Académique Documents
Professionnel Documents
Culture Documents
D
n
Y
iD1
(q
i
r
i
)
bi
. or
UD
n
X
iD1
b
i
log(q
i
r
i
). (3)
where U D logU
, b
i
is the marginal budget
share or marginal propensity to consume for
product i of supernumerary in any period and
b
X
iD1
b
i
D 1, q
i
is the supernumerary quantity
purchased of the product i during the period,
andr
i
is the minimumnecessary quantityof the
product i in the period. (1) includes both sub-
sistence income and supernumerary, that is
p
i
Q
i
D p
i
r
i
Cp
i
q
i
D p
i
r
i
Cb
i
(Y
X
p
i
r
i
). (4)
Differentiating p
i
Q
i
with respect to Y
P
p
i
r
i
gives
op
i
Q
i
o(Y
P
p
i
r
i
)
Db
i
. (5)
By dividing (4) by price, p
i
, and introducing
the time period concept, the basic formulation
of demand function for product i is obtained as
follows:
Q
i
Dr
i
C
b
i
p
i
(Y
X
p
i
r
i
). (6)
(6) is a static formulation of demand, which
depends on the minimum necessary quantity,
the marginal budget share for the product, the
price and the money income.
94 Byungryong Kang et al. ETRI Journal, volume 18, number 2, July 1996
Now a dynamic model can be considered
with a time lag to examine characteristics of
changes in the quantity demanded. Let Q
t
be
the quantity to be purchased for product i dur-
ing time t. As stated in the assumptions of pre-
vious Section III, the quantity of product i pur-
chased in any period depends on the level of
money income, relative price, stock adjusted
for depreciation or habit of consumption, and
preferences or tastes. With these assumptions,
the basic demand (or purchase) function is for-
mulated as follows: (For a notational simplic-
ity, i is suppressed.)
Q
t
D f (r
t
. Y
t
. P
t
. S
t
. T
t
). (7)
where S is the level of stock, and T is pref-
erences or tastes. (7) can be transformed into
a linear form of demand function [18], which
can be easily expressed as
Q
t
Dr
t
Cb
1
Y
t
Cb
2
P
t
Cb
3
S
t
Cb
4
T
t
. (8)
Thus, the quantity to be purchased during time
t can be represented as a form of distributed-
lag demand function as follows;
Q
t
DQ
t1
CLr
t
Cb
1
LY
t
Cb
2
LP
t
Cb
3
LS
t
Cb
4
LT
t
.
(9)
Substituting (6) for time t 1 into (9) yields
Q
t
Dr
t1
C
b
t1
p
t1
(Y
t1
X
p
t1
r
t1
)CLr
t
Cb
1
LY
t
Cb
2
LP
t
Cb
3
LS
t
Cb
4
LT
4
. (10)
Dividing both sides of (10) by the level of
potential quantities demanded by ultimate
adopters of the period, denoted by Q
p
, we
obtain the following rate equation of purchase
during time t
Q
t
Q
p
D
r
t1
b
t1
p
t1
Y
t1
X
p
t1
r
t1
Q
p
C(Lr
t
Cb
1
LY
t
Cb
2
LP
t
Cb
3
LS
t
Cb
4
LT
t
)Q
p
. (11)
Now let us transform all terms of (11) into
those with more common expressions so that
empirical estimation becomes feasible. First,
the second term in the rst bracket of (11) is
examined. The consumers money income in
(4) can be divided into two parts, subsistence
income and supernumerary.
r
i.t1
Q
p
C
b
i.t1
p
i.t1
Y
t1
X
p
i.t1i
r
i.t1
Q
p
DaQ
t1
Q
p
C(1a)Q
t1
Q
p
.
