RAPPORTER FRA STATISTISK SENTRALBYRÅ 83/27
GROSS CAPITAL, NET CAPITAL, CAPITAL SERVICE PRICE, AND DEPRECIATION A FRAMEWORK FOR EMPIRICAL ANALYSIS
BY ERIK BIØRN
STATISTISK SENTRALBYRÅ OSLO KONGSVINGER 1983 ISBN 8253719957 ISSN 03328422
EMNEGRUPPE Economic and Statistical Theory and Analysis
PREFACE
Time series of capital stocks play an important role in macroeconomic modelbuilding and analysis. They are also basic elements in the calculation of depreciation in the different production sectors for national accounting purposes. This report presents a theoretical framework for the construction of capital stock figures from investment data. The results will be utilized in an empirical project which has been started recently in the Central Bureau of Statistics.
Central Bureau of Statistics, Oslo, 8 November 1983
Arne Olen
FORORD
Tidsserier for kapitalbeholdninger spiller en viktig rolle i makrookonomisk modellbygging og analyse. I arbeidet med nasjonalregnskaper beregnes slike tidsserier blant annet som ledd i beregningen av kapitalslitet i de enkelte produksjonssektorer. I denne rapporten presenteres et teoretisk opplegg for beregning av kapitaltall på basis av investeringsdata. Resultatene vil danne grunnlaget for et empirisk analyseprosjekt som nylig er satt i gang i Byråets forskningsavdeling.
S tatistisk Sentralbyrå, Oslo, 8. november 1983
Arne !Dien
CONTENTS Page Abstract
6
1.
Introduction
7
2.
The gross capital: Capital as a capacity concept. Retirement (replacement)
3.
Two useful functions
10
4.
A probabilistic interpretation
5.
The net capital: Capital as a wealth concept. Capital service price. Depreciation
6.
The relationship between depreciation, gross capital, net capital, and capital service price  further results
7. Parametric survival functions
8
11
13 18
20
Appendices A.
The price interpretation of depreciation
65
B.
Proof of the recurrence formulae (70) and (84)
67 68
References Issued in the series Reports from the Central Bureau of Statistics (REP)
69'
6 ABSTRACT The construction of time series for capital stocks from data on gross investment is an essential element in the analysis of the firms' investment behaviour as well as in national accounting. In this report a general framework for the construction of such data is presented. Two capital concepts are involved  the gross capital  representing the capital's capacity dimension  and the net capital representing its wealth dimension. The two associated concepts retirement (replacement) and depreciation are also dicussed, as is the formal relationship between the measurement of the capital volume and the measurement of the price of capital services. Finally, we propose and discuss some parametric survival profiles which may be useful in empirical applications.
7 1. INTRODUCTION *) The measurement of real capital has been characterized as "one of the nastiest jobs that economists have set to statisticians" (John R. Hicks (1969, p. 253)). Closely related to it is the problem of measuring capital services, capital value, capital prices, capital service prices, and depreciation. The problem is not only one of measurement in the narrow statistical sense  a substantial part of the difficulty lies in the definition of useful concepts for empirical work. The reason for this lies in the fact that capital as an economic theoretical concept has at least two 'dimensions'. First, it is a capacity measure, a representation of the potential volume of capital services which can be 'produced' by the capital existing at a given point of time. Second, it is a wealth concept; capital has a value because of its ability to produce capital services today and in the future. The former concept is the one usually needed for production function studies, analyses of the firms' investment decisions, research on productivity issues, etc. The latter concept will be involved in analyzing the profitability
.
of the production sectors, financial market studies, national accounting, etc. Obviously, both concepts have relevance to the building of largescale macroeconomic models  a priori, there is, of course, nothing which implies that they should be numerically equal. In this paper, we give a theoretical framework for constructing capital stock data (and data on related variables) from data on gross investment. Our approach will be a fairly general one, in that we work with generally specified survival profiles in all sections but one. Attention will be focused on two capital measures: the gross capital, which indicates the instantaneous productive capacity of the capital objects, and the net capital, which indicates their prospective capacity. Both variables can be constructed from previous investment data by applying two different, but related, weighting schemes. This is also the case for the two derived variables retirement  which is related to gross capital  and depreciation  which is based on net capital. The fifth variable with which we .shall be concerned is the capital service price, which turns out to have a fairly close and empirically interesting relationship to the other variables. The problems and concepts involved in the measurement of capital are, to some extent, equivalent to those encountered in demography. We may consider capital as a 'population' of•capital units, associate investment with the 'birth' of a capital unit and retirement with 'death', etc. Demographic concepts as age, age distribution, survival probability, expected life time etc. are also useful when dealing with physical capital objects, and we shall make explicit reference to this equivalence at some places in the paper.
