A Unified Model of Investment Under Uncertainty Andrew B. Abel, Janice C. Eberly The American Economic Review, Volume 84, Issue $ (Dec., 1994), 1369-1384. Stable URL: hutp:/inks,jstor.org/sici?sici=0002-8282% 28 1994 12%2984%3A5%3C1369%3A AUMOIU%3E2.0,.CO%3B2-0 ‘Your use of the ISTOR archive indicates your acceptance of JSTOR’s Terms and Conditions of Use, available at hhup:/www.jstororg/about/terms.huml. JSTOR’s Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission, The American Economic Review is published by American Economic Association, Please contact the publisher for further permissions regarding the use of this work. Publisher contact information may be obtained at hup:/www.jstor.org/journals/aea.himl The American Economic Review (©1994 American Economic Association ISTOR and the ISTOR logo are trademarks of ISTOR, and are Registered in the U.S. Patent and Trademark Office. For more information on ISTOR contact jstor-info@umich.edv. ©2002 JSTOR upslwww jstor.org/ Wed an 16 12:33:51 2002A Unified Model of Investment Under Uncertainty By Anprew B. Apet AND Janice C. Eerty* This paper extends the theory of invest went under uncertainty to incorporate {fixed costs of investment, a wedge between the purchase price and sale price of capital, and potential irreversibility of inv investment is a nondecreasing function restment. In this extended framework, of q the shadow price of installed capital. The optimal rate of investment is in one of three regimes (positive, zer, ‘or negative gross investment), depending on the value of q relative to two critical values. In general however, the shadow price q is not directly observable, so we present two examples relating q to observable variables. (JEL E22) If a firm can instantaneously and cost lessly adjust its capital stock, then, as shown by Dale W. Jorgenson (1963), its decision about how much capital to use is essentially a static decision in which the marginal prod- uct of capital is equated to the user cost of capital. The firm’s investment decision be- comes an interesting dynamic problem, in which anticipations about the future cco- nomic environment affect current invest- ment, when frictions prevent instantaneous and costless adjustment of the capital stock. ‘The investment literature of the last three decades has focused on two types of fri tions: adjustment costs and irreversibility. In this paper, we present a simple, more general framework that encompasses irre- versibility as well as adjustment costs that may include a fixed component. Within this more general framework, the opti “Department of Finance, The Wharton School of the University. of Pennsivania, 10104-6367. The authors thank Avner Bar Tan, Brainard, James Foster, Michel Habib, and) partici ‘the University of Pennsyivania, Johas Hopkins University, New York University, North Carolina State. University, Princeton University, Rice University, Texas A&M University, the University of British: Columbia, the University of Chicago, the Unversity of Maryland, the University of Quebec at Montreal, the Uniersity of South Florida, the University of Texas, the University fof Washington, Vanderbilt University. and Yale Uni- ‘ersity for thelr comments. Financial support from the National Science Foundation and the Wharton Junior Faculty Revearch Fund is gratefully acknowledged 1309 ‘mal investment behavior of the firm poten- tially comprises three regimes: (i) a regime of positive gross investment; (i) @ regime of zero gross investment; and (ii) a regime of negative gross investment. Most of the adjustment-cost literature tends to focus, either implicitly or explicitly, on the first of these regimes. The irreversibility literature is more explicit in its recognition of regimes of positive gross investment and zero gross investment, and it rules out the regime of negative gross investment by assumption. ‘The more general model presented here allows a simple characterization of the con- tions giving rise to each of these regimes. In the adjustment-cost literature, based on the seminal work of Robert Eisner and Robert H, Strotz (1963), the adjustment-cost function is typically assumed to be strictly convex and to have a value of zero at zero investment. Although a few studies mention the possiblity of fixed costs (see Michael Rothschild, 1971; Stephen J. Nickell, 1978), there is virtually no formal analysis of these fixed costs. The model presented in this paper incorporates fixed costs. During the 1970's and 1980's, the adjust- ment-cost literature began to merge with the literature on Tobin’s q. James Tobin (1969) argued that the optimal rate of in- vestment is an increasing function of the ratio of the market value of the firm to the replacement cost of the firm’s capital—a ratio that he called q and that has come to be known as “average q.” Michael Mussa (1977) showed in a deterministic model, andBm THE AMERICAN ECONOMIC REVIEW Abel (1983) showed in a stochastic model, that the optimal rate of investment is the rate that equates the marginal adjustment cost with the marginal value of installed capital, a concept known as “marginal 4.” While average q is a potentially observable ‘number, it is marginal g that is relevant for investment decisions. Fumio Hayashi (1982) presented conditions under which average q and marginal q are equal, As indicated earlier, the assumption that investment is irreversible is another type of friction that makes the investment decision an interesting dynamic problem. In a semi nal paper on irreversibility, Kenneth J. Arrow (1968 pp. 8-9) argued that “there of capital goods cannot be accomplished at the same price as their purchase. ... For simplicity, we will make the extreme assumption that resale of capital goods is impossible, so that gross investment is constrained to be non-negative.” Arrow showed that, in a deterministic model, o mal investment behavior under irreversi ity will be characterized by alternating inter- vals of time corresponding to regimes of positive gross investment and regimes of zero gross investment. When the shadow rice of capital is smaller than the cost of new capital, the firm will have zero invest- ment; when the fim undertakes positive ‘gross investment, the shadow price of capi- tal equals the cost of new capital." We incorporate both adjustment costs and irreversibility in an extended model of adjustment costs. We note that ad- justment costs and irreversibility are ex- amined together in a deterministic mo- ‘he same relationship among sros investment, the shadow price of capital, and the price of new capital was derived