Abstract
We study the optimal capital budgeting policy of a firm taking into account the choice between internal and external financing. The manager can dedicate effort either to increase short-term profitability, thus generating greater immediate cash-flow, or to improve long-term perspectives. When both types of effort are observable, low productivity firms end up using internal funds, while high productivity firms use external capital markets. When effort to boost short-term cash flow is observable, while effort to boost long-term profitability is not, non-monotonic policies may be optimal. In such cases financing switches back and forth between internal and external funds as the quality of the project increases.
Similar content being viewed by others
Notes
We will show however that the optimal contract does not have large fines, but this is a result rather than an assumption.
Ironing techniques were introduced by Myerson (1981) in his seminal article.
It can be proved that a sufficient condition to make the interval empty (i.e. \(e_{p} \left( \theta \right) >0\) is optimal for each \(\theta \)) is that \( \alpha < 0.5\).
References
Almazan, A., Chen, Z., Titman, S.: Firm investment and stakeholder choices: a top-down theory of capital budgeting. J Finance 72(5), 2179–228 (2017)
Almeida, H., Campello, M.: Financing frictions and the substitution between internal and external funds. J Financ Quant Anal 45(3), 589–622 (2010)
Antle, R., Eppen, G.: Capital rationing and organizational slack in capital budgeting. Manag Sci 31(3), 163–174 (1985)
Begenau, J., Palazzo, B.: Firm selection and corporate cash holdings. NBER working paper 23249 (2017). http://www.nber.org/papers/w23249
Bernardo, A., Cai, H., Luo, J.: Capital budgeting and compensation with asymmetric information and moral hazard. J Financ Econ 61(3), 311–344 (2001)
Bernardo, A., Cai, H., Luo, J.: Capital budgeting in multi-division firms: information, agency, and incentives. Rev Financ Stud 17, 739–767 (2004)
Bernardo, A., Luo, J., Wang, J.: A theory of socialistic internal capital markets. J Financ Econ 80, 485–509 (2006)
Bernardo, A., Cai, H., Luo, J.: Motivating entrepreneurial activity in a firm. Rev Financ Stud 22(3), 1089–1118 (2009)
Brusco, S., Panunzi, F.: Reallocation of corporate resources and managerial incentives in internal capital markets. Eur Econ Rev 49, 659–681 (2005)
Gertner, R., Scharfstein, D., Stein, J.: Internal versus external capital markets. Q J Econ 109(4), 1211–30 (1994)
Harris, M., Raviv, A.: The capital budgeting process: incentives and information. J Finance 51(3), 1139–1174 (1996)
Harris, M., Raviv, A.: Capital budgeting and delegation. J Financ Econ 50(3), 259–289 (1998)
Harris, M., Kriebel, C., Raviv, A.: Asymmetric information, incentives and intrafirm resource allocation. Manag Sci 28, 604–620 (1982)
Holmstrom, B., Milgrom, P.: Multitask principal-agent analyses: incentive contract, asset ownership and job design. J Law Econ Organ 7(1), 24–52 (1991)
Inderst, R., Klein, M.: Innovation, endogenous overinvestment, and incentive pay. Rand J Econ 38(4), 881–904 (2007)
Kadan, O., Reny, P., Swinkels, J.: Existence of optimal mechanisms in principal-agent problems. Econometrica 85(3), 769–823 (2017)
Laffont, J.-J., Tirole, J.: A Theory of Incentives in Procurement and Regulation. MIT Press, Cambridge (1993)
Myers, S., Majluf, N.: Corporate financing and investment decisions when firms have information that investors do not have. J Financ Econ 13(2), 187–221 (1984)
Myerson, R.: Optimal auction design. Math Oper Res 6, 58–73 (1981)
Rajan, R., Servaes, H., Zingales, L.: The cost of diversity: the diversification discount and inefficient investment. J Finance 55(1), 35–80 (2000)
Scharfstein, D., Stein, J.: The dark side of internal capital markets: divisional rent-seeking and inefficient investment. J Finance 55(6), 2537–64 (2000)
Shin, H.H., Stulz, R.: Are internal capital markets efficient? Q J Econ 113(2), 531–52 (1998)
Stein, J.: Internal capital markets and the competition for corporate resources. J Finance 52(1), 111–33 (1997)
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This paper was partly written while Brusco was visiting the Department of Economics of the Hong Kong University of Science and Technology. We would like to thank seminar participants at the Singapore Management University, National University of Singapore, Deakin University, the University of Hong Kong, the Asian Econometric Society 2017, the 2018 Decentralization Conference and the 2018 Québec Political Economy Conference.