There are, however, notable differences, especially when it comes to the defini
tion of the wealth dimension of the capital stock, service prices,etc. Price variables, interest rates, and related concepts have, of course, no demographic counterparts. The paper is organized as follows: In section 2, we introduce the concept survival function and give a formal definition of the variables gross capital and retirement (replacement). Two functions which are convenient for the following discussion are introduced in section 3. In section 4, we interpret the model probabilistically and show, inter alia s that the auxiliary functions introduced in section 3 are closely related to the moment generating function of the probability distribution of the capital's life time. Section 5 is concerned with the capital value and the associated variables net capital and depreciation. A corresponding definition of the capital service price is also given. In section 6, we take a closer look at the relationship between gross and net capital, depreciation and capital service price, both in the deterministic and stochastic interpretation of the model. Finally, in section 7, we present a selection of parametric specifications of the survival functions which may be useful in empirical applications. First, we consider the familiar exponential decay hypothesis  which has the remarkable property that gross capital and net capital coincide. Then we discuss four classes of twoparametric survival profiles, two of which are convex, two are concave, and some of their most interesting special cases. In this paper, no attention will be devoted to the possible distortive effects of the corporate income tax system on the firm's investment decisions, through its impact on the capital service price. This issue is dealt with a related paper (Biorn (1983)), and we therefore disregard taxes altogether here. *) I wish to thank Petter Frenger and Øystein Olsen for their constructive comments on an earlier version of the paper, and Jørgen Ouren for his efficient programming of the computer routines.
8
2. THE GROSS CAPITAL: CAPITAL AS A CAPACITY CONCEPT. RETIREMENT (REPLACEMENT) Let J(t) denote the quantity invested at time t,measured in physical units or as a quantity index 1 ), where time is considered as continuous. More precisely,J(t) has the interpretation as the intensity of the investment flow at time t, and J(t)dt is the investment effectuated from time t to time t+dt. The proportion of an investment made s years (periods) ago which still exists as productive capital is denoted by B(s). The function B(s) represents both the physical wear and tear, and the time profile of the retirement of old capital goods. We shall consider it as a time invariant technical datum, in the following to be referred to as the technical survival function. In principle, B(s) may be decomposed as B(s) = B S (s)B E (s),
where B S (s) represents the relative number of capital units surviving at age s (the survival curve) and B E (s) indicates the efficiency of a capital unit of age s in relation to its efficiency at the time of investment, i.e. at age 0 (the efficiency factor). We shall not, however, make use of this decomposition in the following. We imagine that each capital good at each point of time contains a certain number of 'efficiency units', each having the same current productive capacity. The survival function B(s) indicates the relative number of efficiency units which are left s years after the initial investment was made. The function thus represents both the loss of efficiency of existing capital objects and physical disappearance, or retirement, of old capital goods. It is continuous and differentiable 2 ) and has the following properties: (1)
0 < B(s) < 1, B(0) = 1,
dB(s) < 0 d s 
for all s < 0,
lim B(s) = 0. sØ
A typical survival function, with a finite maximal life time N, is illustrated in figure 1 below. B(s) 1
•s FIGURE 1. A typical curvature of the technical survival fuction B(s). N = maximal life time
1) Assuming that J(t) is an aggregate of homogeneous capital goods. 2) At least in the interior of the interval on which B(s) is strictly positive. Confer figure 1 and the examples given in section 7.
9 The service flow from this capital stock is an argument in a static production function, together with labour services and other inputs, and we assume throughout than the units of measurement and the form of the production function are chosen in such a way that one capital (efficiency) unit produces
one unit of capital services per unit of time. Then (2)
K(t, ․ ) = B(s)J(t s)
s
>
0
has the double interpretation as the volume of the capital which is s years of age time t (i.e. the capital of vintage ts existing at time t) and the service flow produced at time t by capital of age s. Aggregation over capital vintages gives the following expression for the total volume of capital (flow of capital services) at time t: ^
(3)
t
^
K(t) = I K(t, ․ )ds = I B(s)J(ts)ds = I B(te)J(e)de 0
(e=ts).
0
Capital thus defined is a technical concept; K(t) represents the current productive capacity of the total capital stock at time t. We shall refer to it as the gross capital stock. Differentiating (3) with respect to t we find that the rate of increase of the capital stock can be written as 3 )
(4 )
Kt K(t)
)
t B(0)j(t) __ dK ( ) = dt
+
fdB(t e) dt J( e ) de =J(t) ^
+
f d B s J(ts)ds s 0
^
CO
= J(t) 
I 0
b(s)J(ts)ds,
where
b(s) = (5)
dB
i which implies, since B(0)=1,
B(s) = I b(z)dz s
s>
0.