ina stochastic general-quiibrium model by (980) Sima Be ‘behavior of an individual firm under uncertainty and find that the firm equates the marginal product of capital and the user cost of capital whenever it Is undertaking gross investment; when the firm is not investing, the marginal product of capital is below the DECEMBER 1994 del by Robert E. Lucas, Jr. (1981) and in a stochastic model by Lucas and Edward C. Prescott (1971). Curiously, both of these papers introduce the constraint that gross investment is nonnegative in the formal op- timization problem, yet neither paper com- ments on this assumption, nor does either paper use the term “irreversibility.” In ef- fect, these papers take as a postulate that gross investment cannot be negative. In con- trast, our model incorporates Arrow’s ob- servation that the resale price of capital may be below the price of new capital, and the model includes the special case in which the resale price is zero. If we were simply to postulate that gross investment cannot be negative, then it would be easy to impose irreversibility in an adjustment-cost_ framework by simply as- suming that infinite adjustment costs are incurred at any negative rate of investment, as in Caballero (1991 p. 281). Our approach avoids treating irreversibility as a postulate but rather allows for (and characterizes) ceases in which the optimal rate of invest- ment by the firm is never negative. We introduce an augmented adjustment-cost function that includes traditional convex ad~ justment costs, as well as the possibility of fixed costs and the possibility that the resale rice of capital goods is below their pur- chase price and may even be zero. In this augmented adjustment-cost framework, in- vestment is a nondecreasing function of the shadow price q, which is always positive. There are three regimes of optimal invest- ment behavior characterized by two critical values of 4, 4, = q2- Optimal gross invest- ment is positive for q > qz, zero for values of q between q, and q,, and negative for q 0, mx,¢(Ky€,) $07 and that £, evolves according to a diffusion process: () de, =ule,)de+o(e,) de where 2 is a standard Wiener process. Capital is acquired by undertaking gross investment at rate J, and the capital stock depreciates at a fixed proportional rate 5, so the capital stock evolves according to Q) dK, =(,-8K,) at. When the firm undertakes gross invest- ‘ment, it incurs costs that we can describe in This formulation of the profit function allows the firm tobe either a pricetaker or a price-setter. In the examples in Section I, the frm is assumed to be price taker. ABEL AND EBERLY: INVESTMENT UNDER UNCERTAINTY a terms of three components: (i) purchase or sale costs, (ii) costs of adjustment, and (ii) fixed costs per unit time. ( Purchase sale costs are the costs of buying or selling uninstalled capital. Let py be the price pet unit at which the firm can buy any amount of uninstalled capital, and let pg be the price per unit at which the firm can sell any amount of uninstalled cap- ital, We assume that pz 2 py 0. The sale price of capital may be strictly less than the Purchase price of capital if, for example, capital is firm-specific." “The purchase /sale cost function is pg for [>0 and pj for 1 <0. It is a (weakly) convex and nondecreasing function that takes the value zero when gross investment is zero. Note that the purchase /sale cost function is twice differentiable everywhere except possibly at 1 = 0. (i) Adjustment costs are nonnegative costs that attain their minimum value of zero when I= 0. As is typical in the adjust- ment-cost literature, we assume that adjust- ‘ment costs are continuous and strictly con- vex in 1° In some formulations, adjustment costs also depend on the capital stock K, with the "alternatively, the sale of capital may be less than its purchase prise if there is adverse selection in the Market for used capital goods. The adverse-selecton framework, however, implies heterogeneity in acquis tiog and sales prices across firms. "im addition to Arrow (1968) cited in the Introduc- tion; Nickell (1978 p. 40), Bertola and Caballero (1991 . 1h and Robert S. Pindyck (1991 p. 111) recognize that ig_may be lower than pz and choose to make the extreme assumption that pe = O- Inthe iterature fn consumer durables, Pok-sing Lam (1989), Sanford 4 Grossman and Guy Larogue (1990), and Eberly (i994) include a proportional transaction cost when ‘consumers resell durables, which corresponds (©. Py being smaller than py “Notable exceptions are Alan $. Manne (1961) and Rothschild (1971, who analyze investment behavior under concave adjustment costs In addition, the partial devivative of the adjst- rent-cost function with respect to investment goes 0 infinity as investment goes to infty, and this derivative goes to negative infinity as invesiment goes to negative infinityBn THE AMERICAN ECONOMIC REVIEW partial derivative of the adjustment-cost function with respect to K being negative.” To accommodate this case as well as the case in which adjustment costs do not de- pend on K, we assume that the partial derivative of the adjustment-cost function with respect to K is nonpositive. We assume that the adjustment-cost func- tion, is twice diferentiable with respect to I everywhere except possibly at ‘The assumptions made so fat imply that the partial derivative of the adjustment- ‘cost function with respect to investment is positive for 1>0 and is negative for <0. If the adjustment-cost function is differenti- able at I=0, the partial derivative of the adjustment-cost function is zero at I= 0; ‘more generally, the left-hand partial deriva- tive is nonpositive and the right-hand derivative is nonnegative at 1=0. (ii) Fixed costs of investment are nonneg- ative costs that are independent of the level of investment and are incurred at each point in time when investment is nonzero. Thus, a firm can avoid the fixed cost of investment at a particular point of time by setting in- ‘vestment equal to zero at that point of time. We take account of all three of these types of costs associated with capital invest- ‘ment. The total cost of investment equals the product of a dummy variable » and an “augmented adjustment-cost function” (I,K). The dummy variable y takes the value 0 when I=0 so that the total invest- ment cost is zero when I= 0. When #0, the dummy variable v equals I so that the total investment cost equals the augmented adjustment cost cI, K). ‘The augmented adjustment-cost function (J, K) represents the sum of purchase /sale costs, adjustment costs, and fixed costs, We