GROWTH CURVES AND THEIR FITTING

Various growth curves , viz., the modified exponential, Gompertz and Logistic curves cannot
be determined by the principle of least squares since in all these cases, the number of
parameters (i.e., constants to be determined) is more than the number of variables. Special
techniques have been devised for fitting theses curves to the given set of data.
I. Modified Exponential Curve and its fitting:
Modified exponential curve is given by:

Yt = a + bc t ,

where Yt represents the time series value at time t, and a, b, c are constants called its
parameters, b and 0<c<1. Y0 = a + b and as t →  , Yt → a.

We shall discuss two methods for fitting modified exponential curve:

A. METHOD OF PARTIAL SUMS

To determine the constants of the curve, the whole range of t covered by the data is divided
into three equal parts , each including ,say, m points of time. Equating the totals:
m 2m 3m
S1 =  Yt , S2 =  Yt , S3 =  Yt
t =1 t = m +1 t =2m +1

 ( a + bc t )
m
 S1 =
t =1

Page 1 of 7
 m
= ma + b  c t
t =1
= ma + bP,
where
m
P=  ct
t =1

= c+c 2 + c3 + ... + c m
 
= c 1 + c + c 2 + ... + c m −1
= cP1 ,

P1 = 1 + c + c 2 + ... + c m−1 
cP1 = c+c 2 + c3 + ... + c m
 (1 − c)P1 = 1-c m
1-c m
 P1 =
(1 − c)
 1-c m 
 S1 = am +bc   .
 (1 − c) 

Page 2 of 7
 2m
S2 =  Yt
t = m +1

 ( a + bct )
2m
=
t = m +1
2m
= am + b  ct
t = m +1
= am + bQ,
where
2m
Q=  ct
t = m +1
m +1
=c + c m+ 2 + ... + c 2m

= c m+1 1 + c + ... + c m+1 
m +1 1 − c  m
=c  
 1− c 
Now
m +1 1 − c 
m
 S2 = am + bc  .
 1− c 
Further,
3m
S3 =  Yt
t=2m+1

 ( a + bct )
3m
=
t=2m+1
3m
= am + b  ct
t=2m+1
= am + bR,
where
3m
R=  c t = c 2m+1 + c 2m + 2 + ... + c3m
t=2m+1

=c 2m +1
1 + c + ... + c  m −1

2m +1 1 − c  m
=c  
 1− c 
2m +1 1 − c m
 S3 = am + bc  .
 1− c 
Thus,
Page 3 of 7
  1-cm 
S1 = am +bc   , (1)
 (1 − c) 

1 − cm 
S2 = am + bcm+1  , (2)
 1− c 
and

2m +1 1 − c  m
S3 = am + bc  . (3)
 1 − c 

m+1 1 − c  1 − c m 
m
 S2 − S1 = bc   - bc  
 1− c   1− c 
1 − c m  m
= bc   (c − 1)
 1 − c 
−bc
( )
2
= 1 − cm
1− c
or
bc
( )
2
S1 − S2 = 1 − cm . (4)
1− c
Similarly

1 − c m  2m+1 1 − c 
m
S2 − S3 = bc m+1   - bc  
 1− c   1− c 
m +1 1 − c 
 
m
= bc   1− c
m

 1− c 
bcm+1
 
2m
 S2 − S3 = 1 − cm . (5)
1− c
(5) (S2 − S3 )
 = cm
(4) (S1 − S2 )
or
1/m
 (S − S ) 
ĉ =  2 3  . (6)
 (S1 − S2 ) 

Substituting for ĉ in (4) we have

Page 4 of 7
 (S1 − S2 )(1 − c)
ˆ
b̂ = . (7)
ˆ − cˆ m ) 2
c(1

Substituting the values of b̂ and ĉ in (1) , we have

1 ˆ  1 − cˆ m  
aˆ = ˆ
S1 - b c   .
m   1 − cˆ  

B. METHOD OF THREE SELECTED POINTS

Here , we take three ordinates Y1 , Y2 , and Y3 (say) corresponding to three equidistant
values of t (say) t1 , t 2 , and t 3 , respectively, such that t 2 − t1 = t 3 − t 2 .

Substituting the values of t= t1 , t 2 , t 3 in Yt = a+bc t , a > 0, b < 0, 0 < c < 1, we have

t
Y1 = a +bc 1 ,
t
Y2 = a +bc 2 ,
and
t
Y3 = a +bc 3 .
t t
 Y2 − Y1 = b(c 2 −c 1)
t t −t
= bc 1 (c 2 1 − 1)
or
t t −t
Y1 − Y2 = bc 1 (1 − c 2 1) . (1)

Similarly
t
Y3 − Y2 = b(c 3 − c t 2 )

= bc
t
2
( t
c 3 - t2 − 1 )
(c − 1)
t t
= bc 2 2 - t1 (2)

(2) (Y2 - Y3 )
 = c t 2 − t1
(1) (Y1 - Y2 )
or
1/(t 2 − t1 )
 Y - Y3 
ĉ =  2  .
 Y1 - Y2 

Page 5 of 7
Substituting for ĉ in (1) we have

Y2 - Y1
b̂ =
 
.
cˆ t1
cˆ (t 2 −t1) − 1

Substituting b̂ and ĉ in Y1 = a+bc t1 , we have

aˆ = Y1 -bˆ cˆ t1 .

FITTING OF GOMPERTZ CURVE

Gompertz curve is given by
t
Yt = a bc
 log Yt = log a + c t ( log b )
or
U t = A + Bc t ; A > 0, B < 0, 0 < c <1.
A > 0  log a > 0 or a > 0.
B < 0  log b < 0 or 0 < b < 1
Use Method of Partial sums or Method of Three Selected points to obtain A, B, and c, from
which retrieve a and b.

FITTING OF LOGISTIC CURVE

k
Yt =
1 + ea + bt
1 1 + ea + bt
 =
Yt k
1 1 ea + bt
 = +
Yt k k
 U t = A + Bc t ,
where
1 1 ea
Ut = , A= , B= , and c = e b .
Yt k k
ea
A  0  k > 0, B < 0   0 which is not possible so B > 0,
k
0 < c < 1  0 < e b < 1 or b < 0.
Page 6 of 7
Use Method of Partial sums or Method of Three Selected points to obtain A, B, and c, from
which retrieve a and b.

Alternative Forms of Growth Curves:

Modified Exponential Curve

Yt = a-be−ct ; (a,b,c)>0

Gompertz Curve
−ct
Yt = ae − be ; (a,b,c)>0

Logistic Curve

a
Yt = ; ( a, b,c )  0
1 + be− ct

References

Granger, C. W. J. and P. Newbold (1986) Forecasting economic time series, Academic

Press, New York, 2nd ed.
Gupta, S.C. and Kapoor, V.K.(1994). Fundamentals of Applied Statistics. Sultan Chand & Sons.

Page 7 of 7