The volume of capital worn out or scrapped (i.e. the number of efficiency units which disappear) at time t is the difference between J(t), the gross investment, and the rate of increase of the (gross) capital stock. From (4) we find that the volume of retirement at time t can be expressed in terms of the previous investment flow as follows: •
(6)
CO
D(t) = J(t)  K(t) = f b(s)J(ts)ds. 0
We can alternatively call D(t) the volume of replacement investment at time t, since it represents the number of efficiency unit which would be required to replace retired equipment. The function b(s) indicates the structure of the wear and tear and scrapping process: b(s)ds is the share of an initial investment of one unit which disappears from s to s+ds years after the time of installation. From (1) and (5) it follows that b(s) is nonnegative for all s and that CO
(7)
I 0
b(s)ds =
1.
3) We utilize the following general formula for differentiating an integral: d b(t)b(t) I dt I a(t) f(t,e)de = b (t)f{t,b(t)}  a'(t)f{t,a(t)} + a(t)
i7 f(t,e)de.
10 This equation expresses the fact that all equipment installed will disappear sooner or later. We shall call b(s) the (relative) retirement (replacement) function in the sequel. Formulae for gross capital and retirement similar to (3) and (6) can be found in e.g. Jorgenson (1974, pp. 191192), and Hulten and Wykoff (1980, p. 100). The terminology,however, does not seem to be consistent in the literature. Some authors (e.g. Steele (1980)) define gross capital as the cumulated volume of past gross investment flow over a period of length N, the capital's life time, i.e. in our notation N K(t) = I J(ts)ds. 0 Others, e.g. Young and Musgrave (1980), use gross capital as synonymous with the capital measure derived from the perpetual inventory method, in stating that "gross capital stock for a given year [is obtained] by cumulating past investment and deducting the cumulated value of the investment that has been discarded". (Young and Musgrave (1980, pp. 2324)). In our notation, this corresponds to CO
K(t) = I B S (s)J(ts)ds. 0 This definition is also used by Johansen and Sørsveen (1967, p.182). It coincides with our definition (3) if B(s)=B S (s), which implies B E (s) = 1 for all s>0, i.e. if the efficiency of the surviving capital goods is the same for all vintages. 4 ) If, moreover, B
(s) = 1 for O 0,
sz0,
B(s)
where p is a positive constant, T =zs, and B(s) and b(s) are defined as above. The numerator of Ø p (s) is the present value of the total flow of capital services produced by one initial unit of capital from the time it passes s years of age until it is scrapped, discounted to the time when it attains age s with a rate of discount equal to p. The denominator represents the share of the initial investment which attains age s. 5 ) The ratio Øp(s) may thus be interpreted as the discounted future service flow per capital (efficiency) unit which is s years of age. Similarly, (s) has the interpretation as the present '4 p value of the remaining retirement flow per capital unit which is s years of age. We then have in particular that 00
Ø p ( 0) = I e Pz B(z)dz 0
4) Our definition corresponds to the efficiency corrected capital stock as defined in section 4 of the JohansenSørsveen paper. 5) Or, more precisely, the relative number of efficiency units left s years after the time of investment.
11 is the present value of the total service flow from one new capital unit, and 00
4) P (0) = I e Pz b(z)dz 0 is the present value of the total replacement flow related to one new capital unit. At this stage, however, it is not necessary to attach an economic interpretation to the functions Ø P (s) and ij (s) and the parameter p; they may be considered as purely mathematical entities. Note, in particular, that we have said nothing so far about the possible relationship between
p
and
economic market variables. Obviously, Ø P (s) and ip(s) are both decreasing functions of
p
for all values of s. From (5)
and (9) it follows that
*0 ( s ) = 1
for all s.
If p>0, it is easy to show, by using integration by parts, that
(10) Ø P (s) _ ^P {l iU (s)} 
s >O , P>O.
P
All expressions which can be written in terms of 4(s) can thus be written in terms of ii P (s), and vice
versa. Differentiating (8) with respect to s, we find
b(s (11) ØP` (s
fe
s 0
t
j
B(T +s)
t B(s)
_ b(T +s) dT. b(s)
This expression will be negative  i.e. Ø P (s) is a decreasing function of s  if the integral in (11) is negative. Then p P (s) will be an increasing function of s, cf. (10). A sufficient condition for this to hold for all s, regardless of the value of
(12)
B(T +s)
B(s)
p,
is that
^ b(T +s) b(s)
for all s and T>0.