assume that lim, ¢(1, K)= lim, , ¢(1,K) and denote the common value ‘of these "For instance, Lucas (1967, 1981), Anthur B. ‘Treadway (1969), Lucas and Prescott (1971), Hayashi (1982), and Abel and Oliver J. Blanchard (1983) all, ‘model adjustment costs as a decreasing function of K fora given I DECEMBER 1994 limits as c(0,K). Note that (0, ) is nor the total investment cost when I= 0, be- cause when /=0 the dummy variable » eq- uals 0 and total investment cost equals zero. Instead, (0, K) is interpreted as the fixed cost of investment because both the purchase/sale cost function and the adjustment-cost function are continuous functions that take on the value 0. when 1=0, Because the fixed cost is nonnegative, we have (0, K)> 0. The augmented adjust: ‘ment-cost function is continuous, strictly convex, and twice differentiable with re- spect to I everywhere except possibly at T= 05 Let ¢(0,K) and c(0,K)" denote the left-hand and right-hand partial deriva- tives, respectively, of e(/, K) with respect to T evaluated at [= 0. It follows from the assumptions made above that cj(0, K)” = 0, but ¢,(0,K)~ may be positive, negative, ot zero. In addition, ¢,(0, K)* = ¢,(0, K)- B. Maximization: The Optimal-Investment Function Assume that the firm is risk-neutral and chooses investment to maximize the ex- pected present value of operating profit w(K,e) less total investment cost ve(, K). ‘The value of the firm is thus 8) Mek uep= mix [CEkeCKontied ~ HeyseUleey Kea Meds where r>0 is the discount rate, and the ‘maximization in (3) is subject to the evolu- mn of #, and K, described in (1) and (2), respectively.” We will solve the maximization problem in (3) using the Bellman equation" (where “The properties noted in footnote 6 imply that lim s2ejT-K) = an im. 2c, K) = ‘While standard, this expression rules out bubbles in the value of the firm "A formal derivation of this Bellman equation i presented in Appendix A of Abel and Eberly (1999),vot. 40 5 we hve suppresed the ime subscript () me) = man (nc 0)-vot1)+( exer). ‘The left-hand side of equation (4) is the required return on the firm, and the right- hand side of (4) is the maximized expected return, which consists of two components: operating profits net of augmented adjust- ment costs, (K,e)~ve(I, K); and the ex- pected “capital gain” represented by the change in the value of the firm, G/dN)E(AV). To calculate the expected capital gain, we observe that the value of the firm, V; depends on K and e, which evolve continuously over time according to (2) and (1), respectively. Thus, we can caleu- late E(4V) using Ito’s lemma, equations (1) and (2), and the facts that (dK)? = (dK Xde) = (dt)? = (dzXdt)= 0 = E(dz) to obtain (5) E(dv) =[Va(1- 8K) + w(e)h, + 4o(e)"V,,.] dt Now define q = Vg, which is the marginal valuation of a unit of installed capital. Sub- stituting this definition and the expected capital gain from equation (5) into equation (4) yields (6) rV = max(n(K,€)~ vel I,K) +a(I- 8K) + ule, +30(e)V, To solve the maximization problem on the right-hand side of (6), notice that the only terms that involve the decision vari- ables [ and v are ~ve(/, K) and gl. There~ fore, the optimal values of J and v solve (1) maxlat—ve(1,K)] Itis convenient to solve the maximization problem in (7) in two steps. First, assume that »=1, and choose the value of J that ABEL AND EBERLY: INVESTMENT UNDER UNCERTAINTY Be ‘maximizes the maximand in (7) conditional on v= 1. Then choose » to be either 0 oF 1 For the moment, assume that v= 1 and let Y(q, K) denote’ the maximized value of the maximand in (7) given that v=. Specifically, (8) W(4,K) = max [al — e(1, K)] Let _I*(q,K) denote the value of I that maximizes the maximand in equation (8). Given that o(/, K) is strictly convex in I and is differentiable everywhere except possibly at I=0, the first-order conditions determin- ing 1*(q, K) are (9a) ¢(I*(4,K),K)=@ for q<¢(0,K)” or g>c,(0,K)* (9) I*(q,K)=0 for ¢,(0,K)" sqs¢,(0,K)”. ‘According to equation (9a) the firm equates the marginal cost of investment and the ‘marginal benefit of investment, measured by 4. Notice that c,,, > 0 implies that 1*(q, K) is a strictly increasing function of q over the range of q in equation (9a). If (I, K) is differentiable at 1=0, then 60, KY" =¢,(0,K)* and ¢,(1*(q, K),K) = for all g. However, if CU, K) is not differentiable at = 0, then for values of q between ¢/(0,K)~ and ¢/(0,K)* there is no corresponding value of the marginal cost of investment. As shown in equation (9b) for values of in this range, /*(q, K)=0. Looking at equations (9a) and, (9b) to- eether, we see that 1*(q, K) is’ a nonde- creasing function over the entire range of q, and that <0 forgey(0,K)* 10) 1%.)13H THE AMERICAN ECONOMIC REVIEW \ Having determined the optimal value of 1 given that » = 1, we now turn to the choice of the optimal value of ». If »=0, gross investment is also zero, and the value of the ‘maximand in equation (7) is zero. If v= 1, the optimal rate of investment is [*(q, K) and the value of the maximand in (8) is, (1) ¥(@.K) = al*(q,K)~c(1*(4,K),K). The firm will therefore choose v = 1 when, and only when, y(g,K) is greater than zero.'' To determine the sign of W(q,K), wwe now characterize the behavior of ‘this function. Recall from equation (9b) that for (0, KY < qs c0,K)*, I*g, K) = Substituting zero investment into the right- hand side of (11) yields (12) ¥(4,K (0,K) if e,(0,K) sase,(0,K)". For values of q outside the interval [c,(0, KY”, ¢(0, KY” ], Wa, K) > ~ ol0, K) because the firm could always choose to set 1=0 and thereby attain a value of ~ (0, K) for gf = cU, K). Thus, the mi of W(q,K) is attained for q in the [c(0, K)”,¢,(0, K)*]. Outside this interval, Wa, K) is twice differentiable with respect to 4g. Differentiating equation (11) with re- spect to q and using equations (9a) and (10) when gq, A= 0 the frm is indiferent between 1=0 and 1 1*(g,K). OF cours, if "Gq, K)= 0, the ‘optimal rate of investment is zero It "Ca, K) #0, we assume thatthe firm chooses to set investment equal to Zero at these points of indiflerence. The time path of Kris unaffected by this assumption because 4 = Vg(Ke) follows a difsion proces, which implies te set of times when gq, K)=0 and Iq, K) 0 has zero measure DECEMBER 1994 Ficune 1. Th Rewano ro IsvesTino yields (13) #4, K) <0 ifa0 ifa>c,(0,K)* (14) e.g 4K) = Tp (a,K) > 0 if qe,(0,K)*. ‘Thus, the function Y(q, K) is a convex function that attains its minimum value of —c(0,K) when q is in the interval [c(0,K)-,¢,(0,K)* ]. Let g; and q; denote the smallest and largest roots, respectively, of (a, K)= 0. It follows from