1. Introduction This paper aims at analyzing the effect of uncertainty on investment. We focus on two types of uncertainty: demand and interest rate uncertainty. From a theoretical point of view we will follow the example of recent economic literature concerning stochastic models of irreversible investment decisions, while empirically we will restrict our attention to the Italian economy. We start by analyzing the influence of interest rates on investment demand. It is well known that, within the traditional neoclassical framework, a higher interest rate level will depress investment demand. It is perhaps less well known that with reversibility higher interest rate volatility will stimulate investment demand due to the convexity of the firm's present value (and so to Jensen's inequality) with respect to the interest rate. With irreversibility, the effect of higher interest rate volatility also depends upon the assumed stochastic process that drives the interest rate. Indeed, under irreversibility the exact way in which interest rate uncertainty affects investment is the result of two opposite effects: a positive one identical to the one just seen in the reversible investment world, and a negative effect due to the increase in the option value of waiting. In this paper we show that under
*Financial support from MURST (fondi 60% and ex-fondi 40%) is gratefully acknowledged. We would like to thank Giuseppe Bertola and two anonymous referees for their very helpful suggestions. The usual disclaimer applies.
491
492
2. Investment under Uncertainty In this section we will analyze the investment decision problem faced by a firm when the interest rate and demand are uncertain. This is done by following the standard approach found in most recent economic literature (for a survey, see Dixit and Pindyck 1994).
~This result may also be due to the smoothness brought about by aggregation and to the consequent disregard for major and infrequent adjustments (see Caballero 1997).
493
We consider a risk neutral firm producing a single final good whose operating profit is:3
Y~(K),
(I)
with 7z' (-) > 0 and re" (.) < 0. In Equation (1) Y is a demand shock, fl is a positive constant and K is the capital stock. We assume that there is no depreciation and that the price of capital is constant and equal to 1. The firm's optimal investment can be determined by maximizing its market value V(.) or, equivalently, the expected present value of its cash-flows:
exp -
s r(u)du ) [Yfln(K) -
I] ds
1,
(2)
where r is the interest rate, Idt = d K is net investment and E t denotes the expectation conditional on information available at time t. In what follows, we will analyze the effect of the two sources of uncertainty on investment separately, so as to keep the analytical problem tractable and to obtain closed-form solutions which allow us to draw clear-cut conclusions.
Interest Rate Uncertainty Interest Rate Uncertainty and Reversible Investment. To keep our
analysis simple, we eliminate the effect of demand shocks by setting fl = 0. Therefore, Equation (2) becomes
V(r, K) = max Et
{t}
II
exp -
r(u)(du
[n(K) - I] ds
(3)
(4)
~E quation (1) can be obtained from a set-up with a technology F(L, K) = L ~ K b mad a constant elasticity demand curve such as Q = Y P - ~ , where Q is the output, L is labor and ~ > 1. Maximizing with respect to labor, it can be shown that the operating profit of the firm, i.e., revenue minus the cost of the variable factors of production, is given by C Y ~ K ~ where fl = [1/a + a(1 - E ) ] a = [b(e - 1)/~ + a(1 - a)] and C is a constant. In the main text we assume that rt" (.) < 0; this requires a < 1, i.e. that the marginal revenue product of capital is decreasing, even if there were increasing returns to scale (a + b ~" 1).
494
where z is a Wiener process and G is a constant that represents the interest rate volatility, assumed to be less than 1.4 Omitting arguments whenever possible for the sake of simplicity, the Bellman equation for this problem is
rVds = max {[Tt(K) I]ds+
1I}
Et (dV)} .
(5)
Equation (5) states that at the optimum the expected return, which is the sum of its cash-flow plus its expected capital gain (or loss), must be equal to the market return calculated at the current interest rate. Applying Ito's Lemma to obtain an expression for G(dV), given the stochastic process assumed for r (see Equation (4)), we obtain:
Et(dV) =
(VkI + -~ 1 V . - ~ r a)dt .