Appendices
Appendix I
Proof of Proposition 1
Consider problem (2). Since \( \alpha \in \left( 0,1\right) \) and \( \gamma \in \left( 0,1\right) \) the marginal return of \(e_p\) and k at zero goes to infinity, so the non-negativity constraints for those two variables can be ignored. Since the function is differentiable, the solution must satisfy the first order conditions of the Lagrangian
and the complementary slackness conditions \(\lambda e_c=0\), \(e_c \ge 0\) and \( \lambda \ge 0\).
The first order conditions are
If the solution is such that \(\lambda = 0\) then \(e_c + e_p =1 \) and (16) implies \( k = \alpha ^{-\frac{1}{\gamma }} e_p^{\frac{1-\alpha }{\gamma }}\). Substituting k into (15) we obtain the equation:
Since \(\alpha +\gamma <1\), the left hand side is strictly decreasing in \(e_{p}\) and it goes from \(+\infty \) to 0 as \(e_{p}\) moves from 0 to \( +\infty \). The right hand side is strictly positive so there is a unique solution. Call \( \hat{e}_p \left( \theta \right) \) the solution and observe, by inspection, that it is strictly increasing in \(\theta \). It is feasible if \( \hat{e}_{c}\left( \theta \right) =1-\hat{e}_{p}\left( \theta \right) \ge 0 \) or \(\hat{e}_{p} \left( \theta \right) \le 1\). There is a unique value \(\theta ^{+}\) such that \(e_{p}^{*}\left( \theta ^{+} \right) = 1\), which is given by
Thus, the solution obtained for \(\lambda = 0\) is acceptable if \( \theta \le \theta ^{+}\).
Suppose now that \(\lambda >0\), so that \(e_c = 0\) and \(e_p > 1\). In this case (16) implies \( k = \alpha ^{-\frac{1}{\gamma }} e_p^{\frac{2-\alpha }{\gamma }} \). Substituting k into (15) we obtain
Equation (21) has a unique solution, call it \(\overline{e}_{p}\left( \theta \right) \), which is strictly increasing in \(\theta \) and it satisfies \(\overline{e}_{p} \left( \theta ^+\right) = 1\). Thus, the solution is feasible only if \( \theta >\theta ^{+} \). We conclude that for each \(\theta \) there is only one stationary point. Let \( e^{*}_p \left( \theta \right) \) be such solution, i.e. \( e^{*}_p \left( \theta \right) = \hat{e}_p \left( \theta \right) \) if \(\theta \le \theta ^{+} \) and \( e^{*}_p \left( \theta \right) = \overline{e}_p \left( \theta \right) \) if \(\theta > \theta ^{+} \).
To show that the stationary point is a constrained maximum, observe that the objective function is strictly concave in \( \left( k,e_p\right) \) for each \(e_c\) (the objective function is not concave in the three variables \( \left( k,e_p,e_c\right) \)). This can be checked looking at the Hessian matrix
The values on the main diagonal are negative and the determinant is positive under the assumption \(\gamma +\alpha <1\). Thus, for any fixed \( e_c\) the unique stationary point is a maximum. Let \( k\left( e_c,\theta \right) \) and \( e_p \left( e_c,\theta \right) \) be the points at which the maximum is attained and define
By inspection, \( \lim _{e_c \rightarrow + \infty } F\left( e_c,\theta \right) = - \infty \). Consider now the problem
with Lagrangian
The envelope theorem implies
so the first order and complementary slackness conditions are
Suppose first \(\theta > \theta ^{+} \). In that case we established that \( \left. \frac{dF }{d e_c} \right| _{e_c=0} <0 \) and there is no point \(\theta > \theta ^+\) at which \( \left( e_c+e_p\left( e_c, \theta \right) \right) =1\). We conclude that F is always decreasing and the maximum is achieved at \(e_c =0\).
