_t{\mathcal {X}}^{\prime }(z^*_t))^2dt\\= & {} {\mathcal {Z}}^{\prime }(X_t)\left[ r{\mathcal {X}}(z^*_t)-\theta ^2z^*_t{\mathcal {X}}^{\prime }(z^*_t)+ \frac{1}{2}{\mathcal {Z}}^{\prime \prime }(X_t)(\theta z^*_t)^2({\mathcal {X}}^{\prime }(z^*_t))^3-c^*_t+Y_0 {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}+Y_1 {\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\right] dt\\&-\,{\mathcal {Z}}^{\prime }(X_t)\theta z^*_t{\mathcal {X}}^{\prime }(z^*_t)dB_t\\= & {} {\mathcal {Z}}^{\prime }(X_t)\left[ (\rho -r)z^*_t{\mathcal {X}}^{\prime }(z^*_t)+ \frac{1}{2}(\theta z^*_t)^2{\mathcal {X}}^{\prime \prime }(z^*_t) + \frac{1}{2}{\mathcal {Z}}^{\prime \prime }(X_t)(\theta z^*_t)^2({\mathcal {X}}^{\prime }(z^*_t))^3\right] dt\\&-\,{\mathcal {Z}}^{\prime }(X_t)\theta z^*_t{\mathcal {X}}^{\prime }(z^*_t)dB_t\\= & {} \left[ (\rho -r)z^*_t + \frac{1}{2}(\theta z^*_t)^2\left\{ {\mathcal {X}}^{\prime \prime }(z^*_t) {\mathcal {Z}}^{\prime }(X_t) + {\mathcal {Z}}^{\prime \prime }(X_t)({\mathcal {X}}^{\prime }(z^*_t))^2\right\} \right] dt\\&-\,\theta z^*_tdB_t\\= & {} z^*_t[(\rho -r) dt -\theta dB_t], \end{aligned}$$
where the second equality comes from the generalized Itô’s rule, the third from (31), (32), and (75), the fifth from the definition of \(c^*_t\) and (80), the sixth from the fact that \({\mathcal {X}}^{\prime }({\mathcal {Z}}(X_t)){\mathcal {Z}}^{\prime }(X_t) = 1\) by the inverse relationship of \({\mathcal {X}}\) and \({\mathcal {Z}}\), and the last holds since \({\mathcal {X}}^{\prime \prime }({\mathcal {Z}}(X_t)) {\mathcal {Z}}^{\prime }(X_t) + {\mathcal {Z}}^{\prime \prime }(X_t)({\mathcal {X}}^{\prime }({\mathcal {Z}}(X_t)))^2 = {\mathcal {X}}^{\prime }({\mathcal {Z}}(X_t))[{\mathcal {X}}^{\prime \prime }({\mathcal {Z}}(X_t)) ({\mathcal {Z}}^{\prime }(X_t))^2 + {\mathcal {X}}^{\prime }({\mathcal {Z}}(X_t)){\mathcal {Z}}^{\prime \prime }(X_t)]= 0\) by the inverse relationship of \({\mathcal {X}}\) and \({\mathcal {Z}}\). Therefore, we have
$$\begin{aligned} z^*_t= & {} z^*_{t_0}e^{(\rho -r - \frac{\theta ^2}{2})(t-t_0) - \theta (B_t-B_{t_0})}\\= & {} \lambda ^*\delta ^*_{t_0} e^{\rho t_0}H_{t_0}e^{(\rho -r - \frac{\theta ^2}{2})(t-t_0) - \theta (B_t-B_{t_0})}\\= & {} \lambda ^*\delta ^*_{t_0} e^{(\rho -r - \frac{\theta ^2}{2})t - \theta B_t}\\= & {} \lambda ^*\delta ^*_{t_0}e^{\rho t}H_{t}, \end{aligned}$$
which means \(\delta ^*_{t}\) in (81) stays constant before \(z^*_{t}\) touching \({\tilde{z}}\). That is,
$$\begin{aligned} d\delta ^*_{t} = 0 ~~ \text{ when }~~ 0< z^*_t < {\tilde{z}}. \end{aligned}$$
However, once \(z^*_t\) hits \({\tilde{z}}\), that is, once \(X_t\) hits 0, \(X_t\) moves in the upward direction by (77) and (78) so that \(z^*_t\) moves in the downward direction. That is, \(z^*_t = {\tilde{z}}\) is a downward reflection boundary since \(X_t = 0\) is an upward reflection boundary. Actually, \(z^*_t\) moves in the downward direction in an infinite instantaneous speed, since \({\mathcal {Z}}^{\prime }(0) \triangleq {\mathcal {Z}}^{\prime }(0+) = 1/{\mathcal {X}}^{\prime }({\tilde{z}}-) = -1/v^{\prime \prime }({\tilde{z}}-) = -\infty \). More concretely, suppose that \(z^*_{t_0} = \lambda ^*\delta ^*_{t_0} e^{\rho t_0}H_{t_0} = {\tilde{z}}\) for \(\exists t_0\ge 0\), that is, \(X_{t_0} = 0\), then, by (77) and (78),
$$\begin{aligned} dz^*_{t_0} = d{\mathcal {Z}}(X_{t_0}) = {\mathcal {Z}}^{\prime }(X_{t_0})\left[ -f_1({\tilde{z}}) + Y_1\right] dt = {\mathcal {Z}}^{\prime }(X_{t_0})\left[ -f_1({\tilde{z}}) + Y_1\right] dt = -\infty dt, \end{aligned}$$
which implies \(\delta ^*_t\) also decreases in an infinite instantaneous speed at \(t_0\). Thus, \(\varvec{\delta }^* \triangleq (\delta ^*_t)_{t\ge 0} \in {\mathcal {D}}\) and it decreases only when \(z^*_t\) hits \({\tilde{z}}\). As in He and Pagès [8], the decreases of \(\delta ^*_t\) when \(z^*_t\) hitting \({\tilde{z}}\), regulate \((z^*_t)_{t\ge 0}\) to stay in \((0, {\tilde{z}})\), thus \(\varvec{\delta }^*\) can be characterized by the local times of \((z^*_t)_{t\ge 0}\) at \({\tilde{z}}\).