If we substitute this expression in Equation (5) and maximize, we get the first-order condition VK = 1. This implies, again from Equation (5), that V(r, K) satisfies the following differential equation:
rV = ~ + 1 Vr,_~r 3 .
(6)
solution to this equation is V - r(1 - ~ ) so that from the first-order condition VK = 1: ~t'(K)
r(1 - a~)
1.
(7)
Equation (7) has two important implications. The first one is that in a de,2 = 0) firms terministic environment with a known interest rate (i.e. with o"r demand capital up to the point where the marginal expected present value derived from an additional unit of capital is equal to the cost of capital. The
4This assumption constrains a firm's value so that, under reversibility, it is finite (see infra Equation (4)) and the rate of interest under irreversibi/ity takes on only positive values (see infra Equation (9) and footnote 5). See also Cox, Ingersoll, and Ross (1980), who uses the same process in a completely" different context (to analyze loan contracts) and impose the same condition (to assure bounded asset prices).
495
Interest Rate Uncertainty and Irreversible Investment We now assume that investment decisions are irreversible, meaning that scrapping is never profitable in the case that firms" past decisions turn out to be wrong. This also means that investment cannot be negative, i.e. I >- 0. The introduction of a lower barrier implies that we must consider the complete solution of the differential Equation (6) (see Dixit 1993, chapter 3 for a discussion of this point). The complete solution is the sum of two components: the particular solution and the solution of the homogeneous part. It can be verified that the complete solution is
V - r(1 - a~) + A1 (K)r"l'
(K)
(8)
where A1 (K) is a constant to be determined and n i = 1 / 2 ~/(1/4) + (2/~) is the negative root of the characteristic equation, i.e. of n 2 - n - 2/a~ = 0. We have eliminated the positive root from the solution since it has no economic sense: it implies that the value of the firm increases when the interest rate increases. The first term of the solution (8) is the same as before and represents what might be called the fundamental value of the firm. The second term is the option value, which is the value of the firm derived from future expansion capacity. Once again, using the first-order condition VK = 1, we now obtain:
~' (K)
r(1 ~)
+ A'I(K)rnl = 1 .
This is called the value-matching condition. The other condition of optimality to be satisfied in this context is the smooth-pasting condition [for an exposition of these conditions, see Dixit 1993, chapter 4]. It requires that Vl~ O, or
=
~'(K)
+ A~ (K)nlr~1-1 = 0 .
which is greater than Jz'(K)/(1 - a~), since n 1 < 0. s In other words, Equation (9) shows that, for a given capital stock, firms require a lower interest rate to invest when irreversibility is present than when it is not. This is due to the option value effect which gives rise to an opportunity cost when the option is exercised, i.e. the investment is undertaken. Equation (9) also allows us to analyze how changes in interest rate uncertainty affect investment demand. Indeed, a greater Gr 2 has two effects: on one hand, as we have already seen in the previous section, investment increases, since (d/OG2r)(x'(K)/1 - a~) > 0; but, on the other, under irreversibility the option value also increases, given that (anl/acr~ > 0. In general, 2 the total effect of an increase in oV is not determined. However, given the stochastic process (4) that drives r, the option value effect is dominant, i.e. Or/Oa~ < 0, s so that an increase in interest rate uncertainty reduces the desired capital stock or makes it more profitable to keep the option alive and to delay investment.
,/(1/4) + (Z/O'r~).
STo see this, use the definition o f n 1 to write:
ao~
(1 -
o-r)
ni(n t + 1) 22 n 2 1
o-4(1 +
L_:
.]
+ 1)
(n~ -
'h
(1
--
O-r) n 1
2 2 2 n l ( n l + 1) -
2--2 - 2)'h(~l
] - 1!/
~r
a~ (i
-
497
Giorgio Calcagnini and Enrico Saltari and constant value. Demand uncertainty derives from the motion of Y which follows a geometric Brownian diffusion process: dY Y -/zdt
+ ardz ,
(10)
where z is a Wiener process, t is the instantaneous mean of the demand growth rate and ay is a constant that represents the volatility of the demand growth rate. Accordingly, Equation (2) becomes (assuming now fl > 0)
{;(
exp - r ( s - t)
[Y~Jz(K) - I] ds ,
(11)
I]dt + Et[dV]}.