If instead \( \theta < \theta ^+\) then we established \( \left. \frac{dF }{d e_c} \right| _{e_c=0} > 0 \), so that the solution must involve \( e_c >0\). Since there is a unique stationary point at which \( e_c>0\) and the objective function goes to \( - \infty \) as \( e_c\) gets large, the point must be a maximum. \(\square \)
Proof of Proposition 2
Existence of a solution follows from the results in Kadan et al. (2017). Given an optimal mechanism \( w \left( V,\hat{\theta }\right) , k \left( \hat{\theta }\right) , e_c \left( \hat{\theta }\right) , e_p \left( \hat{\theta }\right) \), define \( w^e \left( \theta , \hat{\theta }, e_p\right) \), \( U\left( \theta , \hat{\theta }, e_p \right) \), \( U\left( \theta \right) \) as in main text and \( w^e \left( \theta \right) \equiv w^e \left( \theta , \theta , e_p \left( \theta \right) \right) \).
Observe that
can be written, using the change of variable \(v=\left( \theta +e_{p}^{\alpha }\right) k^{\gamma }\left( \widehat{\theta }\right) u\), as
so \( w^e \left( \theta , \hat{\theta }, e_p\right) \) is differentiable with respect to \(\theta \) and \(e_p\). The derivative with respect to \(\theta \) is
The integrals in (25) and (26) must be well defined in an optimal mechanism in order to ensure that the maximization problem for the agent is solvable.
\(\underline{\mathbf{Step 1: } U\left( \theta \right) \hbox {is continuous}}\). Suppose not. Then there is a sequence \(\left\{ \theta _n \right\} \) converging to a value \(\theta ^*\) and \(\varepsilon >0\) such that \( \left| U \left( \theta ^*\right) - U \left( \theta _n\right) \right| > \varepsilon \) each n. This in turn implies
Continuity of \(w^e\) in its first argument implies that there is N such that, for \(n>N\) we have
In turn, incentive compatibility implies
and
Summing the two inequalities side by side we obtain
which can be written as
Since \(w^{e} \left( \theta ,\hat{\theta },e_p \right) \) is continuous in \(\theta \), both terms on the LHS go to zero as n goes to infinity, a contradiction. This establishes the continuity of \(U \left( \theta \right) \).
\(\underline{\mathbf{Step 2: } U \left( \theta \right) \hbox {is non-decreasing}.}\) Since the mechanism maximizes expected profit, it should not be possible to implement the same allocation at a lower expected cost. We will show that if U is strictly decreasing on an interval \(\left( \theta _a, \theta _b\right) \) then it is possible to offer a strictly lower expected salary and implement the same allocation, thus contradicting the optimality of the mechanism.
Individual rationality implies \(U\left( \theta _a \right) > U\left( \theta _b \right) \ge 0\). Since U is continuous and strictly decreasing on \(\left( \theta _a, \theta _b\right) \), it is almost everywhere differentiable on that interval. Let
where \( U'_{-}\left( \theta \right) \) and \(U'_{+}\left( \theta \right) \) are the left and right derivative of U. The interval \(\left( \theta _a, \theta _b\right) \) can be chosen so that \( \delta < 0\).