We first consider the case where \(\gamma _1 > 1\). Fix \(T > 0\). We have
$$\begin{aligned}&\int _0^{T} e^{-\rho t}\Big \{\left( {\widetilde{u}}_0(z^*_t)+Y_0z^*_t\right) {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}} +\left( {\widetilde{u}}_1(z^*_t)+Y_1z^*_t\right) {\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\Big \}\,dt\nonumber \\&\qquad + \,e^{-\rho T}v(z^*_{T}) \end{aligned}$$
(82)
$$\begin{aligned}= & {} \int _0^{T} e^{-\rho t}\Big \{\left( {\widetilde{u}}_0(z^*_t)+Y_0z^*_t\right) {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}} +\left( {\widetilde{u}}_1(z^*_t)+Y_1z^*_t\right) {\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\Big \}\,dt + v(\lambda ^*)\nonumber \\&\,+\int _0^{T}e^{-\rho t}\Big \{-\rho v(z^*_t) + {\mathcal {L}}v(z^*_t)\Big \}\,dt+\int _0^{T}v^{\prime }(z^*_t)\lambda ^*H_t\,d\delta ^*_t - \int _0^{T}e^{-\rho t}v^{\prime }(z^*_t)\theta z^*_t\,dB_t\nonumber \\= & {} v(\lambda ^*) - \int _0^{T}e^{-\rho t}v^{\prime }(z^*_t)\theta z^*_t\,dB_t, \end{aligned}$$
(83)
where the first equality comes from the generalized Itô’s rule using (81) and (15), the second from the fact that v satisfies the Bellman equations (30) and that \(\int _0^Tv^{\prime }(z^*_t)\lambda ^*H_t\,d\delta ^*_t = 0\) since \((\delta ^*_t)_{t\ge 0} \in {\mathcal {D}}\) decreases only when \(z^*_t\) hits \({\tilde{z}}\) with \(v^{\prime }({\tilde{z}}) = 0.\) By using (29), we have
$$\begin{aligned} zv^{\prime }(z)=\left\{ \begin{array}{ll} n_+C_1z^{n_+}-L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}z^{1-\frac{1}{\gamma _1}}+ \frac{Y_0}{r}z, \quad 0<z\le {\bar{z}},\\ n_+D_1z^{n_+}+n_-D_2z^{n_-}-L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}z^{1- \frac{1}{\gamma _1}}+\frac{Y_1}{r}z, \quad {\bar{z}}<z\le {\tilde{z}}, \end{array}\right. \end{aligned}$$
(84)
which is bounded for \(z \in (0, {\tilde{z}}]\) since \(\lim _{z\downarrow 0}zv^{\prime }(z) = 0\) with \(\gamma _1 > 1\). Therefore, with \(\gamma _1 > 1\),
$$\begin{aligned} {\mathbb {E}}\Big [\int _0^{T}\big (e^{-\rho t}v^{\prime }(z^*_t)\theta z^*_t\big )^2\,dt\Big ] < \infty . \end{aligned}$$
Thus, the process \((\int _0^{t}e^{-\rho s}v^{\prime }(z^*_s)\theta z^*_s\,dB_s)_{t\ge 0}\) is a martingale under \({\mathbb {P}}\), with \(\gamma _1 > 1\), so that
$$\begin{aligned} {\mathbb {E}}\Big [\int _0^{T}e^{-\rho t}v^{\prime }(z^*_t)\theta z^*_t\,dB_t\Big ] = 0. \end{aligned}$$
Hence, by taking the expectation operator on (82) and (83), we have, with \(\gamma _1 > 1\),
$$\begin{aligned} {\mathbb {E}}\Big [\int _0^{T} e^{-\rho t}\Big \{\left( {\widetilde{u}}_0(z^*_t)+Y_0z^*_t\right) {\mathbf {1}}_{\{0<z^*_t \le {\bar{z}}\}} +\left( {\widetilde{u}}_1(z^*_t)+Y_1z^*_t\right) {\mathbf {1}}_{\{z^*_t>{ \bar{z}}\}}\Big \}\,dt\Big ] + {\mathbb {E}}\Big [e^{-\rho T}v(z^*_{T})\Big ] = v(\lambda ^*). \end{aligned}$$
(85)
When \(\gamma _1 > 1\), the function v defined as (29) is bounded for \(z \in (0, {\tilde{z}}]\) since \(\lim _{z\downarrow 0}v(z) = 0\), which implies
$$\begin{aligned} \lim _{T\uparrow \infty }{\mathbb {E}}\Big [e^{-\rho T}v(z^*_{T})\Big ] = 0. \end{aligned}$$
(86)
Letting \(T\uparrow \infty \) on both sides of the equality (85), by applying the monotone convergence theorem and by (86), we get, when \(\gamma _1 > 1\),
$$\begin{aligned} v(\lambda ^*) = {\mathbb {E}}\Big [\int _0^{\infty } e^{-\rho t}\Big \{\left( {\widetilde{u}}_0(z^*_t)+Y_0z^*_t\right) {\mathbf {1}}_{\{0<z^*_t \le {\bar{z}}\}} +\left( {\widetilde{u}}_1(z^*_t)+Y_1z^*_t\right) {\mathbf {1}}_{\{z^*_t>{ \bar{z}}\}}\Big \}\,dt\Big ]. \end{aligned}$$
(87)
By the generalized Itô’s rule using (10), we have, with \(T > 0\),
$$\begin{aligned}&\int _0^T\left( c^*_t-Y_0 {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}-Y_1 { \mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\right) \delta ^*_tH_t\,dt\\&\quad = x - H_TX_T\delta ^*_T + \int _0^TH_tX_t\,d\delta ^*_t + \int _0^T(\sigma \pi ^*_t-\theta X_t)\delta ^*_tH_tdB_t\\&\quad = x + \frac{1}{\lambda ^*}e^{-\rho T}v^{\prime }(z^*_T)z^*_T - \int _0^TH_tv^{ \prime }(z^*_t)\,d\delta ^*_t + \frac{1}{\lambda ^*}\int _0^Te^{-\rho t}\big (\sigma z^*_tv^{ \prime \prime }(z^*_t)\frac{\theta }{\sigma }+\theta v^{\prime }(z^*_t)\big )z^*_t\,dB_t\\&\quad = x + \frac{1}{\lambda ^*}e^{-\rho T}v^{\prime }(z^*_T)z^*_T + \frac{1}{\lambda ^*}\int _0^Te^{-\rho t}\big (\sigma z^*_tv^{\prime \prime }(z^*_t)\frac{\theta }{\sigma }+\theta v^{\prime }(z^*_t)\big )z^*_t\,dB_t, \end{aligned}$$