(12)
Given the stochastic process (10), and making use of Ito's Lemma to evaluate Et(dV), we obtain
Et(dV) =
If we substitute this expression in Equation (12) and maximize, we again obtain the first-order condition VK = 1. This implies that V satisfies the following differential equation:
(13)
A particular solution of this equation is
YP~(K)
v ?"
-7Tt
(14)
In Equation (14) V is the expected present value associated with K since (r - m ) - 1 is a discount factor with m equal to the expected growth 1 rate of profits, i.e. m = dt [E~(dYP)~]: it is the fundamental value of the firm. 498
Given the first-order condition V~: = 1, and Equation (14), we obtain 1):
4
r = Y~n' (K) + tilt + - ~ t ( t -
1).
(15)
Since r is an exogenous variable for the firm, Equation (15) determines the optimal or desired capital stock. F r o m Equation (15) we see that (a) an increase in the expected d e m a n d growth rate g always increases firm desired capital stock; (b) an increase in d e m a n d volatility ~ will determine a higher (lower) capital stock when t > i ( t < 1) .7 Finally, an increase in a~,- will have no effect on K when t = 1.s In the case of a Cobb-Douglas production technology, it can be shown (see footnote 3) that re(K) = K ~ (with ~ < 1), and so from Equation (15) firm d e m a n d for capital is 1 K= - r m ~-1 y l - ~ eyr
1-a
(16)
ylt + Y(7 -
1)~2 "
(17)
Equation (17) shows that an increase in the expected demand growth rate # always has a positive effect on investment, while the precise effect of demand volatility oar on investment depends on 7: if 7 = 1, that is if ~ + fl = 1, I will be independent of o'~. In particular, it can be shown that this holds when returns to scale are constant. 9 Conversely, when returns to scale are increasing (decreasing) ~ + t > 1 (a + t < 1) and y > 1 (y < 1). D e m a n d Uncertainty and Irreversible Investment. H e r e we assume that there is no secondary market for capital, or that reducing K is infinitely eostly. Consequently, the firm's problem can still be represented by Equation (11), subject to the additional constraint that I -> 0. Just as before, the introduction of the barrier requires us to consider the complete solution of the differential Equation (13):
vfl should also be less than n 1 otherwise it can be shown that r - m becomes negative. SIntuitively, given the level of demand elasticity, a higher substitutability between inputs means a higher value of ff On the other hand, a higher fl implies higher expected profits. 9Given e~and fl as in footnote 3, it easy to show that ~x + p ~ 1 if a + b ~ 1. 499
/" -m
B1
(K)Y ~I ,
(18)
where n 1 is the positive root of the characteristic equation, i.e. of n [(o'~/2) - / 2 ] - r --- 0, and B 1 (K) is a constant. The negative root has been eliminated since it implies that the firm's value decreases when Y increases (remember that Y has positive effect on profit). According to solution (18), the firm's value has two terms: the first is once again the fundamental value of the firm, while the second term represents the option value. To get a solution, we again make use of the value matching condition:
-
VK -'t" - m
+ B~ (K)Y~ = 1 ;
(19)
YP- 1~'
m
+ B~(K)nlY'Z~-I = 0 ;
(20)
(r - m) = YP~'(K) .
(21)
Given our assumption of capital irreversibility, Equation (21) still does not allow us to determine the expected accumulation rate, as we did in the previous section. Indeed, in the presence of negative demand shocks we obtain YPn'(K) <
n 1
n ---2 -1 (r - m ) .