equation (13) that (15) W(a,K)>0 — ifgay The function yg, K) is depicted in Fi ure I for a given value of K. The flat segment of Y(q, K) for values of q between ¢(0,K)~ and’ ¢\(0,K)* corresponds. to equation (12). Figure 1 is drawn under the assumption that the fixed cost, (0, K) is positive, so that the minimum value of W(q, K) is negative, and the flat segment lies below the horizontal axis. According to equation (13), #(q, K) is strictly decreasing to the left of the flat segment and strictly increasing to the right of the flat segment. ‘Thus, in the case depicted in Figure 1, the =I"(a,K)VOL. 84 NO. equation iq, K)=0 has two dist agyand dx: Wa. K)> Oil @ ap ‘Thus, optimal investment behavior Kg, K) is characterized by ct roots, P@,K)<0 ita0 a> ar ©. Characteristics of Optimal Investment and q () As shown in equation (16), the opti- mal rate of investment is zero when q is in the interval [q,, 4g]. We will show that q > 4qy. 80 that this range of inaction for invest- ment is nondegenerate, if there are fixed costs of investment [c(0, K)>0] or if the ‘augmented adjustment-cost function is non- differentiable with respect to I at In general y(q, K)=0 may have either a unique root or more than one root. If there is more than one root, there are either two roots of a continuum of roots. We describe each of these three cases below. Case I: Yq, K)=0 has a unique root so that 41 = 42 and the range of inaction is degen- erate. The equation Y(q,K)=0 has a unique root if (a) c(J, K) is differentiable at = 00 that c(0, K)- =¢,(0, K)*, and hence there is no flat segment at the bottom of the Y(q, K) function; and (b) the fixed cost c(0, K) = 0/so that the mini- mum value of #(q, K) is zero. These two assumptions are fairly standard in the adjustment-cost literature, and they ac- ‘count for the absence of a range of inac- tion in much of this literature (see e.g., John P. Gould, 1968; Richard Hartman, 1972; Abel and Blanchard, 1983). Case II: Wg, K)= 0 has exactly two roots so that there is a nondegenerate range of inac~ tion. The equation W(q,K)=0 has ex- actly two distinct roots if the fixed cost (0, K) > 0 so that the minimum value of ‘W(q, K) is negative, With a positive fixed cost of investment, a nondegenerate range of inaction will arise regardless of the ity of UI, K) at 1 ABEL AND EBERLY: INVESTMENT UNDER UNCERTAINTY 137 Case HII: Wa,K)=0 has a continuum of roots so that there is a nondegenerate range of inaction. If (a) the augmented adjust- ment-cost function c(/,K) is not differ- entiadle at 7=0 so that ¢(0,K)< ¢(0,K)* fimplying that there is a flat segment at the bottom of W(q, K)] and (b) the fixed cost c(0,K)= 0, so that the flat segment at the bottom of #(q, K) lies along the horizontal axis, then W(q, K)= 0 has a continuum of roots extending from 4q, 10 4. For any value of q in this range, the optimal rate of investment is zero, (Gi) The largest and smallest roots of the equation Wq,K)=0, a and qo, depend only on the specification of the augmented adjustment-cost function (J, K). They are independent of the specification of the operating-profit function =r(K,e) and the specification of the diffusion process for e, (iii) If there are positive fixed costs or if cI, K) is not differentiable at [= 0, there is ‘a nondegenerate range of inaction. If there fare positive fixed costs, the function fq, K) is discontinuous; the optimal rate of invest- ment jumps from a negative value to zero at = qj, and it jumps from zero to a positive value'at 4 = 4. (iv) If min; U1, K)2 0, it is never opti- ‘mal for the firm to undertake negative gross investment; the firm’s behavior is observa- tionally equivalent to a situation of versible investment.!? Note from the defini- tion of #(q, K) in equation (8) that ¥(0, K) wax, ~ CU, K) = ~min, c(1, K) so that if min, c(/,K)20, then y40,K)<0. But if WO,K)20, then q,, the smallest root of Wg. K)~= 0, is nonpositive. Therefore it is impossible for q, which must be positive [see equation (20)}, to be less than qy, and "=Cahallero (1991 p. 281) specifies the augmented adjustmentcost function CC)™= +1> O}yy® +L < yal where 6 =1, 7,2 0.7220, and the brackets denote the indicator function. Caballero states that “he irreveribe-investment ease of Pindyek (1988) and Bertola (988) coresponds tothe casein which Yano, and B=." In fact, however, if B= 1, ie: {Ersibty will ocur whenever > I There is no need to make 7 infinite o prevent optimal investment from137% THE AMERICAN ECONOMIC REVIEW the optimal rate of investment cannot be negative.'® This result has a straightforward explanation. In order for a firm to find it optimal to give up some of its installed capital, which has a positive value, the aug- mented adjustment cost it incurs must be negative (je., the net sale price of the capi- tal after taking account of the fixed cost and the adjustment cost must be positive). If there is no value of gross investment for which c(I, K) is negative, then it will never be optimal for a firm to undertake negative {ross investment, (W) Note that the Bellman equation in equation (6) holds identically in K at a point in time so the partial derivative of the left-hand side with respect to K equals the Partial derivative of the right-hand side with Tespect to K. Differentiating both sides of (©) with respect to K yields (17) = Ke) —Beq( LK) ~8q+ax(f- 6K) + MW, +3000) Kaa where 1 is optimal investment from equa- tion (16) and # is the optimal choice of v. Recall that = Vy so that ¢,—V- and doe Veug: NOW" apply Ito lem and ‘equations (1) and (2) to calculate E(dq}: (18) E{da) = ax(f- 8K) dt FMCW, dt + 40 (0)°V, x dt ‘Substituting (18) into (17) and rearranging "To show that gz, the largest oot of (a, K)= 0, is ot negative, we suppose that 4; < 0 and show that this assumption leads 10-2 contradiction. If q,<0, then according to equation (16), (0, K)= 170, )> 0, Re: call that €/(0,K)" 20 so that equation (10) implies that 7°00, K)'50, whichis a coatadieion, Therefore 4220. DECEMBER 1994 yields (19) (r+8)q mtx Kye) — fex( 1K) + Elda} a Equation (19) is essentially an Euler equa- tion from the calculus of variations. The leftchand side of equation (19) is the re- quired return (gross return before subtract- ing depreciation) on the valuation of the ‘marginal unit of capital, and the right-hand side is the expected return, whi ating profit 7<(K,e), the marginal redue- tion in. the augmented adjustment cost ~fex(/,K), and the expected capital gain Eldq}/ dt. In the special case in which there is no uncertainty, equation (19) becomes (r+ 8)q = m4(Ky 0) = ell, K) + da/ dt, which is widely used in the’ deterministic literature on the q theory of investment Gee e.g, Blanchard and Stanley Fishcher, 1989 p. 62). (vi) The marginal valuation of installed capital, q, is the expected present value of the stream of marginal products of capital This result can be shown formally using the following lemma, which is a special case of the Feynman-Kac formula (see Toannis Karatzas and Steven E. Shreve, 1988 p. 267). A simple proof is given in appendix B of ‘Abel and Eberly (1993), LEMMA 1: Suppose that x, is a diffusion and that a>0 is constant. Then. x, ELJ5814€° ds) is a solution to the differ ential equation E(dx)/dt ~ ax, + g, =0. Using the fact that q, is a diffusion and applying this lemma to equation (19) yields! 