where the second equality holds by (75), (81), and the formula of \(\pi ^*_t\) in the theorem, the third by the fact that \(\int _0^TH_tv^{\prime }(z^*_t)\,d\delta ^*_t = 0\). Similarly to the derivation of (87), we can derive
$$\begin{aligned} {\mathbb {E}}\left[ \int _0^\infty \left( c^*_t-Y_0 {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}-Y_1 {\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\right) \delta ^*_tH_tdt\right] = x. \end{aligned}$$
(88)
By (79), (34), (88), and Lemma 3, the triple \((\varvec{\varTheta }^*, {\mathbf {c}}^*, \varvec{\pi }^*)\) of the policies given in the theorem is optimal and
$$\begin{aligned} V(x) = \varPhi (\varvec{\delta }^*, \lambda ^*) = v(\lambda ^*) + \lambda ^*x = v({\mathcal {Z}}(x)) + x{\mathcal {Z}}(x), \end{aligned}$$
where the second equality comes from (87) and the third from (76).
Now we consider the case where \(0< \gamma _1 < 1\). Note that \(z^*_t = {\mathcal {Z}}(X_t)\) has continuous sample paths and \(z^*_0 = \lambda ^*\). Let \(\xi _{\epsilon } \triangleq \inf \{t\ge 0: z^*_t = \epsilon \}\) where \(0< \epsilon < \lambda ^*\) is fixed. Let \(\tau = T \wedge \xi _{\epsilon } \triangleq \min [T, \xi _{\epsilon }]\) where \(T > 0\) is fixed. Since \(zv^{\prime }(z)\) is bounded for \(z \in [\epsilon , {\tilde{z}}]\), similarly to (85), we get
$$\begin{aligned} {\mathbb {E}}\Big [\int _0^{\tau } e^{-\rho t}\Big \{\left( {\widetilde{u}}_0(z^*_t)+Y_0z^*_t\right) {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}} +\left( {\widetilde{u}}_1(z^*_t)+Y_1z^*_t\right) {\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\Big \}\,dt\Big ] + {\mathbb {E}}\Big [e^{-\rho \tau }v(z^*_{\tau })\Big ] = v(\lambda ^*). \end{aligned}$$
(89)
Letting \(\epsilon \downarrow 0\) and \(T\uparrow \infty \), then \(\tau \uparrow \infty \). When \(0< \gamma _1 < 1\), \({\widetilde{u}}_i(z) \ge 0\) by (5), and we can show that \(v(z) \ge v({\tilde{z}}) \ge 0\) by using (27), (29), \(n_+ > 1\), and the fact that \(D_2>0\) as mentioned in Remark 7. Thus, by the monotone convergence theorem, the first term on the left-hand side of (89) has a limit as \(\epsilon \downarrow 0\) and \(T\uparrow \infty \), and the limit is finite since it is between \([0, v(\lambda ^*)]\). Therefore, \({\mathbb {E}}[e^{-\rho \tau }v(z^*_{\tau })]\) also has a finite limit as \(\epsilon \downarrow 0\) and \(T\uparrow \infty \), which is nonnegative. However, we have, by (29),
$$\begin{aligned}&\lim _{T\uparrow \infty , \epsilon \downarrow 0}{\mathbb {E}}\Big [e^{-\rho \tau }v(z^*_{\tau })\Big ]\nonumber \\&\quad = \lim _{T\uparrow \infty , \epsilon \downarrow 0}{\mathbb {E}}\Big [e^{-\rho \tau }v(z^*_{\tau }){\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}} + e^{-\rho \tau }v(z^*_{\tau }){\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\Big ]\nonumber \\&\quad = \lim _{T\uparrow \infty , \epsilon \downarrow 0}{\mathbb {E}}\Big [e^{-\rho \tau }v(z^*_{\tau }){\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}\Big ]\nonumber \\&\quad =\lim _{T\uparrow \infty , \epsilon \downarrow 0}{\mathbb {E}}\Big [e^{-\rho \tau }L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{\gamma _1}{(1- \gamma _1)K_1}(z^*_{\tau })^{-\frac{1-\gamma _1}{\gamma _1}}{\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}\Big ]\nonumber \\&\quad =\frac{\gamma _1}{(1-\gamma _1)K_1}\lim _{T\uparrow \infty , \epsilon \downarrow 0}{\mathbb {E}}\Big [e^{-\rho \tau }L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}(z^*_{\tau })^{-\frac{1-\gamma _1}{ \gamma _1}}{\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}\Big ], \end{aligned}$$
(90)
where the second equality comes from the fact that v(z) is bounded for \(z \in ({\bar{z}}, {\tilde{z}}]\), and the third from that \(z^{n_+}\) is bounded for \(z \in (0, {\bar{z}}]\). By (79) and Remark 3, we have
$$\begin{aligned} {\mathbb {E}}\left[ \int _0^\infty \left( c^*_t-Y_0 {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}-Y_1 {\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\right) \delta ^*_tH_tdt\right] \le x. \end{aligned}$$