In the next section, we give an interesting probabilistic interpretation of Ø P (s) and i P (s).
4. A PROBABILISTIC INTERPRETATION So far, we have considered the process generating the deterioration and retirement of the capital units as a deterministic process and we have established the functions B(s), b(s),4 (s), and 4) (s) on this basis. In this section, we shall give an alternative probabilistic interpretation and establish a correspondence between the two interpretations which will be useful for later reference. 6 ) When a capital good is installed, the investor does not normally know its actual life time. Ex ante it may be considered as a stochastic variable S, the function B(s) representing the survival probabilities, i.e. B(s) is the probability that a new capital good 7 ) will survive for at least s years, (13)
B(s) = P(S>s)
s?0.
Since B(s) is continuous, the distribution function of the life time is P(S0.
b(s)= s {lB(s)} =  B'(s)
The variable S represents the total life time of a capital good. Consider also the remaining life time of a capital good which has already attained age s, i.e. T=Ss. Using basic rules in probability calculus, we find that (14)
P(T>TiS>s) = P(S>T+sIS>s) = BB T+s ) =
s
B(TIS),
sz0, Tzs ,
)
where B(TIs) is defined by the last equality. The conditional density function of the remaining life
time of capital which has attained age s is thus (15)
b(Tis)
 dB(T►s) = b T+s) dT B(s)
s >0, T >0.
(
When this probabilistic interpretation of the retirement process is adopted, the share of a population of capital goods (efficiency units) which survive s years after investment will converge towards B(s) with a probability of one as the number of capital goods increases, according to the "law of the large numbers"  i.e. the former is a consistent estimator of the latter. Correspondingly, b(s)ds is (approximately) the proportion of the capital goods (efficiency units) whose life time is between s and s+ds years, and b(01s)ds = b(s)ds/B(s) represents the proportion of the capital goods having attained age s which will disappear before age s+ds. The latter is thus a formal analogue to the concept 'mortality rate' in demography, i.e. the probability that a person of a certain age will die during a given future period, e.g. the next year. Which interpretations can then be given to the functions Ø (s) and i (s), defined in eqs. (8)
P
P
and (9)? Let us first recall the definition of the concept Laplace transform. The Laplace transform of a stochastic variable X with a density function f(x), defined on [0,..), is 8) 00
(16)
L f (x) = fe xx f(x)dx, 0
where A is a parameter. Letting E denote the expectation operator, this is equivalent to
(17)
Lf(X) = E(e").
Using (15), eq. (9) can be written as
(18)
*P(s) = fe 0
pT b(Tl_s)dT.
This is an expression of the form (16), with f(x) set equal to b(Tls) and x set equal to p. Thus i p (s)
stochastically interpreted is simply the Laplace transform of T=Ss, the remaining life time of a capital good which has attained age s. This expression represented the present value of the remaining retirement flow per capital unit of age s in the deterministic interpretation of the model. Eq. (18) can alternatively be written as (cf. (17))
( 19 )
P(Ss) IS>s) = E(e PT ; s), P (s) = E(e
using ";s" as a shorthand notation for "IS>s". For s=0 we have in particular
( 20 )
b P ( 0 ) = L (p) = E(e PS ),
8) See Feller (1966, Ch. XIII.l). The Laplace transform has a close relation to the moment generating function of the distribution. The moment generating function of X is simply L f ( a) = E(e ). Confer Feller (1966, p. 411), or Cox (1962, p. 9).
i.e. iP p (0) stochastically interpreted is the Laplace transform of the total life time of a new capital unit, S. Expanding e pT in (19) by Taylor's formula, we obtain
(21)
= E (1 p T +
i
= 1 +
E
i =1
^T _
p2

p6 3 T 3 .
2 
. . . ,• s)
. . , (1) 1 pE (T i ; s),
If we combine (21) and (10) we find
(22)
.
E (_ 1) ^ 1 i =2
Ø(s) = E(T; s) + P
p i1 r
E(T^ ,^ s
s >0.
By using this equation we can determine all the moments of the (conditional) distribution of the remaining life time T once we know the function Ø p (s) for a value of p different from zero. All information about the distribution of T is thus "condensed" in this function. If p=0, the second term of (22) vanishes i.e. all moments of second and higher order are "swept out"  and we get simply
(23)
E(T; s) = Ø (s)  o
1 f B(z)dz. ^`"j ^s s
For s=0, we have in particular
(24)
E(S) = E(T;0) = 4) 0 (0) = 1' B(z)dz. 0
Equations (23) and (24) reveal an interesting correspondence between the deterministic and the stochastic interpretation of the replacement process: What emerges as the undiscounted future service fl ow from one capital unit of age s in the deterministic framework 9) is the expected remaining life time of a capital unit of age s in the stochastic version of the model, and vice versa. In particular, the total service flow from a new capital unit, deterministically interpreted, finds its counterpart in the expected total life time in the probabilistic interpretation.