(12)
where a is the marginal budget share of in-
come for the minimum necessary quantity of
a group of products. In order to express the
above equation in terms of aggregate diffusion
rates, we make some mathmatical arrange-
ments. Multiplying both denominator and nu-
merator by Q
t
, the level of quantity which
consumers would eventually wish to hold at
current levels of income and price, the right-
hand side of (12) takes the following form :
Q
t1
(1a)
Q
p
D
Q
t1
Q
p
Q
t
Q
p
t
Q
p
aQ
t
Q
p
. (13)
Note that the quantity purchased in any pe-
riod is the sum of replacement demand, equal
to the depreciation or consumption of the pe-
riod, and net investment. Consumption is as-
sumed to be a continuous process which uses
part of both initial stock of the period and
purchases made during the period. Similarly,
due to consumers limitation of disposable in-
come, they purchase only a certain fraction of
the desired level of quantity which they would
ETRI Journal, volume 18, number 2, July 1996 Byungryong Kang et al. 95
eventually wish to hold at current levels of in-
come and credit. The purchase made during
the period is equivalent to the sum of the min-
imum necessary quantity and net addition of
the quantitywhich consumers wishto purchase
during the period. It is further assumed that
consumers are always likely to keep their mini-
mumunits of stock and the adjustment to a new
level of stock and consumption is normally
spread over more than a single period. There-
fore, the quantity purchased, Q
t
, is a fraction
of the stock, Q
s
t
, and the minimumpreliminary
arrangement stock for consumption at time t,
Q
s
t1
, is a fraction of the expected upper lim-
its of stock at time t, Q
t
. These relations can
be specied as Q
t1
D h
t1
Q
s
t1
and Q
s
t1
D
a
t1
Q
t
. For the present applications, the pa-
rameters, h and a, are assumed to be constant
over time. (13) can then be rewritten as
Q
t1
Q
p
Q
t
Q
p
t
Q
p
aQ
t
Q
p
D
h Q
s
t1
Q
p
Q
t
Q
p
t
Q
p
Q
s
t1
Q
p
. (14)
where Q
s
t1
Q
p
is the aggregate diffusion rate
revealed from potential purchases for product
i at time t 1, and Q
t
Q
p
is the upper limit of
the aggregate diffusion rate. Setting Q
s
t
Q
p
D
y
t
and Q
t
Q
p
Dk, (13) can be rewritten more
compactly as
Q
t1
(1a)
Q
p
D
h y
t1
k
(ky
t1
). (15)
Similarly, using properties stated above, a
compact expression can also be obtained for
the rst term of the rst bracket in (11) as fol-
lows:
r
t1
Q
p
Da h y
t1
. (16)
Now let us consider b
3
LS
t
Q
p
, the third term
of the right-hand side of (11). The stock at time
t, S
t
, can be expressed as
S
t
DS
t1
CQ
t
D
t
. (17)
where D
t
is the quantityof depreciation or con-
sumption over a period of time t. Introduc-
ing the rate of depreciation or consumption
for a certain product, d, which is assumed to
be a constant proportionate ratio, as a mean
value during its durable year, n, the relation-
ship among the stock, the purchase, and the de-
preciation or consumption can be expressed as
given below.
S
t
DQ
t
CQ
t1
C CQ
tnC1
[dQ
t
C2dQ
t1
C CndQ
tnC1
]. (18)
From a slight adjustment of (18), a difference
equation of stock responding the passage of
time as follows:
S
t
S
t1
DQ
t
d
tnC1
X
iDt
Q
i
. 1i n. (19)
where i is the time period backward, starting
from t. It is impossible to survey all of the
quantities purchased so as to nd the level of
depreciation or consumption. This suggests
that another approach be required for a practi-
cal purpose. By using an articial mean value
of the rate of depreciation over time in existing
stocks for the same product, an indirect mea-
surement method can be developed and an ap-
proximate value can be calculated. Since the
difference of depreciations between period t
and t1 in (19) is dQ
t
CdQ
t1
C CdQ
tnC1
,
the level of depreciation during period t has the
96 Byungryong Kang et al. ETRI Journal, volume 18, number 2, July 1996
following relations:
D
t
Dd
tnC1
X
iDt
Q
i
. (20)
Then, D, mean value of dQ
i
, can be proved
as having the following function of current
stock:
DD
D
t
j
D
d
j
tnC1
X
iDt
Q
i
D
2d
2 j dj
2
S
t
D S
t
. (21)
where
D
2d
2 j dj
2
1
b
3
(Q
t
r
t
b
1
Y
t
b
2
P
t
b
4
T
t
) j
(23)
Following this result, we convert
b
3
LS
t
Q
p
of
(11) as given below:
b
3
LS
t
Q
p
D
b
3
Q
t
Q
p
(24)
(Q
t
r
t
b
1
Y
t
b
2
P
t
b
4
T
t
) j
Q
p
Using (15), (16), and (8), (24) can be trans-
formed into the following simple expression:
b
3
LS
t
Q
p
D
h b
3
1a
y
t
k
(ky
t
)
h(j)
1a
y
t
k
(ky
t
) j.