20) = [Era Kseorties) ~ Haste Ines Ky Jem ds >0. We have chosen the solution o equation (19) that oes not contain bubbles,VOL. 84 NO. 5 Thus, q, is the present value of the stream of expected marginal profit of capital which consists of two components: 7,(K,e) is the marginal operating profit accruing to cay tal, and —ve,(J,K) is the reduction in the augmented adjustment cost accruing to the marginal unit of capital. The assump- tions made above that ,(K,e)>0_and cx(L,K) <0 imply that 4, is always positive. IL, Relating the Shadow Price q to Observable Variables ‘We have shown that optimal investment is a nondecreasing function of the shadow price of capital, which is called g. In gen- eral, we cannot directly observe shadow prices. In this section we restrict our atten- tion to perfectly competitive firms with lin- early homogeneous production functions and derive expressions for q in terms of observable variables. In the first example, q ‘equals the value of the firm divided by its capital stock (Tobin's q), and in the second example q is a function of the price of output, the real interest rate, and parame- ters describing the price of output and the production function. Consider a competitive firm that uses cap- ital, K, and a vector of costlessly adjustable inputs, L, to produce output according to the production function F(K,L,e). Assume that the production function F(K,L,«) is linearly homogeneous in K and L, and note that the production function may be subject to stochastic shocks. In addition, the com- petitive prices of output and inputs may be subject to stochastic shocks. It is well known that if the firm is a price-taker in output and factor markets, the operating profit can be written as cy where H(e)> 0. (Kye) = H(e)K "She operating profit in this case cam be written as ‘x(K, 6) = maxy| ple, O)FUK,L, 6) ~ We EL, where 0) isthe given price of the fem’s output, we) i the vector of sven prices of the costessly adjustable inputs, and Q is industry output. Note that all of the prices’ may be” potentially random. Let = L/K ABEL AND EBERLY: INVESTMENT UNDER UNCERTAINTY a7 Case I: o( 1, K) is linearly homogenous in I and K.—We can show that if the operating profit function satisfies equation (21), and if (I,K) is linearly homogeneous in I and K, then (2 V(Kye) =a(e)K. In this case, the shadow price of capital, q(e), equals’ the average value of capital, V(K,6)/K, which is observable using secu: rity market prices and is known as Tobin’s 4. This result extends Hayashi’s (1982) re- Sult, which was derived in a deterministic model, to a stochastic model that admits The value function in equation (22) is implied by the following lemma, proved in Appendix A. LEMMA 2: Suppose that w(K, e) and CUI, K) are homogeneous of degree p in Land K. Then the value function can be written as V(K,e)= Me)K?, and q = V4(K,0)= p VK €)/K. ‘Thus, when the operating profit function and the augmented adjustment-cost func- tion are of the same degree of homogen ‘marginal q and average q are proportional, Jn the special case where p=1 [so that equation (21) holds, and the augmented adjustment-cost function is linearly homoge- neous}, Lemma 2 indicates that average and marginal q are equal, as in equation (22) We now discuss the content of the as- sumption that c(l, K) is linearly homoge- neous. Recall that (I, K) has three com- ponents: (i) a purchase/sale cost; (ii) an adjustment cost; and (iii) a fixed cost (As we discussed in Section I, the pur- chase/sale cost is pg 1 for 7>0 and x1 for <0, Obviously, a doubling of Te the vector of raion of the comlenly_ adh able inputs tothe capital sock. It fellows rom the lnsat"homogenciy of FCK; Lye) that 3(K,e) = mat(p(e,O)FU,Xye)~weYAIK. The matimand in Sauste brackets i independent of the india fms pital stock, K, and thus the operaig-prott function am be writen ap in equation CD), where 1H) = mace, O)FU,A ©) meV}137 THE AMERICAN ECONOMIC REVIEW 1 and K doubles the purchase /sale cost, so the purchase /sale cost func- tion is a linearly homogeneous function of Fand K. Gi) In the literature in which the adjust- ment-cost function depends on K as well as on. J, it is commonly assumed that the adjustment-cost function is lin- carly homogeneous in I and K.'° ii) The fixed cost of investment, e(0, K), is independent of the amount of invest- ment I. If this fixed cost reflects the cost of stopping production while new capital is installed,” it is proportional to the operating profit function H(e)K which is, of course, proportional to K. In this case, the fixed cost, c(0, K), is a linearly homogeneous function of / and K (even though itis independent of 1). If the purchase/sale cost, the adjustment cost, and fixed cost are all linearly homoge- neous functions of f and K, then c(1, K) is linearly homogeneous in J and K and can be written as 1 exe! (e) (z) where G(-) is continuous and convex, and except possibly at zero, is twice differen- tiable. In this case, ¢,(/,K)=G'U/K), so that equations (16) and (9a) yield (23) (1. K) Gq) <0 ifa0 ifa>a,. Notice that the optimal investment-capital ratio depends only on q, and since q is independent of the capital stock, the opti- mal investment-capital ratio is independent "Lucas (1967,1981), Lucas and Prescott (197), Hayashi (1982) and Abel and Blanchard (1983) all make this asumption. Rothschild (1971p, 608) and Nickell (1978 p. 37) both sugses that the eos of stopping production would rise toa fixed cost of investment. In addition Rothschild suggests that breaking in new equipment or procedures is costly DECEMBER 1994 of the scale of the firm.'* If g, <0, then the negative investment regime is never opera- tive, and as explained in Section I, invest- ‘ment would appear to be irreversible. Case I: ¢ (1, K)=0.—Now assume that the augmented adjustment-cost function does not depend on the capital stock (for- mally, cx(,K)=0)."