-fl
(22)
but firm capital stock cannot be adjusted. In the opposite case, firms will adjust capital to the new demand level and investment occurs. If we specify firm production technology as we did earlier, then the operating profit is YPK~, with oL< 1; moreover, making use of the theory of Brownian motions, we are able to determine the long-run distribution of the marginal profit and the corresponding accumulation rate. Applying the analysis already carried out for a similar case by Dixit and Pindyck (1994, 373), it can be shown that the expected accumulation rate is
500
(23)
where, as before, ~ = (13/1 - c0. Equation (23) shows that, given a Cobb-Douglas technology and investment irreversibility, in the long-run: 1. an increase in the expected demand growth rate always has a positive effect on expected investment; 2. an increase in demand uncertainty has always a negative effect on expected investment.
where th is the economy-wide average, v, is the firm-specific disturbance term with Et(v~) = O, Et(v~) = cr~t, and Et(') is the mean of firm-specific disturbances. If we define with At, Bt, Ct, respectively, the probabilities that the perceived rate of change was up, same, down, then it can be shown that 501
I1"1
-1
-2
-3
-4
1 ~ ExpectedGrowthRate ~Volatility I
fit
(~ a t ct
-F --
ct at
-2 aye = ~ - - ;
ct at
(25)
where at = F-l(1 - At), ct = F - t ( C t ) , 6 = fllt for all i, t, and F is the cumulative distribution of vit. Assuming that vlt is normally distributed, Pt and ayt can be readily calculated from sample proportions A t and Ct. Finally, 6 (i.e. what is known as the just-noticeable difference in the rate of change in the key variable) is a scale factor which, for our purposes, we arbitrarily set equal to 1.1 In Figure 1 we graphed both Art and art. 11 Two points are worth noticing. The first one is the negative relationship between the two time series, even if this correlation is not very strong (corr~ut, ayt) = - 0.21). In other words, there is no reason to expect higher (lower) uncertainty when expected
1See Dasgupta and Lahiri (1993) and references cited therein for how to calculate 6. XlTo make the graph clearer, the two series have been standardized.
502
d e m a n d growth rate is low (high). The second point is that, as we would expect, uncertainty is higher near a turn in the business cycle. As for interest rate variables, we first constructed a series of the expected real interest rate (r~) as the difference between the 3-month Italian Treasury bill rate and the CPI inflation.12 We focused on short-term interest rates because of the difficulty in measuring medium- or long-term expected inflation and, hence, expected real interest rates. Second, to obtain a measure of its volatility, we estimated the data generation process for rt, modelling it as an Ornstein-Uhlenbeck process, henceforth OU process. According to the O U process, the instantaneous change in the real interest rate (drt) is described as
(26)
where (~b/2) represents the long-term mean, 2 is the speed of adjustment coefficient a~ is the variance of dry, d z is a Wiener process (see Chan et al. 1992; Evans, Keel, and Okunev 1994; Calcagnini and Evans 1995). Equation (26) offers considerable flexibility in modelling real interest rates and implies that the conditional volatility of changes in rt is constant, as opposed to Equation (4). W e tested this restriction in two ways: first, we estimated Equation (26) by OLS and checked for heteroscedasticity and, second, we tested the more general model:
~'t+l -- f t = (~ -'}- 21"t + Ct+l ;
E(Et+l) = 0 ;
(27)
as a set of overidentifying restrictions on a system of'moment equations using the Generalized Methods of Moments (GMM) (see Hansen 1982 and Chan et al. 1992). By estimating model (27) we are able to directly obtain values
lVFo obtain a better discrete appro~dmationof the instantaneous change in real interest rates we constructed a monthly series for rv We followedthe strategy to measure r t as it - Pt, where it is nominal interest rate and Pt the inflation rate, since our measure of the monthly inflation rate was estimated to be a random walk. Consequently, Pt = Et(Pt+l). We did not use survey data on inflation expectations to obtain a series for the expected inflation, since in the first part of the eighties the sample proportion Ct, i.e. the probability of a decrease in the perceived inflation, was close to zero. In a case like this, assuming, as we did, that v~tis normallydistributed, ct and expressions for ,ut and cryt as in (25) could only be obtained making use of ad hoc procedures (see Visco 1985). Finally, we did not show a graph for r, and a n since the time period for the two series was too short to be able to analyze their behavior over the business cycle. 503
(7.23)
-0.54* ( - 9.53)
(6.83)
-0.67* ( - 7.49)
-
(12.96)
0.68* (-lO.91) 0.18 (1.77)
(7 r
"2
0.041" (4.92)
ARCH test p-value White her. test p-value %2(6) test p-value Nobs
NOTE: (a) = Variance of residuals from OLS. Student-t values are in parentheses. Significance level: * = 0.05.