(25)
where j(D j
0
Cj
1
Cj
2
Cj
4
) is the sum
of inuence coefcients from individual com-
ponents of the demand function consisting
of minimum necessary quantity (r
t
), income
(Y
t
), price (P
t
), and tastes (T
t
). Finally, let
us consider the remaining terms of (11). Since
these terms can be expressed as a certain por-
tion of
Q
t
Q
p
, they can be denoted by
nQ
t
Q
p
. That
is,
Lr
t
Cb
1
LY
t
Cb
2
LP
t
Cb
4
LT
t
Q
p
D
(n
0
Cn
1
Cn
2
Cn
4
)
(1a)
h
y
t
k
(ky
t
). or
LQ
t
b
3
LS
t
Q
p
D
n
1a
h
y
t
k
(ky
t
). (26)
where n(n D n
0
Cn
1
Cn
2
Cn
4
) is the sum
of inuence coefcients from the individual
components of the change in demand consist-
ing of change in minimum necessary quan-
tity (Lr
t
), change in income (LY
t
), change in
price (LP
t
), and change in tastes (LT
t
). Sub-
stituting (15), (16), (25), and (26) into (11)
yields the following nal formula for the pur-
chase rate relative to potential purchases:
Q
t
Q
p
Da h y
t1
Ch
y
t1
k
(ky
t1
)
C
h(b
3
Cn)
1a
y
t
k
(ky
t
)
h(j)
b
3
(1a)
y
t
k
(ky
t
) j. (27)
which can be referred to as the purchase rate
equation relative to potential purchases. This
ETRI Journal, volume 18, number 2, July 1996 Byungryong Kang et al. 97
equation implies that purchase rate is deter-
mined by level of stock and difference between
rate of depreciation or consumption () over
the useful life of a product and the sum of in-
uence (j) of income, prices, and tastes and
change (n) in them on purchase.
From (25), the formula for the aggregate
diffusion rate equation of current stock relative
to potential purchases can be approximately
derived the following continuous form:
y
t
D
S
t
Q
p
D
tnC1
X
iDt
LS
i
Q
p
D
Z
t
iDtnC1
h
1a
y
i
k
(ky
i
)
h(j)
b
3
(1a)
y
i
k
(ky
i
) j
di
D
k
1Ce
a
0
h(1)
1a
tC
h(1)(j)
2b
3
(1a)
t
2
. (28)
(28) is our nal formula for the aggregate
diffusion rate equation of the current stock of
the model, which can be widely applicable
for forecasts of the aggregate diffusion rate of
technological products in the market, the rate
of stock, and market penetration process of
various new goods and services. The rst ex-
ponent of (28) represents the level of initial
value for y
t
, the second exponent represents
the positive inuences by the changes of in-
come, price, stock, depreciation or consump-
tion, and tastes, and the third exponent repre-
sents their negative inuences by other substi-
tutes. The nal formula for the current stock
contain the term e
a
0
h(1)
1a
iC
h(1)(j)
2b
3
(1a)
i
2
, which
determines the limiting value of S
t
, That is, as
t goes innite, S
t
DkQ
p
, if (depreciation co-
efcient) -j (purchase coefcient), or S
t
D0,
if >j.
Therefore, S
t
seems to be asymptotic in
either manner. In the latter case, however,
asymptotic nature is not realized during the life
cycle of a product although S
t
eventually goes
to zero after a certain saturation point in the
very long run. f (t) D e
a
0
h(1)
1a
iC
h(1)(j)
2b
3
(1a)
i
2
,
S(t) D S
t
, and s(t) D LS
t
can be graphically
presented as one of the following two sets de-
pending on parameters involved. For example,
the left three graphs depict the case where S(t)
approaches a constant kQ
p
as time goes in-
nite, and the right three graphs depict the case
where S(t) approaches zero as time goes in-
nite, as shown in Fig. 3.
The model can accommodate different pat-
terns of technological substitution frequently
made in technological product market. It al-
lows the S-curve to be symmetrical and nonsy-
metrical as well as asymptotic and nonasymp-
totic, with the point of inection responding
to the substitution process. It incorporates de-
mand determinants for technological products.
By doing so, as far as technological products
are concerned, our model can be widely ap-
plicable for forecasting diffusion rate in target
market, rate of change in stocks and purchases,
and market penetration process.
V. EMPIRICAL TESTS AND
RESULTS
In order to illustrate the effectiveness and
generalityof our model, time-series data of dif-
fusion rate for some typical products in elec-
tronics and telecommunications market were
examined. We collected data from several
sources, most of which are published by the
98 Byungryong Kang et al. ETRI Journal, volume 18, number 2, July 1996
Fig. 3. Model curve.
respective governments, and selected some
durable and service products with available
data [20]-[22]. The durables included were
black and white televisions, color televisions,
personal computers, and VCRs in Korea and
Japan, and service products included were tele-
phone service subscribers, paging subscribers,
and cellular service subscribers in both coun-
tries. Furthermore, in order to assess the rela-
tive performance of the proposed model, these
products were also analyzed by using the ex-
isting famous models such as the Pearl Curve,
the Gompertz Curve, the Weibull Distribution
Model, the Bass Model, the NSRL Model, and
ETRI Journal, volume 18, number 2, July 1996 Byungryong Kang et al. 99
Table 1. Estimates of the proposed models parameter, MSE, and R
2
for some technological durable products.