° We continue to as- sume’ that the firm is perfectly competitive and has @ linearly homogeneous production function so that the operating profit func- tion is proportional to the capital stock [equation (21)}. Under these assumptions, we show in Appendix B that the value func- tion is a linear function of the capital stock regardless of the specification of the diffu- sion for e. In particular, (25) Vike a(e)K+J(e) where J(e)> 0. To get an explicit expres sion for q(e) in terms of the underlying stochastic process, we will focus on pat lar parametric specifications of the operat- ing profit function and the diffusion for e. It is not necessary to restrict c(/, K) further. Consider a competitive firm that uses cap- I and labor to produce output according to the Cobb-Douglas production function vL"K'~*, where 00 is a productivity parameter that may be stochas- tic. The firm pays a constant wage rate w per unit of labor and sells its output at a price P that may be stochastic. Define p Po and observe that the instantaneous oper- ating profit equals the revenue from selling ‘output minus the cost of labor so that (26) (Kp) = maxl pL*K!~* — wh] = ho"K where h =(1~ aa®/-° o=1/0-a)>1 «0-2 > 0 and "Lucas (1967) high istic model with a linearly homogeneous operating ‘rope function and convex costs of adjustment this assumption is adopted by Eisner and Strotz (1963), "Gould. "(1968), Rothschild (1971), Richard Hartman (1972), Mussa (1977), Nickell (1978), Pindsck (1982), Abel (1983), and Cabeliero (199),VOL. 4 NO. S At time 1, the present value of marginal profits accruing to the undepreciated por- tion of currently installed capital is” (21) qa hf El ptjerrras We calculate the expectations in equation (27), and the value of g,, under the assump- tion that p evolves according to the geomet- ric Brownian motion dp 28) Pods (28) FF where z follows a standard Wiener process. In this case, the distribution of In p,.., con- ditional on p, is NUn p, ~ 307s, 77s) So that (29) Ep? pf exp[10(8 —1)os] mn (27) ‘Substituting equation (29) into equ: and simplifying yields** hop [rr8-oo= 0") (30) 4, Now suppose that 0 0 so that .jmegral in equation 7) converses. "Recall that min, c{l,K) <0 is necessary and suf- ficient for qy>0. Either e(0,K)>0 oF ¢y(0,K)” < €1(0,)" is suicent for 43 > 45 When we consider the effécts of a change in a parameter such a6 , we are actully comparing the behavior of two others intial firms with diferent Constant values of the partmeter in question. This fnalsis does not apply 10 the effect on 2 given firm of ‘change in the parameter because the firm's optimiza tion problem assumes thatthe parameters are known with Certainty to be constant over time ABEL AND EBERLY: INVESTMENT UNDER UNCERTAINTY Bm siven p,, If the initial value of q, is less than q, or greater than or equal to qo, the increase in g, increases investment, which is consistent with Hartman (1972), Abel (1983), and Caballero (1991). But note that if the initial value of q, is in the interval [q,,42), small increase in will not move q, out of this interval, and investment will ‘remain unchanged and equal to zero. Thus, with the more general adjustment-cost function introduced in this paper, we have the result that investment is a nondecreasing function of @ fora given p,. IIL. Competitive Equilibrium” In a recent paper Pindyck (1993) ques- tions the relevance of adjustment costs in competitive equilibrium for firms with con- stant returns to scale. Pindyck also points out that considerations of industry equilib- rium may reverse the findings of Hartman (1972), Abel (1983), and Caballero (1991) concerning the effects of uncertainty on in- vestment by competitive firms. We first address Pindyck’s argument that adjustment costs are irrelevant in a per- fectly competitive industry in which firms have constant returns to scale, His argu- ment applies to the case in which the adjustment-cost function depends only on the rate of investment, and not on the capi- tal stock (see Pindyck, 1993 p. 274). Observe from equation (25) that, under constant re- turns to scale and this form of the adjustment-cost function, the value of a ‘competitive firm with capital stock K™ is ae)K* + J(e). If this firm could costlessly divide itself into two firms with capital stock K* /2, each of the two firms would be worth GedK* /2+ He); the total value of the two firms would be q(e)K* +2J(e), which is greater than the value of the original firm. ‘Thus, provided that new firms can be freely created, firms would have an incentive to We thank an anonymous referee for raising the {issues that motivated us to write this section.1390 THE AMERICAN ECONOMIC REVIEW divide into smaller parts. In addition, if there is free entry, potential entrants would enter the industry because even with no capital a firm has a positive value J(e). indyck argues that “in the limit, the indus- try would be composed of an infinite num- ber of infinitesimally small firms, and so each firm would have no adjustment costs” (p. 274) because they would each have in- finitesimally small rates of investment. Pindyck’s conclusion that each firm would have no adjustment cost is based on the assumption that lim,_.9c*UI,K)=0 and lim; 9 ¢7U1, K)=0, where ¢*(, K) is the adjustment-cost function, rather than the augmented adjustment-cost function. While this assumption is fairly standard in formu- lations of the adjustment-cost function in which cy is identically zero, adjustment costs will not become irrelevant in our Case 1, in which the augmented adjustment-cost function is linearly homogeneous in J and K. As we have shown, the value of the firm is strictly proportional to its capital stock in this case. Therefore, a firm with zero capital has zero value, so that even with free entry there are no rents to be earned by potential entrants with zero capital. In this case, the size distribution of firms is indeterminate. It is possible that some firms will have in- finitesimally small capital stocks and rates of investment, but even for these firms ad- justment costs are not irrelevant. Arbitrarily small firms will have arbitrarily small values of [and K, but the value of 1/K will stil be given by equation (24), which depends on the augmented adjustment-cost function.” Thus, Pindyck’s (1993) argument about the irrelevance of adjustment costs under con- stant returns to scale and perfect competi- tion does not apply when the augmented adjustment-cost function is linearly homoze- neous in J and K, as in our Case I Recall from equation (23) that in this ease the augmented adjustment cost function