for the parameter x (x ~ 0 implies that the conditional volatility of rt is not constant), and parameters and as in model (26). Results for models (26) and (27) are shown in table 1.13 Estimates for q5 and 2 are statistically significant and quite similar in both cases, while x is not statistically different from 0. The latter result is the most significant for our purposes and, besides justifying the use of OLS to estimate model (26), implies that the volatility of the real interest rate is constant) 4 In other words, we should expect that changes in the real interest rate volatility play no role in determining the dynamics of the accumulation rate. To verify this interpretation, we constructed a series for the volatility
13We present two OLS estimates over different time periods because we were not able to use the GMM estimation technique {'or the longest period, due to the presence of negative values for rt in the first part of the '80s. 14We also estimated a GARCH(1,1) for rt, but the model was rejected by our data.
504
of rt, o-~n,as the squared residuals from Equation (26) which we will use as an explanatory variable in model (32). We note that the two stochastic processes for r ~ t h e first assumed to obtain analytical results and the second estimated for the period 19801995--are different. Indeed, as shown by Metcalf and Hassett (1995) in the case of the output price, we could expect that cumulative investment is generally unaffected by the use of a mean reversion process rather than a Brownian motion of the type (4). Therefore we expect that the difference between the process we assumed in Section 2 and the one we estimated in Section 3 does not imply the impossibility of correctly identifying the effect of the interest rate volatility on investment. Moreover, given our estimates of model (26), we should expect and do find that changes in investment demand are not affected by changes in a~ since the latter is constant during the period we are analyzing.
3. Model Estimation
This section focuses on the empirical relationship between investment and demand uncertainty using as a benchmark model (17) or, equivalently, model (23)_is Both models state that investment reacts positively to the expected growth rate of demand and negatively to its volatility, even if we require more stringent assumptions in the case of reversibility. In particular, we should require that ~ < 1 to always find a negative effect of demand uncertainty on investment. This restriction can easily be tested by looking at the estimated coefficient of the expected demand growth rate (]zt). To control for the presence of time to build effects (i.e., partial capital adjustment and therefore serial correlation), the level of the investment rate (dK/K = I/K) lagged one quarter has been added to model (17). t6 Therefore, the estimated model may be written as
bo + blot t + bz(o-~) t + b 3 ~
t--1
--
_..}._
St,
(28)
lSThe investment rate is defined as the ratio of net investment (equipment + means of transport) at time t to net capital stock at time t - 1. Quarterly capital stock data axe constructed using a perpetual inventory and a constant rate of depreciation (10% for equipment and 16.5% for means of transport), starting from the 1970 value published in Annunziato, Manfroni, and Rosa (1992). However, while the investment rate refers to the whole economy" (since disaggregated quarterly data are not available for gross investment), demand variables refer to industrial
firms.
16In two cases, we corrected our estimates for residual serial correlation by using an AR1 estimation procedure (see Table g).
505
1995:iii
1. ~/ K 2./--(~) K 0.643* (3,91) 0.215" (4.58) - 0,607" ( - 4.20) 3. A / K 0.534* (3.14) 0.192" (5.25) - 0.498* ( - 3.25)
. . . . . .
Const.
/2t (a~)t
[,~
~] K t - ~ R~
1.026.
(31.30) 0.957 0.124 1.516 0.025 99
1.005.