5. THE NET CAPITAL: CAPITAL AS A WEALTH CONCEPT. CAPITAL SERVICE PRICE. DEPRECIATION Gross capital as defined in section 2, by aggregating the surviving shares of the different capital vintages expressed in efficiency units, is a capacity concept: K(t) represents the number of capital (efficiency) units at time t on the one hand, and the instantaneous service flow from this capital stock on the other. We now consider the vaZue dimension of the capital. The market value of the capital goods will, in general ,reflect the cost of producing new investment goods on the one hand, and the capital users' expectations about future productivity on the other. Let q(t) denote the price of investment goods at time t. The value of the investment outlay is then q(t)J(t), which is, of course, also the value of the new capital installed at time t. The value of an old capital good does not, in general, reflect its historic cost, but rather the service flow that it is likely to produce during its remaining life time. Let q(t, ․ ) be the price of one capital unit (efficiency unit) of age s at time t and K(t, ․ ), as before, the number of such units. The value of the capital which is of age s at time t is then V(t, ․ ) = q(t, ․ )K(t, ․ ), and the value of the total capital stock can be written as 9) Confer the interpretation of (8) above.
14 00
(25)
00
00
V(t) = I V(t, ․ )ds = I q(t, ․ )K(t, ․ )ds = I q(t, ․ )B(s)J(ts)ds, 0
0
0
the last equality following from (2).
The decomposition of V(t, ․ ) into a price and a quantity component is however, in a sense arbitrary. An alternative decomposition is V(t, ․ ) = p(t, ․ )J(ts), where p(t, ․ ) = q(t, ․ )B(s) has the interpretation as the price of capital of age s at time t per capital unit originally invested at time ts.
10)
The corresponding expression for the capital value,
(26)
V(t) = I V(t, ․ )ds = I p(t, ․ )J(ts)ds, 0
0
will be convenient for the purpose of defining depreciation, as we shall see in appendix A. How is q(t, ․ ), or p(t, ․ ), determined? A reasonable assumption is that q(t, ․ ) is an increasing function of the current investment price (the replacement price) q(t) for all s>0, and a decreasing function of the age s for each given t  the older a capital unit is, the lower will its price be, cet. par. Obviously, we have q(t,0) = p(t,0) = q(t), and V(t,0) = c;(t)J(t). In this paper, we shall make the specific assumption that the relative prices per unit of capital goods of different ages perfectly reflect the differences in their prospective service flows. More precisely, the price per unit of the (discounted) future flow of capital services is assumed to be the same for all capital vintages at each given point of time. Interpreting p as the rate at which future capital services are discounted (cf. section 3), we can formalize this hypothesis as
(27)
Ø p (s) )
Øp
for all t and all s>O.
^^^
It implies a sort of 'lain of indifference' to hold between the different capital vintages: A firm buying at time t a capital unit (efficiency unit) of age s at the price q(t, ․ ) pays the same price per unit of discounted prospective capital services as a firm which buys a new capital unit at the price q(t). If (27) is satisfied, the firm will be indifferent between expanding its capital stock by investing in new and old equipment, or by changing the age composition of the capital stock by investing in one vintage and disinvesting in another.
(28)
11)
The common price per unit of (discounted) capital services is
c(t) = q(t) _ q(t) Ø^ p s fe 0
B(s) ds
The 'law of indifference' (27) can alternatively be stated in terms of the price p(t, ․ ) _ q(t, ․ )B(s). ( 29 )
It then says
. p(t, ․ ) I e p(zs) s
_ ^
q(t) for
all t and all s >0,
e p Z B(z)dz
I 0
i.e. p(t, ․ ), considered as a function of s, declines in equal proportion to the decline in the discounted remaining flow of capital services. 10)
A third decomposition would be the following: Let B(s) = B S (s)B E (s), where B s (s) represents the
survival curve and B E (s) the efficiency factor. We could then interpret B
s (s)
as belonging to the
quantity component and B E (s) as belonging to the price component of V(t, ․ ). The price variable, gE(t, ․ ) = q(t , ․ )BE(s) = p(t, ․ )/B s (s), would then represent the price per capital unit of age s at time t, corrected for loss of efficiency. 11) The latter conclusion, of course, presumes a neoclassical (puttyputty) production technology, with full substitutability between the different capital vintages.