Therefore,
$$\begin{aligned} 0 \le {\mathbb {E}}\left[ \int _0^{\infty }c^*_t\delta ^*_tH_tdt\right]\le & {} x + {\mathbb {E}}\left[ \int _0^\infty \left( Y_0 {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}+Y_1 {\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\right) \delta ^*_tH_tdt\right] \\\le & {} x + (Y_0 + Y_1){\mathbb {E}}\left[ \int _0^\infty H_tdt\right] \\= & {} x + \frac{Y_0 + Y_1}{r} < \infty , \end{aligned}$$
where the last inequality comes from the fact that \(0 < \delta ^*_t \le 1\) for \(t \ge 0\), and the equality from \({\mathbb {E}}\left[ \int _0^\infty H_tdt\right] = \int _0^\infty {\mathbb {E}}\left[ H_t\right] dt = 1/r\). That is, \({\mathbb {E}}\left[ \int _0^{\infty }c^*_t\delta ^*_tH_tdt\right] \) is finite. However, by (34), we have
$$\begin{aligned} {\mathbb {E}}\left[ \int _0^{\infty }c^*_t\delta ^*_tH_tdt\right]= & {} {\mathbb {E}}\left[ \int _0^\infty \left( L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( z^*_t\right) ^{-\frac{1}{\gamma _1}}{\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}+L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( z^*_t\right) ^{-\frac{1}{\gamma _1}} {\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\right) \delta ^*_tH_tdt\right] \\= & {} \frac{1}{\lambda ^*}{\mathbb {E}}\left[ \int _0^\infty e^{-\rho t}\left( L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( z^*_t\right) ^{-\frac{1-\gamma _1}{\gamma _1}}{\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}\!+\!L_1^{\frac{\gamma _1\!-\!\gamma }{\gamma _1}}\left( z^*_t\right) ^{-\frac{1-\gamma _1}{\gamma _1}}{\mathbf {1}}_{\{z^*_t>{\bar{z}}\}}\right) dt\right] . \end{aligned}$$
Therefore, \({\mathbb {E}}\left[ \int _0^\infty e^{-\rho t}L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( z^*_t\right) ^{-\frac{1-\gamma _1}{\gamma _1}}{\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}dt\right] = \int _0^\infty {\mathbb {E}}\left[ e^{-\rho t}L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( z^*_t\right) ^{-\frac{1-\gamma _1}{\gamma _1}}\right. \left. {\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}\right] dt\) is finite, which implies
$$\begin{aligned} \lim _{t\rightarrow \infty }{\mathbb {E}}\left[ e^{-\rho t}L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( z^*_t\right) ^{-\frac{1-\gamma _1}{\gamma _1}}{\mathbf {1}}_{\{0<z^*_t\le {\bar{z}}\}}\right] = 0. \end{aligned}$$
Combining this with (90), we get
$$\begin{aligned} \lim _{T\uparrow \infty , \epsilon \downarrow 0}{\mathbb {E}}\Big [e^{-\rho \tau }v(z^*_{\tau })\Big ] = 0. \end{aligned}$$
Therefore, letting \(\epsilon \downarrow 0\) and \(T\uparrow \infty \) on both sides of (89), we find that (87) holds also for the case where \(0< \gamma _1 < 1\). Similarly, we can show that (88) holds also for the case where \(0< \gamma _1 < 1\). Thus, the theorem is proved also for the case where \(0< \gamma _1 < 1\). \(\Box \)
G Proof of Lemma 5
By (20) and the fact that \(n_+ > 1\), (37) is well defined and \({\tilde{z}} > 0\). By using (36), we can show \({\tilde{z}} \le {\bar{z}}\). By calculation, we can show that
$$\begin{aligned} E_1 = -\frac{Y_0}{rn_+(\gamma _1n_+-\gamma _1+1)}{\tilde{z}}^{1-n_+} < 0, \end{aligned}$$
where the inequality comes from (20). The function w is continuous at \(z = {\tilde{z}}\) since it is defined to be \(w(z) = w({\tilde{z}})\) for \(z\ge {\tilde{z}}\). The Bellman equation (39) is a second order ordinary differential equation, which has two homogeneous solutions \(z^{n_+}\) and \(z^{n_-}\), and a particular solution \(L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{\gamma _1}{(1-\gamma _1)K_1}z^{-\frac{1-\gamma _1}{\gamma _1}}+\frac{Y_0}{r}z\), so that the general solution of it is
$$\begin{aligned} const_1\cdot z^{n_+} + const_2z^{n_-} + L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{\gamma _1}{(1-\gamma _1)K_1}z^{-\frac{1-\gamma _1}{\gamma _1}}+\frac{Y_0}{r}z, \end{aligned}$$
where \(const_1\) and \(const_2\) are arbitrary constants. Therefore, w satisfies the Bellman equation (39). Here, we are conjecturing that the term \(const_2z^{n_-}\) is discarded due to the rapid growth of \(z^{n_-}\) as \(z\downarrow 0\).