can be writen a5 KGU/K), where G(-) is continuous and convex. Al though the augmented adjustment cost KOU/ I) goes to zero as K goes to zero, the marginal avgmented adjustment cost G°U/), evaluated at optimal J, does ot goto 2er0 a5 K goes 10 zero DECEMBER 1904 Pindyck’s second argument is that, even if for some reason firms cannot be arbitrarily small, the response of existing firms and free entry will cause the equilibrium price to respond endogenously t0 shocks. Most studies of investment behavior by competi- tive firms under uncertainty ignore this en- dogenous response of equilibrium price. Al- though a competitive firm is a price-taker, a competitive industry is not a price-taker. Specifically, a shock that hits all firms in an industry is likely to affect industry output and thus the equilibrium price. However, a shock that hits only one competitive firm in an industry will not affect industry output or the equilibrium price. Pindyck (1993) analyzes endogenous price responses to industry-wide shocks to reex- amine the results of Hartman (1972), Abel (2983), and Caballero (1991), who find that increased uncertainty increases the invest- ‘ment of competitive firms with constant re~ turns to scale. Pindyck shows that if all firms in an industry face identical realizations of the random variable(s) impacting the indus- try, then taking account of the endogenous response of the equilibrium price tends to reverse the findings of Hartman, Abel, and Caballero. However, it should be noted that if competitive firms face only idiosyncratic shocks, then the results of Hartman, Abel, and Caballero continue to hold. Our analy- sis in Case I would be subject to Pindyck’s criticism if we interpret the uncertainty about p = Pv as arising from demand shocks that affect the competitive price of output P, which is identical for all firms in a com- petitive industry; however, our analysis in Case Il is immune to Pindyck’s criticism if the uncertainty arises from a productivity shock 0 that is idiosyncratic to a particular firm. The isue of the endogenous response of equilib. tum price to shocks does ot arse in our analysis of (Case 1, because we need not specif the relationship between price and the source of uncertainty. Indeed, four anabsis of Case I didnot use any specification for the evolution of the price of output. Whatever the behavior of the price of output, and however it re- sponds to shocks, competitive firms take the price of ‘output a given,VOL. 84 NO. 5 WV. Conclusion In this paper we have extended the ad- justment-cost framework under uncertainty to incorporate fixed costs of investment, a wedge between the purchase price and sale price of capital, and potential irreversibility of investment. In this extended framework, investment is a nondecreasing function of 4, the shadow price of installed capital, and there are potentially three investment regimes which depend on the value of ¢ relative to the critical values g, and q3. Conveniently, these critical values depend only on the Specification of the augmented adjustment-cost function. If q is greater than q,, then, as is standard in the q-theory branch of the adjustment-cost literature, in- vestment is positive and is an increasing function of q. If q is between g, and gy, then the investment is zero. Although this regime features prominently in the irre- versibility literature, it is largely ignored in the adjustment-cost literature. Finally, if q is less than q,, gross investment is negative, a possibility that is simply ruled out by as- sumption in the irreversibility literature ‘The shadow price q is in general not observable, so we presented two examples relating q'to observable variables. In one example, restrictions on the production function and the augmented adjustment-cost. function guarantee that q is identically equal to the average value of the capital stock, which is observable using security prices. In the other example, we tightly specify the production function and the diffusion pro- cess for the random variable p (the product of the output price and a productivity pa- rameter) and derive an expression for q as a function of the contemporaneous value of p. In this example, p does not have a sta- tionary distribution, and hence g does not have a stationary distribution. In ongoing research we are examining the behavior of q and investment in the presence of mean-reverting process for p so that q will have a stationary distribution. The ultimate goal of this line of research is to derive an econometric specification to apply these models to aggregate and disaggregate data on investment, ABEL AND EBERLY: INVESTMENT UNDER UNCERTAINTY 281 Appenpix A PROOF OF LEMMA 2: ‘The operating-profit function and the ‘augmented adjustment-cost function are ho- ‘mogeneous of degree p in I and K so that (A) (Ke) = He) KP and (A2) (LK o(z)«° ‘Then the value function in equation (3) can be written as (A3) VK.) max [EMH 6.05) MeO NIKE beds where i... Jp.,/Koy isthe (gross) growth rate of the capital stock. Consider a firm with capital stock K{? at time f, and let v{2, and if, denote the optimal values of the dummy variable » and the investment-capital ratio chosen by this firm at time £ + s. This optimal behavior leads to 4 capital stock of K‘?, at time ¢. The value of the firm at time f is V(K!,e,). Now consider a second firm with a capital stock at time f equal to K®=aK0” with a> 0. ‘This firm has the option of choosing exactiy the same values of the dummy variable v and the investment-capital ratio I/K at ‘every point of time as chosen by the firm with capital stock K{. If the second firm were to set v2, =0{0, and i, =i, for all s> 0, then K®, would equal aK’, for all s>0. Because’ the cash flow at’ time 1-45 is proportional to K/., in equation (A3), the second firm has the option of obtaining an expected present value of cash flows equals to a”V(K{?,¢,). Therefore, (AS) V(aK,,£,) 2 @V(K,,€,)- Equation (A4) holds for any K, and for any positive factor a. In particular, consider a1380 THE AMERICAN ECONOMIC REVIEW first firm that has a capital stock of aK, at time ¢, and a second firm that has a capital stock of K,=(/a)aK, at time 1. There- fore, the argument preceding equations (A4) implies that (48) V(K,.