For each positive integer n define the function
where \( V=\left( \theta + e^{\alpha }_p\right) k \left( \hat{\theta }\right) u \). Consider now polynomial functions of the form
When \( e_p = 0 \) we have \( {\mathbb {E}} \left[ q_n \left( V,\hat{\theta }\right) \right] = \theta ^n \). This implies that we can build any polynomial function
By the Stone–Weierstrass theorem any continuous function on the interval \( \left[ \underline{\theta },\overline{\theta }\right] \) can be approximated to any degree desired by a polynomial function. Thus, we can define a function
such that
The derivative of \(\eta \) on the interval \( \left( \theta _a, \theta _b \right) \) is given by \( \eta ' \left( \theta \right) = \theta _a + \theta _b - 2 \theta \). Now consider a new mechanism which is identical to the initial one except that the wage function is replaced by
where \(\varepsilon >0 \) is chosen to be sufficiently small so that \( \varepsilon \eta ' \left( \theta \right) > \delta \) for each \(\theta \in \left( \theta _a, \theta _b\right) \). This implies that \( U \left( \theta \right) \) remains strictly decreasing on \( \left( \theta _a, \theta _b \right) \).
The new mechanism gives the same incentives as the previous one. In particular notice that, since \({\mathbb {E}} \left[ \eta \left( V,\hat{\theta } \right) \right] \) does not depend on \(\hat{\theta }\), truth-telling must remain optimal. Also, the choice of \(e_p\) remains optimal and in particular it remains optimal to select \(e_p=0\) in the interval \(\left( \theta _a,\theta _b\right) \).
We conclude that the new mechanism satisfies both incentive compatibility and individual rationality and it implements the same allocation at a strictly lower cost, contradicting the optimality of the original scheme. Thus, \(U \left( \theta \right) \) is non-decreasing.
\(\underline{\mathbf{Step 3: } \hbox {A formula for } U' \left( \theta \right) \hbox {when } e_p \left( \theta \right) >0}\). Since \( U \left( \theta \right) \) is continuous and non-decreasing on the compact interval \(\left[ \underline{\theta },\overline{\theta }\right] \) it is differentiable almost everywhere. By the envelope theorem the derivative must be
Using the expression of \(w^{e}\left( \theta ,\widehat{\theta } ,e_{p} \right) \) given in (25) and the assumption that the density \(f \left( u\right) \) is differentiable it is immediate to see that whenever \(e_p > 0\) we have
Suppose now that at \(\theta \) the project improving effort is \(e_{p}\left( \theta \right) >0\). A necessary condition for implementability is that \( e_{p}\left( \theta \right) \) maximizes the expected utility of the manager when the true \(\theta \) is reported, i.e.
Since \(w^{e}\) is differentiable, a necessary condition for optimality is
where we made use of (27). This implies
whenever \(e_{p}\left( \theta \right) >0\).
\(\underline{\mathbf{Step 4: } \frac{\partial w^e}{\partial \theta } =0 \hbox { when } e_p \left( \theta \right) =0}\). Suppose the optimal policy prescribes \( e_p \left( \theta \right) = 0 \) on an interval \(\left( \theta _a,\theta _b\right) \). A necessary condition for \(e_{p}=0\) to be an optimal choice is
Since \(e_{p}^{\alpha -1}\) goes to \(+\infty \) as \(e_{p}\) goes to zero, inequality (29) implies \(\frac{\partial w^{e}\left( \theta ,\theta ,e_{p} \right) }{\partial \theta }\le 0\). Suppose the inequality is strict on \(\left( \theta _a,\theta _b\right) \). Since \( U \left( \theta \right) \) is non-decreasing, it must be the case that \( e_c \left( \theta \right) \) is strictly decreasing.
Furthermore, it must be \(e_c \left( \theta \right) \le 1 \). In fact, since \(e_c \left( \theta \right) \) is strictly decreasing on the interval, it must be \(e_c \left( \theta \right) <1 \) almost everywhere. If not, reducing \(e_c\) by \(\varepsilon < 2\left( e_c-1\right) \) and decreasing the wage by \(\varepsilon ' = e_c \varepsilon - 0.5 \varepsilon ^2\) would leave the manager with the same utility, would not affect incentive compatibility and it would yield a higher profit. This is sub-optimal. Increasing the effort \(e_c\) by \( \varepsilon \) and increasing the compensation by \( \varepsilon ' = 0.5 \varepsilon ^2 +e_c \varepsilon \) in order to keep utility constant results in a higher profit for \(\varepsilon \) sufficiently small. The change does not violate incentive compatibility.