Product Market
Period
covered
a
0
a
1
a
2
MSE R
2
Black and White TV Korea 1970-84 -4.1348 0.6632 -0.0292 0.0537 0.9538
Black and White TV Japan 1952-66 -6.7844 1.1620 -0.0403 0.0697 0.9898
Color TV Korea 1981-88 -2.1241 0.8161 -0.0218 0.0114 0.9965
Color TV Japan 1965-85 -6.2711 1.1676 -0.0306 0.0589 0.9950
Personal Computer Korea 1985-93 -7.4939 1.5383 -0.0652 0.0156 0.9981
VTR Japan 1977-89 -5.3646 0.4673 0.0009 0.0163 0.9961
Table 2. Estimates of the proposed models parameter, MSE, and R
2
for some technological service products.
Product Market
Period
covered
a
0
a
1
a
2
MSE R
2
Telephone service Korea 1976-93 -1.4749 0.1487 0.0079 0.0341 0.9732
Telephone service Japan 1947-89 -4.8878 0.1836 0 0.0337 0.9939
Paging service Korea 1984-92 -5.8283 0.2857 0.0343 0.0084 0.9979
Paging service Japan 1981-91 -1.4949 0.1236 0.0096 0.0053 0.9929
Cellular service Korea 1984-92 -7.5117 0.3480 0.0234 0.0110 0.9968
Mobile service Japan 1979-91 -7.5856 0.6549 -0.0118 0.1201 0.9735
the NUI model. The best model is considered
to be the one that shows lowest mean squared
error (MSE) and highest adjusted R
2
. The as-
sociated parameters of above models were esti-
mated throughappropriatelinear and nonlinear
regression analysis algorithms, like the least
squares method and Gauss-Newtonmethod. In
each case the deterministic model that best t-
ted the data was obtained. That is, these algo-
rithms searched for the parameter values that
most closely approximated the actual diffusion
curve. MSE is calculated by sample average
of sum of squared one-step-ahead forecasting
errors. Estimates of the presented models pa-
rameter, MSE, and R
2
for some technological
products are shown in Tables 1-2.
For each durable and service product se-
lected, MSE and R
2
are shown in Tables 3-
6. Tables indicate that proposed model t the
data on diffusion of durable and service prod-
ucts remarkably well in all cases. For all cases
applied, adjusted R
2
statistics of our model
show the higher values than any other exist-
ing model. Specically, the proposed model
100 Byungryong Kang et al. ETRI Journal, volume 18, number 2, July 1996
Table 3. Comparision of the proposed model and other diffusion models for durable products.
Product
Period
covered
R
2
and
MSE
Proposed
model
Pearl
curve
Gompertz
curve
Weibull
model
Black and 1970-1984 R
2
0.9538 0.7171 0.6845 0.8758
White TV (Korea) MSE 0.0537 0.3253 0.0870 0.1136
Black and 1952-1966 R
2
0.9898 0.9082 0.9509 0.9522
White TV (Japan) MSE 0.0697 0.5808 0.0844 0.2185
Color TV 1981-1988 R
2
0.9965 0.9916 0.9948 0.9894
(Korea) MSE 0.0114 0.0228 0.0090 0.0101
Color TV 1965-1985 R
2
0.9950 0.8954 0.9564 0.9660
(Japan) MSE 0.0589 1.1595 0.2576 0.1552
Personal 1985-1993 R
2
0.9981 0.9711 0.9917 0.9865
Computer (Korea) MSE 0.0156 0.2005 0.0109 0.0766
VTR 1977-1989 R
2
0.9961 0.9961 0.9513 0.9360
(Japan) MSE 0.0163 0.0150 0.0384 0.0360
Table 4. Comparision of the proposed model and other diffusion models for durable products.