By calculation, it can be shown that w is twice continuously differentiable at \(z = {\tilde{z}}\). Since \(n_+ > 1\), we have
$$\begin{aligned} \lim _{z\downarrow 0}w^{\prime }(z) = \lim _{z\downarrow 0}[E_1n_+z^{n_+-1}-L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}z^{-\frac{1}{\gamma _1}}+\frac{Y_0}{r}] = -\infty . \end{aligned}$$
Since w(z) is twice continuously differentiable at \(z={\tilde{z}}\) and constant for \(z\ge {\tilde{z}}\), we have
$$\begin{aligned} w^{\prime }(z) = w^{\prime \prime }(z) = 0 ~~\text{ for } z\ge {\tilde{z}}. \end{aligned}$$
(91)
Now we show that w is strictly decreasing and strictly convex for \(z \in (0, {\tilde{z}}]\). Note that
$$\begin{aligned} w'(z)&=n_+E_1z^{n_+-1}-L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}z^{-\frac{1}{\gamma _1}}+\frac{Y_0}{r},\\ w''(z)&=z^{-\frac{1}{\gamma _1}-1}\left[ n_+(n_+-1)E_1z^{n_++\frac{1}{\gamma _1}-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}\right] . \end{aligned}$$
Let
$$\begin{aligned} g(z):=n_+(n_+-1)E_1z^{n_++\frac{1}{\gamma _1}-1}+L_0^{\frac{\gamma _1- \gamma }{\gamma _1}}\frac{1}{\gamma _1 K_1}, \end{aligned}$$
then \(g(\cdot )\) is strictly decreasing since \(E_1<0\), so that
$$\begin{aligned} g(z) > g({\tilde{z}}) = 0, ~~\text{ for }~~ 0< z < {\tilde{z}}, \end{aligned}$$
where the equality holds since \(w''({\tilde{z}}) = 0\). Thus \(w''(z) > 0\) for \(0<z<{\tilde{z}}\). That is, w is strictly convex for \(z \in (0, {\tilde{z}}]\). Since \(w''(z) > 0\) for \(0<z<{\tilde{z}}\), \(w^{\prime }(z)\) is strictly increasing for \(0<z\le {\tilde{z}}\). Therefore, we have
$$\begin{aligned} w'(z)< w'({\tilde{z}}) = 0. \end{aligned}$$
Thus, w is strictly decreasing for \(z \in (0, {\tilde{z}}]\). \(\square \)
H Proof of the assertion that \({\bar{x}}^b<0\) when the IBR is not greater than the LBR in Remark 9
Let us define
$$\begin{aligned} \varLambda \triangleq \frac{\frac{1-\gamma _1}{\gamma _1}+n_+}{K_1}+\frac{1-n_+}{r}>0,~~\varGamma \triangleq \frac{L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}}{L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}-L_1^{ \frac{\gamma _1-\gamma }{\gamma _1}}}\frac{1-\gamma _1}{\gamma _1K_1}>0, \end{aligned}$$
and rewrite the threshold \({\bar{x}}^b\) in Shim and Shin [16] as
$$\begin{aligned} {\bar{x}}^b=\left( -\frac{n_-}{n_+-n_-}\varLambda +\varGamma \right) (Y_1-Y_0)-\frac{Y_1}{r}. \end{aligned}$$
(92)
The inequality \(Y_1/Y_0\le \varPsi \) is then equivalent to
$$\begin{aligned} (Y_1-Y_0)\le \frac{n_+-1}{-n_-\varLambda +\frac{n_+\gamma _1-\gamma _1+1}{ \gamma _1}\varGamma }\frac{Y_1}{r}. \end{aligned}$$
(93)
According to (92) and (93), we obtain
$$\begin{aligned} {\bar{x}}^b\le \left( \frac{-\frac{n_-}{n_+-n_-}\varLambda +\varGamma }{-\frac{n_-}{n_+-n_-} \varLambda +\frac{(n_+-1)\gamma _1+1}{(n_+-1)\gamma _1}\varGamma }-1\right) \frac{Y_1}{r}<0. \end{aligned}$$
\(\square \)
I Proof of Proposition 1
Note that \(({\mathbf {c}}^{b}, \varvec{\pi }^{b})\) without the borrowing constraints is given by (see Shim and Shin [16])
$$\begin{aligned} c_t^b=\left\{ \begin{array}{ll} L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( z_t^{b,\lambda _1}\right) ^{-\frac{1}{\gamma _1}}, &{}\quad \text{ for } -{Y_1}/{r}<X_t<{\bar{x}}^b,\\ L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\left( z_t^{b,\lambda _0}\right) ^{-\frac{1}{\gamma _1}}, &{}\quad \text{ for } X_t\ge {\bar{x}}^b, \end{array}\right. \end{aligned}$$