\(\underline{\mathbf{Step 5: } \frac{\partial w^e \left( \theta , \hat{\theta }, e_p \right) }{\partial \theta } \hbox { is absolutely continuous wrt } \theta }\). A sufficient condition for absolute continuity of a function over a compact interval is that the derivative is bounded. Equation (28) and the fact that \(\frac{\partial w^e}{\partial \theta } =0 \) when \(e_p \left( \theta \right) =0\) imply that the expression (28) holds both when \(e_p \left( \theta \right) >0 \) and when \(e_p \left( \theta \right) =0 \). Thus, \( \frac{\partial w^e}{\partial \theta } \) is bounded below by zero and above by
which must be bounded in any optimal mechanism.
\(\underline{\mathbf{Step 6: } \hbox {The partial derivative of} w^{e} \hbox {is non-decreasing}}\). Since \(w^e \left( \theta , \hat{\theta }, e_p \right) \) is differentiable with respect to \(\theta \), for each pair \( \left( \hat{\theta },e_p\right) \) we can write
Incentive compatibility implies
and
Now set \(\hat{\theta }= \theta +\varepsilon \). Summing (31) and (32) side by side and using expression (30) for \(w^e \left( \theta ,\hat{\theta },e_p\right) \) we obtain
for each \(\varepsilon > 0\).
Define the functions
and suppose that there is an interval \( \left( \theta _a,\theta _b\right) \) such that \( h \left( \theta _b \right) < h\left( \theta _a\right) \). Define
and notice that \(\kappa < 0\). Notice also that for each value \( \delta _h < \theta _b - \theta _a \) we can find an interval \( \left( \theta '_a, \theta '_b \right) \subset \left( \theta _a, \theta _b \right) \) with the properties that \( \theta '_b - \theta '_a < \delta _h\) and \( \frac{h\left( \theta '_b\right) - h\left( \theta '_a \right) }{\theta '_b - \theta '_a} \le k\).
For each \(\theta \) the function \( h \left( s,\theta \right) \) is absolutely continuous in s. This means that for each \( \varepsilon \) there is \( \delta \) such that \( \left| s - \theta \right| < \delta \) implies \( \left| h(s,\theta ) - h\left( \theta \right) \right| < \varepsilon \).
Now pick an interval \( \left( \theta '_a, \theta _b' \right) \) such that \( \frac{h\left( \theta '_b\right) - h\left( \theta '_a \right) }{\theta '_b - \theta '_a} \le \kappa \) and for each \( s \in \left( \theta _a', \theta _b'\right) \), \(\left| s- \theta \right| < \theta _b' - \theta _a' \) implies \( \left| h(s,\theta _a) - h \left( \theta _a\right) \right| < \frac{\left| \kappa \right| }{4} \) and \( \left| h(s,\theta _b) - h \left( \theta _b\right) \right| < \frac{\left| \kappa \right| }{4} \).
a contradiction. We conclude that we must have
whenever \(\theta ' > \theta \).
At last, notice that this implies that \( U' \left( \theta \right) \) is non-decreasing. \(\square \)
Proof of Proposition 3
In any optimal mechanism the expression \(U^{\prime }\left( \theta \right) =\frac{1}{\alpha } \Big (e_{c}\left( \theta \right) +e_{p}\left( \theta \right) \Big ) e_{p}^{1-\alpha }\left( \theta \right) \) must be non-decreasing. Thus, if \( U^{\prime }\left( \theta \right) \) is strictly positive at \(\theta \) it cannot be zero at any \(\theta ' > \theta \). It immediately follows that \( e_p \left( \theta \right) > 0\) implies \( e_p \left( \theta ' \right) > 0\) for each \(\theta ' > \theta \). Thus, there is a \(\theta ^*\) such that \(e_p=0\) on \(\left[ \underline{\theta }, \theta ^*\right) \) and \( e_p > 0\) for \( \theta \ge \theta ^*\).