(34.66) 0.939 0.115 1.846 ... 98
NOTE: (a) AR(1) correction. Student-t values are in parentheses. Significancelevel: * = 0.05.
w h e r e all variables have b e e n previously specified and et is an error t e r m with the usual properties. Results are shown in Table 2, columns 1 and 2, and strongly support m o d e l (28): the expected d e m a n d growth rate has a positive effect on the investment rate, while d e m a n d volatility has a negative effect. 17 Moreover, since the value of the estimated coefficient of the lagged d e p e n d e n t variable is not significantly different from L we r e - e s t i m a t e d Equation (28) using as a d e p e n d e n t variable the first difference of the investment rate. Making use lVTheoreticalmodels derived in Section 2 are only suitable for cross-sectional predictions. In principle, to analyze the effects of time-varying uncertainty, we should build in the stochastic process for the demand (or interest rate) volatility into the models. However, Pindyck and Solimano (1993) have already noted that, in option valuation models where stochastic volatility has been explicitlytaken into account, the effects of this different model specification are small in absolute terms. Moreover, the explicit consideration of stochastic volatilityintroduces formal complications that have not been completely resolved in financial literature (see Hull and White 1987). 506
lnrt - ln(1 -
a~),
(29)
lnrt -
ln(1-
a~) + ln(1 + 1 ) .
(30)
R e m e m b e r i n g that (3nl/3a~) > 0 and that n 1 < 0, the last two terms of Equation (30) show that d e m a n d for capital depends both positively and negatively on interest rate volatility, respectively: which of the two effects prevails is an empirical matter. Equation (30) has a more general specification than Equation (29); therefore, in what follows we will limit our discussion to Equation (30). Since Equation (30) contains stationary (rt, a~) as well as non-stationary variables (Kt, Qt), we added - In K t - ] to both sides of the equation and _ In Qt- 1 on its right-hand side. After rearranging terms, Equation (30) can be written as
( I ) t = lnO + /h -- InKt_l -
lnrt -
ln(1 -
a~)
In(1 + 1 ) (31)
ln(Q~_l) ,
lSIndeed, using a standard ADF unit root test, we cannot reject the hypothesis that the investment rate is integrated of order 1 [i.e. I(1)] at the 10% significance level, and that the demand gro~-h rate and its volatility are both I(0) at the 5% significance level. 19However, we also note that the restrictSon61 -> 1/2 Ib21implied by Equation (17) or (23) is not satisfied. These beta coefficients are calculated using coefficients reported in column 4 of Table 2. ~To see this, refer to the expression for n(K) in footnote 3 with 13 = 0. 507
where (I/K)t = In KS - In KS_ i, #t = In Qt - In Qt_ t and (In Ks_ t - In Q t - 1) plays the role of the error correction term within an error correction model, as long as the log of the ratio of capital to output is stationary,st Finally, to obtain empirical estimates of the effects of each individual variable on the investment rate, Equation (31) has been specified as
(')
c4)(ln a~)t
d- C 5 ~ t _
(')
(32)
where (I/K)t, tzt, rt, a~, (ln KS_ 1 2.073 In Qt- 1) and at have been previously defined and Co = In 0. 22 Again, we included the dependent variable lagged one quarter among the regressors to control for serial correlation due to capital partial adjustment. While we expect that cl and c5 > 0, c2 and c6 < 0, we cannot predict the sign of the coefficient of rr~: it all depends on whether the positive effect of the interest rate volatility due to In (1 - o'~) in Equation (30), as measured by the coefficient ca, is greater, equal to or smaller than the negative effect due to ln(1 + 1) in Equation (30), as measured by the coefficient c4 .za Since our real interest rate series contains negative values, we estimated Equation (32) without taking the logs of rt and, consequently, of a~. OLS estimates of Equation (32) are shown in Table 3. 24 Two different versions of model (32), with and without demand volatility (Table 3, columns 1 and 2, respectively), have been estimated, since results which exclude a s show that the model is misspecified: both the Durbin-Watson test and the Lagrange multiplier test reveal the presence of serial correlation in residuals. When model (32) was estimated by the AR1 procedure, only c 1 and c5 were statistically significant. Therefore, the augmented version of model (32) can also be seen as a test of the robustness of model (28) and suggests the importance of simultaneously looking at real and financial factors when studying investment.