and
$$\begin{aligned} \pi _t^b=\left\{ \begin{array}{ll} \frac{\theta }{\sigma }\left\{ n_-(n_--1)D^b_2\left( z_t^{b,\lambda _1}\right) ^{n_--1}+L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1K_1}\left( z_t^{b,\lambda _1}\right) ^{-\frac{1}{\gamma _1}}\right\} , &{}\quad \text{ for } {Y_1}/{r}<X_t<{\bar{x}}^b,\\ \frac{\theta }{\sigma }\left\{ n_+(n_+-1)C^b_1\left( z_t^{b,\lambda _0}\right) ^{n_+-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1K_1}\left( z_t^{b,\lambda _0}\right) ^{-\frac{1}{\gamma _1}}\right\} ,&{}\quad \text{ for } X_t\ge {\bar{x}}^b, \end{array}\right. \end{aligned}$$
where
$$\begin{aligned} {\bar{x}}^b&=-n_-D_2^b{\bar{z}}^{n_--1}+L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}{\bar{z}}^{-\frac{1}{\gamma _1}}-\frac{Y_1}{r},\\&=\left[ \frac{\frac{-n_+n_-\gamma _1-n_-+n_-\gamma _1}{\gamma _1K_1}+\frac{n_+n_--n_-}{r}}{n_+-n_-}+\frac{L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}}{L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}-L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}}\frac{1-\gamma _1}{\gamma _1K_1}\right] (Y_1-Y_0)-\frac{Y_1}{r},\\ C^b_1&=D_2{\bar{z}}^{n_--n_+}+\left( L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}-L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\right) \frac{\gamma _1}{(1-\gamma _1)K_1}{\bar{z}}^{-\frac{1-\gamma _1}{\gamma _1}-n_+}+\frac{Y_1-Y_0}{r}{\bar{z}}^{1-n_+}>0,\\ D^b_2&=D_2>0, \end{aligned}$$
and \(z_t^{b,\lambda _0}\) and \(z_t^{b,\lambda _1}\), respectively, have a one-to-one correspondence with \(X_t\) by the algebraic equations
$$\begin{aligned} X_t=-n_+C^b_1\left( z_t^{b,\lambda _0}\right) ^{n_+-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}\left( z_t^{b,\lambda _0}\right) ^{-\frac{1}{\gamma _1}}-\frac{Y_0}{r}, ~~\text{ for } X_t\ge {\bar{x}}^b, \end{aligned}$$
(94)
and
$$\begin{aligned} X_t=-n_-D^b_2\left( z_t^{b,\lambda _1}\right) ^{n_--1}+L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}\left( z_t^{b,\lambda _1}\right) ^{-\frac{1}{\gamma _1}}-\frac{Y_1}{r}, ~~\text{ for } -{Y_1}/{r}<X_t<{\bar{x}}^b, \end{aligned}$$
(95)
respectively. Since \(D_1<0\), it is straightforward to see that \({\bar{x}}^b<{\bar{x}}\) and \(C_1<C_1^b\).
Consider the case where \(X_t\ge {\bar{x}}\). Recall that \({\mathcal {X}}(z) = -v^{\prime }(z)\) in (31) is a decreasing function of z with \(X_t = {\mathcal {X}}({\mathcal {Z}}(X_t))\) by (32). Note that \(X_t\) is a decreasing function of \(z_t^{b,\lambda _0}\) in (94). That is, \(X_t\) can be written as the decreasing functions of \({\mathcal {Z}}(X_t)\) and \(z_t^{b,\lambda _0}\):
$$\begin{aligned} X_t&=-n_+C^b_1\left( z_t^{b,\lambda _0}\right) ^{n_+-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}} \frac{1}{K_1}\left( z_t^{b,\lambda _0}\right) ^{-\frac{1}{\gamma _1}}-\frac{Y_0}{r}\nonumber \\&=-n_+C_1\left( {\mathcal {Z}}(X_t)\right) ^{n_+-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}} \frac{1}{K_1}\left( {\mathcal {Z}}(X_t)\right) ^{-\frac{1}{\gamma _1}}-\frac{Y_0}{r}. \end{aligned}$$
(96)
Thus, since \(C_1<C_1^b\), we have
$$\begin{aligned} z_t^{b,\lambda _0}<{\mathcal {Z}}(X_t) ~ \text{ for } X_t\ge {\bar{x}}, \end{aligned}$$
(97)
which implies the inequality \(c_t^*<c_t^{b}\) for \(X_t\ge {\bar{x}}\).