15 We may interpret p as the rate of interest forgone by a producer who owns the capital and uses its services instead of purchasing interestbearing financial assets. If we set p=ry, where r is the nominal interest rate and y is the rate of increase of q, and if r and Y are constants, then (28) is equivalent to
q(t)
=
ferzc(t+z)B(z)dz. 0
This equation agrees with the firstorder conditions for maximization of the present value of cashflow in a neoclassical model of producer's behaviour, when we replace c(t+z) by the value of the marginal productivity of capital at time t+z,12) For the majority of capital goods, neither second hand markets nor hire markets exist, i.e. q(t, ․ ) (or p(t, ․ )) and c(t) cannot be observed as market variables for s>0. The 'law of indifference' (27)  (29) is then no testable hypothesis; rather, it may be considered as providing an implicit definition of q(t, ․ ) (or p(t, ․ )). It gives a procedure for constructing series for q(t, ․ ), and corresponding indices for c(t), under perfect market conditions, from observed values of the investment price q(t) and given values of the survival rates B(s) and the rate of discount p.13) Returning for a moment to the probabilistic interpretation of the deterioration process, we find, by using (22), that (27) can be expressed in terms of the moments of the distribution of the capital's life time as follows
q(t, ․ )
(30)
_ q(t) for all t and all s>0.
E(T;s) + E(1) 1l P i 1 (S) + E(1) i1 i1 i E i. E(S ) i =2 i.' E(T ' s) i =2
This equation has particular intuitive appeal in the case where the discounting rate
p
is zero. The 'law
of indifference' then simply says that the relative prices of the different capital vintages are equal
to the ratios of their expected remaining life times: q(t, ․ ) _ E(T;s
q(t)
for all t and all s>0, p=O.
E(S
Combining (25), (27), and (28), we find that the value of the capital stock can be written as
(31)
V(t) = q(t)
0
f
Ø P (s)B(s) J(ts)ds = c(t) f Ø P (s)B(s)J(ts)ds. 4)p(0) 0
This equation gives a procedure for computing the capital value from data on q(t), J(ts), B(s), and P
It also indicates two alternative ways of decomposing this value into a price and a quantity com
ponent. First, if we define the price component as equal to the current investment price, the quantity component becomes CO
(32)
U t = 1 f q(t, ․ )K(t, ․ )ds = K(t) = q t N q(t) 0
CO
I G (s)J(ts)ds, 0 P
where OD
= (33) GP(s) 
Ø P (s)B(s) Ø 
I
e P(zs)
B(z)dz s>0.
P f
0
e P z B(z)dz
12) See Bjorn (1983, appendix) for a demonstration of this in a more general context. 13) In the rather few cases where q(t, ․ ) (or p(t, ․ )) are observed market variables  e.g. cars, office buildings, and dwellings  eq. (27) ( or (29)) can be used to estimate  4) 10 (s) and hence, given the rate of discount p, draw conclusions on the form of the underlying survival function B(s). Examples of analyses of this sort are Hall (1971) and Hulten and Wykoff (1981).
16 We see that K N (t), like K(t), is constructed by aggregating the previous investment flow, but the weighting system is basically different. The weight assigned to investment made s years ago in KN (t), G P (s), is the share of the total discounted service flow produced by one unit invested after
it is s years old, whereas K(t) is based on the technical survival rates B(s). Or othervise stated, KN (t) is constructed on the basis of the prospective service flow, K(t) on the basis of the instantaneous service flow each capital vintage. From (33) we see that the weighting function G p (s) satisfies
(34)
dG P (s)
0 < G P (s) < 1,
= 1,
ds
0,
lim G(s) = 0, sØ
i.e. it has the same qualitative properties as B(s), cf. (1). Furthermore, it follows from (11) and (33) that G P (s)s)J(ts)ds (48) DN(t) =
E S =
0 (p=4 ),
^ I P(S>s)ds 0
i.e. depreciation is equal to gross capital divided by the expected life time of a new capital unit (first equality), or equivalently, equal to a weighted average of the past gross investment flow with the survival probabilities B(s) = P(S>s) used as weights (second equality). We get a similar relationship between the price variables. From (23), (24), (27), and (28), we find
(49)
c
_ qt E S }  E S s;s t,s
for all s (p=0),
i.e. the capital service price is equal to the market price per capital unit divided by its expected (remaining) life time. .And this equality holds for all capital vintages. (Again, when p>0, higher order moments should also be taken into account.) Third, as we noticed above (eqs. (41) and(44)) g p (s) has properties which suggest its interpretation as a density function. This function has the same formal relationship to the net capital K N (t) as the function b(s) has to the gross capital K(t). Since b(s), interpreted stochastically, is the density function of the life time of the gross capital, S, this motivates giving g p (s) the interpretation as the density function of the 'life time of net capital', S N . The formal definition of S N would then be
P(S N >s) = Gp(s)
.
for all s >0.