-
ZZlndeed, In Ks and In Y, are cointegrated with vector (1, -2.073). ~eWe also estimated model (32) using In Kt-1 and In Qt-1 as separate explanatory variables and results were statistically identical. :aThese two effects are due to the convexity of the present value function with respect to rt and the option value, respectively. 24In a world (or European) economy with integrated capital markets, we expect that rt is determined by a hypothetical world rate, or by the interest rate prevailing in some of the strongest economies like the U.S. or Germany. This fact would eliminate or reduce the simultaneity problem between investment and interest rates in smaller countries like Italy and, therefore, OLS estimates would still be unbiased.
508
1. ~-K Const.
/zt
(a~),
2, L
( - 3.72)
rt
-
( - 1.69)
1.002"
(23.36) 0.544 (0.95) 0,964 0.087 1.288 0.007* 0.027*
1
0.976*
(29.49) - 0.402 (-0.97) 0.973 0.076 1.558 0.108 0.187
EftR2
(In K - 2.073 In Q)
62
62
NOTE: Student-t values are in parentheses. Significance level: * = 0.05. Model (32) estimates show that investment is more affected by dem a n d variables than financial variables. Both the d e m a n d variables' coefficients have the expected sign and their size is close to values estimated in Table 2, even if the two models were estimated over two different sample periods. T h e real interest rate shows the traditional negative relationship with investment, even though its impact is smaller than that of the d e m a n d variablesY and its coefficient is statistically significant only at the 10% 25Indeed, beta coefficients for ,tit, cr~tand rt are 0.116, -0.114 and -0.042, respectively. 509
4. Conclusions In this paper we studied a stochastic model of investment in the presence of interest rate and demand uncertainty, which allowed us to analyze the role of variable levels, together with their volatility in affecting firms' accumulation decisions. From a theoretical point of view it is important to distinguish between the cases of reversible and irreversible investment. In the former case, while an increase in interest rates and expected demand has the traditional effect on investment (negative and positive, respectively), an increase in interest rate volatility always has a positive effect. The effect of demand volatility depends upon the type of returns to scale assumed: higher demand volatility determines higher (lower) investment demand when returns are increasing (decreasing). When returns to scale are constant, demand volatility has no effect on investment.
z~vVethank one of the two refereesfor pointingout the need for this cheek. 510
OLS Estimates for Model Shown in Equation (32) with FirstDifferenced Variables, 1980:iii-1995:iii.
T A B L E 4.
1. a I_ K
Const.
(2.55)
0 . 3 4 1 "
-9..O7) -0.019
-
1.03) 0.028
rt-1
- 0.033
-
1.67)
- 0.035 -0.16)
511
References Annunziato, P., P. Manfroni, and G. Rosa. "La stima del eapitale per settore e area geografica e alcuni indici di produttivit~." CSC Ricerche 66 (1992). Bernanke, Ben. "Irreversibility, Uncertainty and Cyclical Investment." Quarterly Journal of Economics 98 (1983): 85-106. Caballero, Ricardo J. "Aggregate Investment." Forthcoming in the Handbook of Macroeconomics, 1997. Calcagnini, Giorgio, and Lewis T. Evans. "Modelling Short-Term Interest Rates in Italy." Unpublished WP, University of Urbino, 1995. Carlson, John A., and Michael Parkin. "Inflation Expectations." Economica 42 (1975): 128-38. Chan, K. C., G. Andrew Karolyi, Francis A. Longstaff, and Anthony B. Sanders. "An Empirical Comparison of Alternative Models of the Short-Term Interest Rate." Journal of Finance 47 (1992): 1209-27. Cox, John, Jonathan Ingersoll, and Stephen A. Ross. "An Analysis of Variable Rate Loan Contracts_" Journal of Finance 35 (1980): 389-403. Dasgnpta, Susmita, and Kajal Lahiri. "'On the Use of Dispersion Measures from NAPM Surveys in Business Cycle Forecasting." Journal of Forecasting 12 (1993): 239-53. Dixit, Avinash. The Art of Smooth Pasting. Harwood: Academic Press, 1993. 512
514