Multiplying both sides of the equality (96) by \((n_+-1)\), then we obtain
$$\begin{aligned} -n_+&(n_+-1)C^b_1\left( z_t^{b,\lambda _0}\right) ^{n_+-1}+L_0^{\frac{\gamma _1-\gamma }{ \gamma _1}}\frac{n_+-1}{K_1}\left( z_t^{b,\lambda _0}\right) ^{-\frac{1}{\gamma _1}}\\&\quad =-n_+(n_+-1)C_1\left( {\mathcal {Z}}(X_t)\right) ^{n_+-1}+L_0^{\frac{ \gamma _1-\gamma }{\gamma _1}}\frac{n_+-1}{K_1}\left( {\mathcal {Z}}(X_t)\right) ^{-\frac{1}{\gamma _1}}, \end{aligned}$$
and
$$\begin{aligned} -n_+&(n_+-1)C^b_1\left( z_t^{b,\lambda _0}\right) ^{n_+-1}-L_0^{\frac{\gamma _1-\gamma }{ \gamma _1}}\frac{1}{\gamma _1K_1}\left( z_t^{b,\lambda _0}\right) ^{-\frac{1}{\gamma _1}}+L_0^{ \frac{\gamma _1-\gamma }{\gamma _1}}\frac{\frac{1}{\gamma _1}+n_+-1}{K_1}\left( z_t^{b,\lambda _0} \right) ^{-\frac{1}{\gamma _1}}\\&\quad =-n_+(n_+\!-1\!)C_1\left( {\mathcal {Z}}(X_t)\right) ^{n_+-1}\!-\!L_0^{\frac{\gamma _1- \gamma }{\gamma _1}}\frac{1}{\gamma _1K_1}\left( {\mathcal {Z}}(X_t)\right) ^{-\frac{1}{ \gamma _1}}\!+\!L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{\frac{1}{\gamma _1}+n_+-1}{K_1} \left( {\mathcal {Z}}(X_t)\right) ^{-\frac{1}{\gamma _1}}. \end{aligned}$$
Thus, we get
$$\begin{aligned} \pi _t^b-\pi _t^*&=\frac{\theta }{\sigma }\left[ L_0^{\frac{\gamma _1-\gamma }{\gamma _1}} \frac{\frac{1}{\gamma _1}+n_+-1}{K_1}\left\{ \left( z_t^{b,\lambda _0}\right) ^{- \frac{1}{\gamma _1}}-\left( {\mathcal {Z}}(X_t)\right) ^{-\frac{1}{\gamma _1}}\right\} \right] >0, \end{aligned}$$
where the inequality holds due to (20) and (97).
Similarly, we can show that \(c_t^*<c_t^{b}~ \text{ and } ~ \pi _t^*<\pi _t^b\) for \(0\le X_t < {\bar{x}}^b\). \(\square \)
J Proof of Proposition 2
As mentioned in Remark 9, \({\bar{x}}^b < 0\) in this case. Recall that \({\mathcal {X}}(z) = -w^{\prime }(z)\) in (40) is a decreasing function of z with \(X_t = {\mathcal {X}}({\mathcal {Z}}(X_t))\) by (41). Note that \(X_t\) is a decreasing function of \(z_t^{b,\lambda _0}\) in (94). That is, \(X_t\) can be written as the decreasing functions of \({\mathcal {Z}}(X_t)\) and \(z_t^{b,\lambda _0}\), for \(X_t\ge 0 > {\bar{x}}^b\):
$$\begin{aligned} X_t&=-n_+E_1\left( {\mathcal {Z}}(X_t)\right) ^{n_+-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}} \frac{1}{K_1}\left( {\mathcal {Z}}(X_t)\right) ^{-\frac{1}{\gamma _1}}-\frac{Y_0}{r}\nonumber \\&=-n_+C^b_1\left( z_t^{b,\lambda _0}\right) ^{n_+-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}} \frac{1}{K_1}\left( z_t^{b,\lambda _0}\right) ^{-\frac{1}{\gamma _1}}-\frac{Y_0}{r}. \end{aligned}$$
(98)
Thus, since \(E_1<0<C_1^b\), we have
$$\begin{aligned} z_t^{b,\lambda _0}<{\mathcal {Z}}(X_t)~ \text{ for } X_t\ge 0, \end{aligned}$$
(99)
which implies the inequality \(c_t^*<c_t^{b}\) for \(X_t\ge 0\).