Product
Period
covered
R
2
and
MSE
Proposed
model
Bass
model
NSRL
model
NUI
model
Black and 1970-1984 R
2
0.9538 0.2042 1.446 0.2847
White TV (Korea) MSE 0.0537 29.1563 31.3403 28.5492
Black and 1952-1966 R
2
0.9898 0.6679 0.7890 0.8408
White TV (Japan) MSE 0.0697 27.7152 17.5926 14.5140
Color TV 1981-1988 R
2
0.9965 0.9638 0.9677 0.9706
(Korea) MSE 0.0114 7.2773 6.4263 7.2799
Color TV 1965-1985 R
2
0.9950 0.9636 0.9899 0.9910
(Japan) MSE 0.0589 1.7918 0.4387 0.4678
Personal 1985-1993 R
2
0.9981 0.9109 0.9720 0.9799
Computer (Korea) MSE 0.0156 10.3232 3.6888 3.4698
VTR 1977-1989 R
2
0.9961 0.9551 0.9542 0.9565
(Japan) MSE 0.0163 2.5167 2.5686 2.7065
predicts the technological substitution process
from black and white TV to color TV very ac-
curately reecting actual replacement taking
place in the market. For other products, it took
as if there are little signicant improvements
of the proposed model compared with exist-
ing models. However, if those products are re-
placed by new products, which is commonly
ETRI Journal, volume 18, number 2, July 1996 Byungryong Kang et al. 101
Table 5. Comparision of the proposed model and other diffusion models for service products.
Product
Period
covered
R
2
and
MSE
Proposed
model
Pearl
curve
Gompertz
curve
Weibull
model
Telephone 1976-1993 R
2
0.9732 0.9634 0.9243 0.8992
service (Korea) MSE 0.0341 0.0424 0.0494 0.0539
Telephone 1947-1989 R
2
0.9939 0.9939 0.9509 0.8887
service (Japan) MSE 0.0337 0.0337 0.0962 0.3743
Paging 1984-1992 R
2
0.9979 0.9829 0.9148 0.8414
service (Korea) MSE 0.0084 0.0591 0.0336 0.5040
Paging 1981-1991 R
2
0.9929 0.9787 0.9549 0.8463
service (Japan) MSE 0.0053 0.0140 0.0139 0.0572
Cellular 1984-1992 R
2
0.9968 0.9886 0.9549 0.8544
service (Korea) MSE 0.0110 0.0355 0.0061 0.4222
Mobile 1979-1991 R
2
0.9735 0.9660 0.9601 0.9496
service (Japan) MSE 0.1201 0.1336 0.0116 0.1899
Table 6. Comparision of the proposed model and other diffusion models for service products.
Product
Period
covered
R
2
and
MSE
Proposed
model
Bass
model
NSRL
model
NUI
model
Telephone 1976-1993 R
2
0.9732 0.7566 0.7616 0.7172
service (Korea) MSE 0.0341 10.4510 9.6695 13.4948
Telephone 1947-1989 R
2
0.9939 0.8842 0.8815 0.8867
service (Japan) MSE 0.0337 0.9271 0.9487 0.9287
Paging 1984-1992 R
2
0.9979 0.9846 0.9888 0.9943
service (Korea) MSE 0.0084 0.8079 0.5873 0.3609
Paging 1981-1991 R
2
0.9929 0.9828 0.9888 0.9891
service (Japan) MSE 0.0053 0.7075 0.4599 0.5245
Cellular 1984-1992 R
2
0.9968 0.9528 0.9713 0.9835
service (Korea) MSE 0.0110 0.0844 0.0514 0.0355
Mobile 1979-1991 R
2
0.9735 0.9644 0.9631 0.9646
service (Japan) MSE 0.1201 0.2888 0.2992 0.3232
the case in the technological product market,
the differences in the explanatory power be-
tween the proposed model and existing mod-
els would be greater than they are shown in
Tables 3-6. Further, if empirical tests can be
performed by using more rened categories
102 Byungryong Kang et al. ETRI Journal, volume 18, number 2, July 1996
of products within each product type actually
grouped in the tests (e.g., the models 286, 386,
486, and 586 within personal computer prod-
uct), the proposed model would show more
signicantly higher explanatory power than
existing models.
Althoughnot reported here, the visual t of
our model is the best of all models empirically
tested in this study. Therefore, our model is
obviously better than other existing models in
that it has higher explanatory power and lower
forecasting errors. It can be seen from the em-
pirical results that the historical data conrms
the long-range predictions of our model. It
also demonstrate its ability to handle problems
other than technological substitution.
VI. CONCLUSIONS
This study develops a widely applicable
innovation diffusion model based on a con-
ceptual framework characterizing several fac-
tors determining market demand for techno-
logical products. The model is developed
partly to overcome certain limitations in ex-
isting growth models currently used for fore-
casting market demand for new products. Our
model is developed by usingassumptionscom-
monly employed in the area of market demand
and consumer behavior analyses in economic
theory. Discussedare major underlyingfactors
of market penetration and actual demand, such
as the level of income, price, stock and depre-
ciation, and tastes or consumer preferences.
In order to test the applicability of our
model for actual data, certain durable and ser-
vice products in electronics and telecommuni-
cations market are examined in both Korea and
Japan. The empirical test results indicate that
our model has higher explanatory power than
any other existing model in all cases tested.