If the interval \(e_p=0\) on \(\left[ \underline{\theta }, \theta ^*\right) \) is non-empty then no incentive rent is paid on that interval. Thus, the solution must be the same as in the complete information case under the constraint \(e_p = 0\).
At last, to see that \( \theta ^* < \overline{\theta } \) observe that at \(\overline{\theta }\) the solution must coincide with the complete information solution and that implies \( e_p \left( \overline{\theta }\right) > 0\). By standard arguments, this has to be true also on a left-neighborhood of \(\overline{\theta }\) of strictly positive measure. \(\square \)
Proof of Proposition 4
Assume that when \(e_c \ne e_c \left( \hat{\theta }\right) \) wage is zero, so that it is always optimal to pick \( e_c \left( \hat{\theta }\right) \). Consider the maximization problem of the manager when the wage function is given by (11). The expected utility when the true value of the parameter is \( \theta ,\) the announced value is \( \hat{\theta }\) and \(e_p\) is chosen is
where \( a\left( \widehat{\theta }\right) \) is given by (12). For each \( \hat{\theta }\) such that \(U'\left( \hat{\theta }\right) =0\), the value of \(e_p\) that maximizes expected utility is \(e_p=0\). Since \(U'\left( \hat{\theta }\right) =0\) is possible only if \(e_p\left( \hat{\theta }\right) =0\), this implies that expected utility is maximized at the value of \(e_p\) prescribed by the mechanism.
When \(U'\left( \hat{\theta }\right) >0 \) the optimal \(e_p\) must be obtained solving the first order condition
Using the definition of \(U'\left( \hat{\theta }\right) \), the condition can be written as
The LHS is strictly positive and constant in \(e_p\), while the RHS is strictly increasing in \(e_p\) and it goes from 0 to \(+\infty \). Thus there is a unique solution and, by inspection, the solution is \( e_p = e_p\left( \hat{\theta }\right) \).
With this result, we can now consider the problem of choosing the report \(\hat{\theta }\) for a given observed value \(\theta \). The expected utility can now be written as
Substituting the expression for \( a\left( \widehat{\theta }\right) \) from (12) we obtain
Since \( U'\left( \hat{\theta }\right) \) is non-decreasing, the expression is maximized at \( \widehat{\theta }=\theta \). Thus, the proposed wage function induces truthtelling, it induces the correct choice of efforts and it gives the manager expected utility \(\int _{\underline{\theta }}^{\theta }U' \left( s\right) ds\), i.e. the lowest incentive rent associated to a truthtelling mechanism. \(\square \)
Appendix II
In this appendix we describe how the optimal policy described in Sect. 6.2.1 is computed. The objective function is everywhere differentiable and for each pair \(\left( \theta ,e_{p}\right) \) the level of k that maximizes it is given by
where we used \(\gamma =\frac{1}{2}\), \(\alpha =\frac{1}{3}\). Using this fact, the objective function can be written as a function of \(e_{p}\) and \(e_{c}\) only:
Let \(x=e_{p}^{\frac{1}{3}}\) and \(y=e_{c}\). Then the objective function becomes
Now notice that the function is strictly concave in y for each \(\left( x,\theta \right) \) and that the optimal value of y for each given pair \( \left( x,\theta \right) \) is
At this point we can look at the objective function
and compute numerically the optimal value \(x^{*}\left( \theta \right) \) for each \(\theta \). The optimal policy is then given by
It can also be checked numerically that the expression
is in fact increasing in \(\theta \), as it is shown in the following picture (Fig. 3).
At last, in order to obtain the complete information solution, it is sufficient to set \(\mu \left( \theta \right) =0\) for each \(\theta \) in the problem stated above.
Rights and permissions
About this article
Cite this article
Brusco, S., Panunzi, F. Internal financing, managerial compensation and multiple tasks. Ann Finance 16, 501–527 (2020). https://doi.org/10.1007/s10436-020-00375-z
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10436-020-00375-z