Multiplying both sides of the equality (98) by \((n_+-1)\), then we obtain
$$\begin{aligned}&-n_+(n_+-1)E_1\left( {\mathcal {Z}}(X_t)\right) ^{n_+-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}} \frac{n_+-1}{K_1}\left( {\mathcal {Z}}(X_t)\right) ^{-\frac{1}{\gamma _1}}\\&\quad =-n_+(n_+-1)C_1^b\left( z_t^{b,\lambda _0}\right) ^{n_+-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}} \frac{n_+-1}{K_1}\left( z_t^{b,\lambda _0}\right) ^{-\frac{1}{\gamma _1}}, \end{aligned}$$
and
$$\begin{aligned}&-n_+(n_+-1)E_1\left( {\mathcal {Z}}(X_t)\right) ^{n_+-1}-L_0^{\frac{\gamma _1-\gamma }{\gamma _1}} \frac{1}{\gamma _1K_1}\left( {\mathcal {Z}}(X_t)\right) ^{-\frac{1}{\gamma _1}}+L_0^{\frac{\gamma _1-\gamma }{ \gamma _1}}\frac{\frac{1}{\gamma _1}+n_+-1}{K_1}\left( {\mathcal {Z}}(X_t)\right) ^{- \frac{1}{\gamma _1}}\\&\quad =-n_+(n_+-1)C_1^b\left( z_t^{b,\lambda _0}\right) ^{n_+-1}-L_0^{\frac{\gamma _1-\gamma }{\gamma _1}} \frac{1}{\gamma _1K_1}\left( z_t^{b,\lambda _0}\right) ^{-\frac{1}{\gamma _1}}+L_0^{\frac{\gamma _1-\gamma }{ \gamma _1}}\frac{\frac{1}{\gamma _1}+n_+-1}{K_1}\left( z_t^{b,\lambda _0}\right) ^{-\frac{1}{\gamma _1}}. \end{aligned}$$
Thus we get
$$\begin{aligned} \pi _t^b-\pi _t^*&=\frac{\theta }{\sigma }\left[ L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{ \frac{1}{\gamma _1}+n_+-1}{K_1}\left\{ \left( z_t^{b,\lambda _0}\right) ^{-\frac{1}{\gamma _1}}- \left( {\mathcal {Z}}(X_t)\right) ^{-\frac{1}{\gamma _1}}\right\} \right] >0, \end{aligned}$$
where the inequality holds due to (20) and (99). \(\square \)
K Proof of Proposition 3
Recall that \(X_t = {\mathcal {X}}({\mathcal {Z}}(X_t)) = -v^{\prime }({\mathcal {Z}}(X_t))\). That is, for \(0<X_t<{\bar{x}}\),
$$\begin{aligned} X_t = -n_+D_1\left( {\mathcal {Z}}(X_t)\right) ^{n_+-1}-n_-D_2\left( {\mathcal {Z}}(X_t) \right) ^{n_--1}+L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}\left( { \mathcal {Z}}(X_t)\right) ^{-\frac{1}{\gamma _1}}-\frac{Y_1}{r}.\nonumber \\ \end{aligned}$$
(100)
Since \(D_1<0\) and \(D_2>0\) we have \({\mathcal {Z}}(X_t)>z_t^{M_1}\), which yields \(c_t^*<c_t^{M_1}\) for \(0<X_t<{\bar{x}}\). On the other hand, we can obtain, for \(0<X_t<{\bar{x}}\),
$$\begin{aligned} \frac{\sigma }{\theta }\left( \pi _t^*-\pi _t^{M_1}\right) =(n_+-1)\frac{Y_1}{r}-F \left( {\mathcal {Z}}(X_t)\right) -\left( \frac{1}{\gamma _1}+n_+-1\right) L_1^{ \frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}\left( z_t^{M_1}\right) ^{-\frac{1}{\gamma _1}}, \end{aligned}$$
if we exploit (35) along with (100), and (42) with (43), where the function F is defined as (71). Since \({\mathcal {Z}}(X_t)\) and \(L_1^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}\left( z_t^{M_1}\right) ^{-\frac{1}{\gamma _1}}\) go to \({\tilde{z}}\) and \(Y_1/r\), respectively, as \(X_t\) goes to 0, and \(F\left( {\tilde{z}}\right) =0\), we have
$$\begin{aligned} \lim _{X_t\downarrow 0}\frac{\sigma }{\theta }\left( \pi _t^*-\pi _t^{M_1}\right) =\left[ (n_+-1)\frac{Y_1}{r}-F\left( {\tilde{z}}\right) -\left( \frac{1}{\gamma _1}+n_+-1\right) \frac{Y_1}{r}\right] =-\frac{1}{\gamma _1}\frac{Y_1}{r}<0, \end{aligned}$$
\(\square \)
L Proof of Proposition 4
If \(C_1>0\), we see that \({\mathcal {Z}}(X_t)<z_t^{M_0}\) for \(X_t\ge {\bar{x}}\), so we easily obtain \(c_t^*>c_t^{M_0}\) for \(X_t\ge {\bar{x}}\). We also have
$$\begin{aligned} \pi _t^*&=\frac{\theta }{\sigma }\left\{ n_+(n_+-1)C_1\left( {\mathcal {Z}}(X_t)\right) ^{n_+-1}+L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{\gamma _1K_1}\left( {\mathcal {Z}}(X_t)\right) ^{-\frac{1}{\gamma _1}}\right\} \\&>\frac{\theta }{\sigma \gamma _1}L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}\left( {\mathcal {Z}}(X_t)\right) ^{-\frac{1}{\gamma _1}} >\frac{\theta }{\sigma \gamma _1}L_0^{\frac{\gamma _1-\gamma }{\gamma _1}}\frac{1}{K_1}\left( z_t^{M_0}\right) ^{-\frac{1}{\gamma _1}}=\pi _t^{M_0}, \end{aligned}$$
for \(X_t\ge {\bar{x}}\). It is straightforward to see that \(c_t^*<c_t^{M_0},~\pi _t^*<\pi _t^{M_0}\) for \(X_t\ge {\bar{x}}\) given \(C_1<0\). \(\square \)