Results further suggest that the model can be
widely applicable for trend analyses and fore-
casts for diffusion rates of technological prod-
ucts in markets, rate of change in stocks and
purchases, and market penetrationprocesses of
various new goods and services.
The results of this study should be consid-
ered within the content of its limitations. The
rst limitation is certain parameters including
purchases coefcient and depreciation coef-
cient are assumed to be constant over time.
This assumption is employed to enhance the
mathematical tractability of model derivation
process. The second limitation is that the dif-
fusion model in this study eventually degener-
ates into a function of time only with all eco-
nomic factors considered being parameters to
be estimated.
Technological forecasting on the basis of
trend extrapolationimplies the assumptionthat
things are going to change in the future more
or less the same way as they did in the past.
Even though there exist a very large number
of diversied factors, all interacting and jointly
contributing to the process of change, the as-
sumption is normally valid, since some fac-
tors accelerate the process while others retard
it, and the overall tendency normally devi-
ETRI Journal, volume 18, number 2, July 1996 Byungryong Kang et al. 103
ates little from an extrapolated central trend.
However, when there is a signicant change in
the circumstances inuencing the elements of
forecast, a practical forecasting model has to
be dynamic in nature and responsive to these
exogenous inuences.
APPENDIX
This appendixpresents the detailed charac-
teristics of representative existing models dis-
cussed in the body of the paper. The typical
equation for the Gompertz curve is
Y Dk a
b
x
. (29)
where k is an asymptote. The growth rate of
Gompertz curve can be represented by
LF
t
D(lnY)
tC1
(lnY)
t
Db
t
(b1).
and
LF
tC1
LF
t
D
(lnY)
tC2
(lnY)
tC1
(lnY)
tC1
(lnY)
t
Db.
Death of railway employees, accidents in fac-
tories, specic death rates, and other declining
series might be described by a Gompertz curve
having a lower asymptote at the right. When
0 - a - 1 and 0 - b - 1, the value of Y ap-
proaches k, the upper asymptote, and shows S-
shaped asymptotic growth curve. This curve
has a xed inection point, f D0.367, where
f is the fraction of the upper limit achieved,
and shows relatively fast ratio of increase at the
early stage.
The Pearl curve was developed to describe
the growths of an albino rat, of a tadpoles tail,
of the number of yeast cells in a nutritive solu-
tion, of the number of fruit ies in a bottle, and
the population growth in a geographical area.
The equation for the Pearl curve [3] is
Y D
K
1Ce
abx
. 0-b-1. (30)
Setting e
abx
Dj, the growth rate of j can be
expressed as
dj
dx
xDtC1
.
dj
dx
xDt
D
b e
ab(tC1)
b e
abt
De
b
.
When the value of coefcient b is positive and
less than 1, the value of jand Y approaches the
asymptote, 0 and k, respectively. This curve
has a xed inection point, f D 0.5, and the
rst difference of the trend values of the curve
forms a curve resembling a normal frequency
distribution (symmetry).
Sharif and Islam [13] suggested the
Weibull distribution as a general model for
forecasting technological changes. This
Weibull distribution model (or Sharif-Islam
model) is based on Weibull distribution
function, which can accommodate different
patterns of technological changes. This curve
provides nonsymmetry with the point of
inection that responds to the substitution
process. The idea to describe the new product
growth via the Weibull distribution has been
suggested and illustrated earlier by Pessemier
[23]. The Weibull probability density function
(p.d.f.) is (31), its probability distribution
function (cumulative p.d.f.) is (32), and the
Weibull distribution model function is (33).
f (t) D()(t)
(1)
e
(t)
. (33)
104 Byungryong Kang et al. ETRI Journal, volume 18, number 2, July 1996
where L is the upper limit, F(t) is the fraction
of the upper limit achieved at time t, >0 is a
scale parameter, and > 0 is a shape parame-
ter. Here, and together determine the steep-
ness of the curve, while alone determines the
shape of the curve. Depending on the value of
the Weibull curve becomes left skewed, sym-
metrical, or right skewed. Thus, the Weibull
distributionmodel can be applicable for a wide
variety of technological forecastings. But it
cannot select any reference year as t D 0 for
the purpose of regression analysis, because in
the Weibull distribution, F(t) D0 at t D0 and
F(t) approaches its limit as t approaches inn-
ity. Therefore, the base year can only be esti-
mated by trial and error with the correlation co-
efcient as the criteria [13].