20 Using (41), we find that its expectation is in general 00
I sB(s)ip (s)ds (50)
P
E(S ) = I sg (s)ds  0 N 0 p
Ø p (0)
For p=0 we get in particular 00
I sB(s)ds
(51)
E(SN) = I s g 0 (s)ds = 0 0
2 O
(p=0).
B(s)ds
Thus defined, the expected life time of the net capital would then emerge as a weighted average of the life time with the survival probabilities used as weights. The latter equation can be given an interesting reformulation. Using integration by parts, it is easy CX)
to show that I sB(s)ds = E(S 2 )/2, provided that lim s 2 B(s) = 0. Hence, recalling (24), we find 0 sØ (52)
E(S )
N
E(S)
where
a2
2
= 1 E(S )= 1 2 [E(S)J 2
tE(S)J2+ Q 2 Q EE(S)32
S = 1 {1 +[ S J 2 } E(S)
(p 0 ) ,
is the variance of S. The ratio between the expected life time of the net capital as defined
above and that of the gross capital thus has its lowest value, 1/2, for
Q S =0, i.e. when there is no
uncertainty with respect to the life time of the gross capital; all units disappear at the same time. The ratio increases with the square of the coefficient of variation of the life time,
as/E(S).
If the
coefficient of variation is unity, the expected life time of gross and net capital coincide.
7. PARAMETRIC SURVIVAL FUNCTIONS The results derived in the previous sections are valid for any survival function B(s) which satisfies the general restrictions (1). In this section, we present aselection of parametric functions which may be useful for empirical applications. For each B(s) we derive the corresponding functions G p (s), Ø p (s), and
1
P
These functions can be used on the one hand for the quantification of gross
and net capital, retirement, depreciation, and capital service price
on
the basis of investment data 
on the other hand for estimating and testing hypotheses about the form of the survival function from data on vintage prices. We present four classes of survival functions, each characterized by two parameters. The first parameter represents the maximal life time of the capital, the second indicates the 'curvature' of the survival profile. Important special cases of these functions are also considered. The results will be presented partly algebraically, and partly in the form of tables and diagrams. For the sake of reference we shall, however, start by considering a one parameter survival function, namely the familiar specification with exponentially declining survival rates.
Exponentially declining survival function: B(s) = e Ss —
Consider the parametrization
(53)
B(s) = e
65
s>0,
where 6 is a positive constant. Probabilistically interpreted, the life time S then has an exponential distribution.
Inserting (53) in (5), (8), and (9), we find
21 (54)
b(s) = 6e6S,
(55)
(1)(0(s) p _ +6,
6 (56) i(s) p = +6
s >0.
This parametrization thus has the particular property that Ø p (s) and i p (s) are constants independent of
s. The (conditional) Laplace transform of the remaining life time is equal to the Laplace transform of the total life time for all ages s. Since
1
p+6
1 [1
6

p + S
(p)2

6
we find, by using (22), that
(57)
E(T;s) = E(S) = E(T 2 ;s) = E(S2) = Z ,
for all s >0,
62
and hence
(58)
var(T 2 ;s) = var (S 2 ) = E(S 2 ) .[E(S)] 2 = 1
for all s>0.
6
In this case, the remaining life time has a (conditional) expectation equal to 1/6 and a (conditional) variance equal to 1/6 2 for all s. From (33), (41), and (55) we find moreover that
(59)
Gp(s) = B(s) = ee5,
(60)
gp(s) = b(s) = se
6s
for s>0,
and hence, using (32) and (42), that 00
(61)
KN(t) = K(t) = Ie S5 J(ts)ds, 0
00
(62)
DN(t) = D(t) = I 6e 6S J(ts)ds = 6K N (t) = 6K(t). 0
These relationships hold regardless of the value of the discounting rate p. Thus, in the exponential
case, gross capital is numerically equal to net capital, and retirement (replacement) coincides with depreciation1 6) The rate of retirement is equal to the rate of depreciation, and the common value is constant and equal to I. This is another particular property of this survival function. Its implication for the price variables is also remarkably simple. From (27) and (55) it follows that
(63)
q(t, ․ ) = q(t)
i.e. the price per capital
for all s>0,
efficiency unit will be the same for all ages. The equivalent relationship ex
pressed in terms of the price per capital unit
originally invested is
16) This conclusion concurs with eq. (52) which implies that E(S
N)
= E(S) when the coefficient of
variation of S is unity. This is in fact the case for the exponential distribution, since' the expec
tation and the standard deviation are both equal to 1/6 in this case, cf. (57) and (58).