¢,) = (1/a)"V(aK,,€,). Putting together equations (A4) and (AS) we have V(aK,,e,)> a’ViK,,,) = V(aK,,e,), which implies (A6) V(aK,,£,)=a°V(Ky6,) Because equation (A6) holds for any posi- tive K, and any positive a, the value of the firm is proportional to the capital stock to the power p, and hence the value function cean be written as (AI) (Ki, = Ae) KP Partially differentiating (A7) with respect to kK, yields V( K€) 9: = Vi Kes 8s) = pe ‘Appenpix B The Value Function When the Augmented Adjustment-Cost Function Does Not Depend on the Capital Stock ‘The optimal program of the firm is gov- erned by the differential equation given in ‘the text equation (6). Here we assume that ex(I,K)=0, so we write the augmented cae function, c(/, K) as simply o(). (BI) MK) = max w(K, 0) ve) + a BK) +mork, Ouan Now suppose that the firm is a price-taker in output and factor markets and has a production function that is linearly homoge- DECEMBER 1994 neous in J and K so that 7(K,e)= H(e)K Gee footnote 15). We will verify that V(K,e)= q(e)K + J(e) satisfies (BI). Sub stituting V(K, 6) = q(e)K + J(e) and x(K,e)= H(6)K into (B1) and recalling the definition of g yields (B2) raleK + ede) = max (HK —¥e(D, 4 a(6U~8K)+ wlo)g, K+ Mey, Hotere K+ o(eFI,) Collecting terms in K yields (B3)_ [er +8)a(e) ~ ate a, ~ jote)"a,.~ Hed] K = maxtateyl—¥e( D1 ~ ey wey ole) Ie In order for (B3) to hold for all K, the term in square brackets on the left-hand side must equal zero, and the right-hand side of (B3) must also equal zero. Note that from equation (8) we can write (BA) max[a(e)1—ve(1)] = max[0,¥(4)]. Setting the right-hand side and the left-hand side of (B3) equal to zero yields (BS) max[0,¥(4(«))] — J(e) + ue). + ioe) he= and (B6) H(e)—(r+8)a(e) + ule)ay +ho(e)a. Note that both differential equations are of the form (BT) g(e)—ax(e) + E(dy /dt) =0.VoL. 8 NO. 5 ‘According to Lemma 1, a solution to the differential equation in (B7) is (88) x(2,) = E,[8(e.e* ds Since equations (BS) and (B6) are both of the form in equation (B7), we substitute from these into equation (B8) to conclude: (B9)_ Je) =f maxl0,H¢ a leas (B10) aCe) = Ef HEE, de ds ‘Therefore, J(c,) can be interpreted as the present value of rents accruing to the firm from the augmented adjustment technology, and q(¢,) is the present value of marginal products of capital. Note that this solution was derived for any diffusion process gov- ening €,. REFERENCES ‘Abel, Andrew B. “Optimal Investment under Uncertainty.” American Economic Re- view, March 1983, 73(1), pp. 228-33. Abel, Andrew B. and Blanchard, Olivier J. Intertemporal Model of Saving and In- vestment.” Econometrica, May 1983, 5103), pp. 675-92. Abel, Andrew B. and Eberly, Janice C. “A Uni- fied Model of Investment under Uncer- tainty.” National Bureau of Economic Research (Cambridge, MA) Working Pa- per No. 4296, March 1993. Arrow, Kenneth J. “Optimal Capital Policy with Irreversible Investment,” in J. N. Wolfe, ed., Value, capital and growth. Papers in honour of Sir John Hicks. Edinburgh: Edinburgh University Press, 1968, pp. 1-19. Bertola, Giuseppe. “Adjustment Costs and Dynamic Factor Demands: Investment and Employment Under Uncertainty.” Ph.D. dissertation (Ch. 2), Massachusetts Institute of Technology, June 1988. Bertola, Giuseppe and Caballero, Ricardo J. “Irreversibility and Aggregate Invest- ABEL AND EBERLY: INVESTMENT UNDER UNCERTAINTY 1388 ment.” Review of Economic Studies, April 1994, 61207), pp. 223-46. Blanchard, Olivier J. and Fischer, Stanley. Lec ‘tures on macroeconomics. Cambridge, MA: MIT Press, 1989. Caballero, Ricardo J. “On the Sign of the Investment-Uncertainty Relation ship.” American Economic Review, March 1991, 8/(1), pp. 279-88, Eberly, Janice C. “Adjustment of Consumers’ Durables Stocks: Evidence from Automo- bile Purchases.” Journal of Political Econ- omy, June 1994, 10203), pp. 4 nants of Business Investment,” in Com- mission on Money and Credit, Impacts of ‘monetary policy. Englewood. Cliffs, NJ: Prentice Hall, 1963, pp. 59-337. Gould, John P.'“Adjustment Costs in the ‘Theory of Investment of the Firm.” Re- view of Economic Studies, January 1968, 35(101), pp. 47-55. Grossman, Sanford J. and Laroque, Guy. “As- set Pricing and Optimal Portfolio Choice in the Presence of Illiquid Durable Con- sumption Goods.” Econometrica, January 1990, 58(1), pp. 25-51. Hartman, Richard. “The Effects of Price and Cost Uncertainty on Investment.” Journal of Economic Theory, October 1972, 5(2), pp. 258-66. Hayashi, Fumio, “Tobin's Marginal and Av- erage q: A Neoclassical Interpretation.” Econometrica, January 1982, 50(), pp. 213-24 Jorgenson, Dale W. “Capital Theory and In- vestment Behavior.” American Economic Review, May 1963 (Papers and Proceed ings), $3(2), pp. 247-59. Karatzas, loannis and Shreve, Steven E. Brown ian motion and stochastic calculus. New ‘York: Springer-Verlag, 1988. Lam, Pok-sang. “Irreversibility and Con- sumer Durables Expenditures.” Jounal ‘of Monetary Economics, January 1989, 23(1), pp. 135-50. Lucas, Robert E., Jr. “Adjustment Costs and the Theory of Supply.” Journal of Politi= cal Economy, August 1967, 75(4), pp. 321-34, “Optimal Investment with Rational Expectations,” in Robert E, Lucas, Jr.,ass ‘THE AMERICAN ECONOMIC REVIEW and Thomas J. Sargent, eds, Rational expectations and economic practice, Vol 1. Minneapolis: University of Minnesota Press, 1981, pp. 55-66. Lucas, Robert E., Jr. and Preseott, Edward C. “Investment under Uncertainty.” Econo- ‘metrica, September 1971, 39(5), pp. 659- “Capacity Expansion and istic Growth.” Econometrica, July 1961, 29(3), pp. 632-49. Mussa, Michael. “External and Internal Ad- justment Costs and the Theory of Aggre- gate and Firm Investment.” Economet- rica, May 1977, 44(174), pp. 163-78. Nickell, Stephen J. The investment decisions of firms, Cambridge: Cambridge University Press, 1978. Pindyck, Robert S. “Adjustment Costs, Un- certainty and the Behavior of the Firm.” American Economic Review, June 1982, 7203), pp. 415-27. ___- “Irreversible Investment, Capacity Choice, and the Value of the Firm.” DECEMBER 1994 American Economic Review, December 1988, 78(5), pp. 969-85. . “Irreversibility, Uncertainty, and Tnvestment.” Journal of Economic Litera- ture, September 1991, 29(3), pp. 1110-48. 2A Note on Competitive Invest: ment under Uncertainty." American Eco- nomic Review, March 1993, 83(1), pp. 273-77. Rothschitd, Michael, “On the Cost of Agjust- ment.” Quarterly Journal of Economics, November 1971, 85(4), pp. 605-22. ‘Sargent, Thomas J. “Tobin's q’ and the Rate of Investment in General Equilibrium.” Carnegie Rochester Conference Series on Public Policy, Spring 1980, 12, pp. 107-54. Tobin, James. “A General Equilibrium Ap- proach to Monetary Theory.” Joumal of ‘Money, Credit, and Banking, February 1969, 1(1), pp. 15-29. ‘Treadway, Arthur B. “On Rational En- trepreneurial Behaviour and the Demand for’ Investment.” Review of Economic Studies, April 1969, 36(106), pp. 227-39.