The Bass model is derived from a hazard
function (the probability that an adoption will
occur at time t given that it has not yet oc-
curred) [8] and is formulated as a new prod-
uct growth model for consumer durables on
the basis of the assumption that the timing
of a consumers initial purchase is related to
the number of previous buyers [7]. The as-
sumptions of the Bass model are similar to
the theoretical concepts emerging in the lit-
eratures on new product adoption and diffu-
sion, as well as to some learning models, and
the theoretical framework stems mathemati-
cally from the contagion models which have
found such widespread application in epide-
mology [4], [7]. The equation [7], [8] for the
Bass model is
f (t) D[ pCdF(t)][1F(t)]. or
n(t) D
dN(t)
dt
D p(mN(t))
C
q
m
N(t)(mN(t)). (34)
where f (t) is the likelihood of purchase at
time t, F(t) D
R
t
0
f (t)dt, n(t) is the number of
adopters at time t(Dmf (t)), N(t) is the cumu-
lative number of adopters at time t(DmF(t )),
m is the potential number of ultimate adopters,
p is the coefcient of innovation, and q is the
coefcient of imitation. The rst term in (34),
p(m N(t)), represents adoption of buyers
who are not inuenced in the timing of their
adoption by the number of people who already
have bought the product, and the second term,
q
m
N(t)(mN(t)), represents adoption of buy-
ers who are inuenced by the number of pre-
vious buyers. The Bass model assumes that
the maximumpenetration rate cannot occur af-
ter the product has captured 50% of the mar-
ket potential. The dynamic model presented
in Norton and Bass extend the basic model of
Bass into the realm of succesive substitution
between generations in order to facilitate mul-
tiperiod ahead forecast up to the three years
[24].
Easingwood, Mahajan, and Mullar have
suggested the nonuniform inuence(NUI) dif-
fusion model that overcomes these limitations
of the Bass model which considers the co-
efcient of imitation as symmetrically vary-
ing over time. The equation [8], [11] of NUI
model is
dF(t)
dt
D[ pCqF
(t)][(1F(t)]. (35)
where F(t) D N(t)m, and is the nonuni-
form inuence factor. The NUI model pro-
vides an adoptionrate curve which can be sym-
ETRI Journal, volume 18, number 2, July 1996 Byungryong Kang et al. 105
metrical or nonsymmetrical, with the point of
inection occurring at any stage of the dif-
fusion process. When p D 0, i.e., the coef-
cient of external inuence is zero, equation
(35) yields a exible extension of the Mans-
eld model termed nonsymmetrical respond-
ing logistic (NSRL) model [10].
REFERENCES
[1] P. H. Franses, Gompertz Curves with seasonality,
Technological ForecastingandSocial Change, vol.
45, pp. 287-297, 1994.
[2] J. P. Martino, Technological Forecasting for Deci-
sion Making. NewYork: American Elsevier, 1975.
[3] R. Pearl, The Biology of PopulationGrowth. Alfred
A. Knopf, Ed., New York, 1925.
[4] E. Manseld, Technological Change and the Rate
of Imitation, Econometrica, vol. 29, pp. 741-765,
1961.
[5] A. W. Blackman, Jr., A mathematical model for
trend forecast, Technological Forecastingand So-
cial Change, vol. 3, pp. 441-445, 1972.
[6] J. C. Fisher and R. H. Pry, A simple substitu-
tion model of technological change, Technologi-
cal Forecasting and Social Change, vol. 3, pp. 75-
88, 1971.
[7] F. M. Bass, A newproduct growth model for con-
sumer durables, Management Science, vol. 15, pp.
215-227, 1969.
[8] V. Mahajan, E. Muller, and F. M. Bass, Newprod-
uct diffusion models in marketing: A review and
directions for research, Journal of Marketing, vol.
54, pp. 1-26, 1990.
[9] C. J. Easingwood, V. Mahajan, and E. Muller, A
nonsymmetric responding logistic model for tech-
nological substitution, Technological Forecasting
and Social Change, vol. 20, pp. 199-213, 1981.
[10] J. A. Sharp, An interpretation of non-symmetric
responding logistic model in term of price and ex-
perience effects, Journal of Forecasting, vol. 3,
pp. 453-436, 1984.
[11] C. J. Easingwood, V. Mahajan, and E. Muller, A
nonuniform inuence innovation diffusion model
of new product acceptance, Marketing Science,
vol. 2, pp. 273-295, 1983.
[12] M. N. Sharif and C. Kabir, A generalized model
for forecasting technological substitution, Tech-
nological Forecasting and Social Change, vol. 8,
pp. 353-364, 1976.
[13] M. N. Sharif and M. N. Islam, The weibull dis-
tribution as a general model for forecasting tech-
nological, Technological Forecasting and Social
Change, vol. 18, pp. 247-256, 1980.