22 (64)
p(t, ․ ) = e ss q(t),
for all s>0,
i.e. this price declines exponentially with age at the rate S. Combining (28) and (55) we find that the capital service price is equal to
(65)
c(t) = q(t)(p+S) •
If we let o=r  ci (t )/q (t) , r denoting the nominal market interest rate, i.e.
i f we make the reasonable equili
brium assumption that the capital users (capital owners) consider the current 'real interest rate' when discounting the future flow of capital services from time t (confer section 5), this expression is identical with the familiar textbook formula for the user cost of capital in a neoclassical model of capital accumulation,
c(t) = q(t) {r+ sg(t)/g(t)1 . From the point of view of empirical applications, the exponential model is very restrictive since it has only one parameter. Its implicit assumption of an infinite maximal service life is also inconvenient and implausible, as is the constancy of the rate of depreciation which it imposes. In the following, we outline four classes of twoparametric survival functions with a finite maximal life time, two of which are convex and two concave.
CLASS I: Convex: B(s) _ (2 1) n First we consider for 0 1
and hence
p(Ns) 1 g p (s)_pN _ Pe p40' N 1 e
•
Finally, when n=0, (76) gives
E(S N ) 1 _
(P=0),
E(S)
which agrees with (52), since the simultaneous exit specification implies, as already remarked, that the life time S has a one point distribution, and, consequently, Q S =O. The ratio of E(S N ) and E(S) cannot take a lower value than it does in this case, so the simultaneous exit assumption is also in this respect an extreme specification. n=1: Linear survival function
When n=1, the survival function is a linearly decreasing function of s, B(s) = 1 
N
,
b(s) = 10 N  {1  e }/p
The vintage prices and the capital service price are in this case
(Ns)p{1 e  p (Ns) 1 N sHp(Ns)
q(t, ․ ) = g(t) _pN s }](1  171) CNp  { 1  e
p(t , ․ ) = q(t)
(N s)p{le p(Ns) _p N Np{ 1e }
c(t) = q (t) pe pN }/(Np) 1 {1 

= cl( t )
[NHp(N)1(1 ^)
NsH p (Ns)
}
= q(t) N
H (N) p
p
= q(t) 1 N H p (N)
Finally, from (76) we find
E(S N ) `_ 2 3 E(S)
(p=0),
i.e. when the survival function is linearly decreasing, the expected life time of the net capital will be two thirds of the life time of the gross capital.
Strictly convex survival functions All members of this class of survival functions in which n>2 are (strictly) convex functions of the age s. Or stated otherwise, the relative retirement (density) function b(s) is a decreasing function of s, since (67) implies
db(s) _ _ n n 1 ds
(1 _ s ) n 2 < 0
N
N
for n>2.
Moreover, b(s) is itself convex for n>2, since
d 2 b(s) _ n(n 1)(n2)
s^n 3 > 0
ds 2 N 3
for n>2.
This situation is illustrated in the upper half of figure 2. In the limiting case where n goes to infinity while N is fixed, the survival function degenerates to
B(s) = B I (s;N,^) =
1 for s=0 0 for s>0,
i.e. the capital is scrapped momentaneously once it has been installed. On the other hand, if n and N both go to infinity while their ratio is restricted to be a finite constant
s,
i.e.,
27 n N n
*
00
^
CO
N= we find from 66
B(s)
and (67), when we recall the definition of e,
= (1 
N ) n
N
N
e ss
,
and hence, using (68) (or (70)),
I
C p (s;N,n)0
e Ss P+s
•
This limiting case is thus simply the exponential case discussed above.
CLASS II: Concave: B(s) = 1

(!) 777
The second class of survival functions we shall consider is
(80)
B(s) = R IT (s;N,m) =
1(N) m
for 00, i.e. a specification with instantaneous scrapping of the capital. When 0 0 ' 2 ds 1e The basic curvature of this class of functions is thus the same as in class I for 2 0.
61
CLASS IV: Concave: Inverse truncated ex2onential We can generate a fourth class of twoparametric survival profiles by reversing the sign of the parameter 6 in class III. This, of course, also implies a reversing of its curvature. Let Y . where y is defined to be positive. This gives the survival function
il r
eN_ e 5 eYN
(108)
B(s) = B IV (s;N,Y) =
for 00, i.e. a specification with infinite service life and no deterioration of the capital. Second, when Y 0, (108) degenerates to the linear function B(s) = 1s/N. ,
Third, when YØ, we get the simultaneous exit specification (B(s) = 1 for 0