[14] X. Tingyan, A combined growth model for trend
forecasts, Technical Forecasting and Social
Change, vol. 38, pp. 175-186, 1990.
[15] A. Floyd, A Methodology for Forecasting of Fig-
ures of Merit, in Technological Forecasting for In-
dustry and Government: Methods and Applica-
tions. James R. Bright, Ed., Englewood Cliffs, NJ:
Prentice-Hall, 1968.
[16] B. H. Bao and D. H. Bao, LIFO adoption: A tech-
nology diffusion analysis, Accounting Organiza-
tions and Society, vol. 14, pp. 303-319, 1989.
[17] R. Stone, Linear expenditure systems and demand
analysis: An application to the pattern of british de-
mand, The Economic Journal, LXIV, pp. 511-527,
1954.
[18] R. Stone and D. A. Rowe, The durability of con-
sumers durable goods, Econometrica, vol. 28, pp.
407-416, 1960.
[19] H. S. Houthakker and L. D. Taylor, Consumer
Demand in the United States. Harvard University
Press, 1966.
[20] Statistical Yearbook of Communications. Ministry
of Communications, Korea (1982-1994).
[21] Annual Report of Telecommunication, The Elec-
tronic Times Inc., Korea (1988-1994).
[22] A Report of Study Group on Broadband ISDN in
Japan. Ministry of Posts and Telecommunications,
106 Byungryong Kang et al. ETRI Journal, volume 18, number 2, July 1996
Japan, 1992.
[23] E. A. Pessemier, Product Management, Strategy
and Organization. New York: John Wiley and
Sons, 1977.
[24] J. A. Norton and F. M. Bass, A diffusion theory
of adoption and substitution for succesive genera-
tions of high technology products, Management
Science, vol. 33, pp.1069-1086, 1987.
[25] T. Hirokazu and J. Dipak, Cross-national analysis
of diffusion of consumer durable goods in Pacic
rim countries, Journal of Marketing, vol. 55, pp.
48-54, 1991.
Byungryong Kang received
the B.S. degree in economics
from Sungkyunkwan Univer-
sity in 1980 and the M.S. de-
gree in theoretical economics
from Kyungpook National
University in 1982. He joined
ETRI in March 1984, and since then he had been
involved in feasibility study, economic analysis, and
project planning for the various projects of large-scale
public switching systems development such as TDX-1,
TDX-10, ISDN, and ATM systems. From 1993 to
1994, he was a Visiting Research Scholar at Georgia
State University in the USA. After his return he has
been a manager for the project of demand forecast and
economic analysis for ATM switching systems. He is
currently a Principal Member of Research Staff in ETRI.
His research interests include econometric models,
innovation process, demand forecast and economic
analysis for new technological products and services in
telecommunications industry.
Chimoon Han received the
B.E. and M.E. degrees in
electronics engineering from
Kyungpook National Uni-
versity in 1977 and Yonsei
University in 1983, respec-
tively. He received the Ph.D.
degree in electronics engineering from the University of
Tokyo, Japan, in 1990. Since he joined Korea Institute
of Science and Technology in 1977, he was involved
in developing optical ber communication and radio
mobile communication systems until 1982. Since 1983,
He has been involved in developing ISDN user-network
interface system, LAN system, ATM switching system
and wireless ATM PCS system in ETRI. He is currently
the Director of Systems Technology Department in
Switching Technologies Division, in ETRI. His research
interests include the system architecture and its perfor-
mance evaluation, system engineering, system design,
and implementation for ATM switching systems and
wireless ATM PCS systems.
Chu-Hwan Yim received his
B.S. and M.S. degrees from
Seoul National University in
industrial education (Electro-
nics) in 1972 and 1979, re-
spectively. He received Dr.
degree in communication sys-
tems from the Technical University of Braunschweig,
Germany, in 1984. He joined ETRI in 1978. Between
1979 and 1984, he was with communications systems in-
stitute of Technical University of Braunschweig. He re-
turned to ETRI in 1984, where he served for two years
(1987-89) as Director of ISDN Department, next two
years (1989-1991) as Director of Switching Technology
Department, and next three years (1991-94) as Direc-
tor of Protocol Engineering Center. Since 1994 he has
been the Vice President of Switching Technologies Di-
vision, in ETRI. The main project being carried out is a
development ATM switching system. Dr. Yim has pub-
lished over 70 papers in the major international and do-
mestic journals and proceedings, and has given several
dozens of invited talks at domestic and international con-
ferences worldwide. He has served many times as chair-
manor as a member of international anddomestic confer-
ence/workshop/symposium committees, program com-
mittees and advisory committees.