Check your understanding: Input-output models

February 1, 2018

Input-output models

Input-Output and Social Accounting Matrices

Input-output was partly inspired by the Marxian and Walrasian anal-

ysis of general equilibrium via interindustry flows. It has been a
mainstay of economics and economic policy and planning through-
out the world for the past half-century and used for economic plan-
ning throughout the developed and developing world.
Leontief was raised in Russia and obtained his Ph.D. in Berlin. He
developed input-output analysis at Harvard after 1932 and he bagan
constructing an empirical example of his input-output system, his
1941 classic, Structure of American Industry. Leontief followed up this
work with a series of classical papers on input-output economics.

Simple Example
Let there be two sectors agriculture and industry. Agriculture re-
quires 2 units of industry to produce 10 units of output while indus-
try requires 3 units of agriculture to produce 20 units of output. Each
uses its output as an input, perhaps 10 percent of its gross output.
The A matrix or input-output matrix in coefficient form
" #
1/10 3/20
2/10 1/10

or " #
0.1 0.15
0.2 0.1
here the coefficient aij is the amount of good i used for the produc-
tion of one unit of good j. The
" coefficient
# matrix must be distinguished
1 3
from the flow matrix à =
2 2
" #
10
Here X is column vector of gross value of production X =
20
Product of A and X is AX is a quantity of intermediate use of both
agriculture and industry as illustrated below.
 check your understanding: input-output models 2

Final demand
The column vector F is final demand, consumption, investment,
" #
8.5
government and final demand, F = C + I + G + Nx F =
16
Note that intermediate demand is not part of GDP and is also
not part of final demand. Input-output analysis solves fundamental
problem of what level of gross outputs X is required if a specific final
demand vector F is to be produced?

The material balance equation

The fundamental problem of input-output analysis is thus to solve
the material balance equation

X = AX + F (1)

which just says that gross output is the intermediates plus final de-
mand. The question is how to solve this equation for X. For this
we need linear algebra since the material balance is only system of
simultaneous linear equations.1 1
see CYU on linear algebra for full
At each stage of production some intermediate goods, raw materi- explanation.

als and other inputs, are required. These are similar to capital goods,
but in fact only last a fraction of the period of production while cap-
ital goods last longer than one period. Let X j be the output of one of
n sectors in a given period. A simple average cost function would be
n m
pj Xj = ∑ pi aij Xj + ∑ wlj llj + r j k j (2)
i l =1

where pi is the price of good i, with i, j = 1, 2, ..., n, w j is the wage

paid, llj is the amount of labor of skill class l = 1, 2, ..., m and r j is
the cost of capital. The amount of capital required per unit of output
is k j . The key parameter for input-output (I-O) analysis is aij , the
amount of good i required as an intermediate or raw material for the
production of one unit of good j.
Capital and labor coefficients are usually considered to be variable
over time and across sectors. The coefficients depend on the wage-
rental ratio, w/r, for each skill category of labor. Given that ratio,
firms choose a combination of capital and labor that minimizes the
cost to produce the level of output X j . The key assumption for I-O
analysis is that, while factor proportions vary with the wage-rental
ratio, the aij proportions of the intermediate and raw material inputs
do not.
By far the most attractive feature of I-O models is their ability to
account for economy-wide effects of project implementation. An
 check your understanding: input-output models 3

infrastructure project is said to “pull” on inputs, both domestic and

foreign, which in turn pull on their own suppliers. To see this let the
n × n matrix
 
a11 a12 ... a1n
 a21 a22 ... a2n 
 
A= . (3)
 
 .. .. .. 
 . . 
an1 an2 ... ann
be the I-O coefficient matrix describing the amount of good i required
for one unit of the production of good j, as noted above. These are
the flow coefficients and have embedded in them the fixed-coefficient
assumption already discussed. The total demand for intermediate
goods is an n × 1 matrix, or column vector.
  
a11 a12 ... a1n X1
 a21 a22 ... a2n   X2 
  
AX = 
 .. .. ..  . 
  (4)
 . . .   ..  Figure 1: Vector diagram of the I-O
model-âĂŞinputs must be ≤ outputs
an1 an2 ... ann Xn for a closed economy

which is the same as ∑nj=1 aij X j . Figure 1 shows the output and input Industry

vectors for a two-sector economy, agriculture and industry. This

X
economy is productive since its inputs are bounded from above by
the dotted lines. If the inputs were greater than either or both upper Inputs
AX
bounds, some level of imports would be indicated.
Figure 2 shows the essence of the I-O model. As inputs are re- 0utput
quired for the outputs, so too are inputs required for the inputs
themselves. This is shown in the figure as the even shorter vector
labeled A2 . The process continues ad infinitum but practically only
to the point that the level of inputs for that round is indistinguishable
from zero. The sum of these vanishing vectors is known as the direct Agriculture

and indirect requirements for the production of X. This beloved concept

among practitioners of I-O analysis shows that it is not possible in
modern economies to produce anything without effectively rippling
through the entire economy, for both output and employment.
How then is X itself determined? It can be deduced from what is Figure 2: Vector diagram of the I-
known as the material balance equation of the I-O framework. Implic- O model–note the shrinking size of
itly define the vector of final demand, F as the input vectors as they chain and
sum to the direct and indirect input
       requirements for the given value of X
X1 a11 a12 ... a1n X1 F1
Industry
 2   a21
X a22 ... a2n   X2   F2 
      
 . = . .. ..  . + .  (5)
   
 .   .
 .   . . .   ..   ..  X

Xn an1 an2 ... ann Xn Fn

Inputs
AX

0utput

A3X
A(AX)=A2X

Inputs7to7
produce7inputs

Agriculture
 check your understanding: input-output models 4

This expression is an expanded version of the material balance

equation above, repeated here for convenience as
X = AX + F
where F is shown in figure 3 as the difference between X and AX.
There are many ways to solve this last expression for X as a function
of final demand, F. In the early days of the Soviet Union they did it
with buildings full of computers, literally women armed with pen-
cils and paper, but now it can be done by matrix inversion methods
available even in Excel. A more instructive solution is the so-called
power-series expansion or approximation to the solution that in-
volves successive powers of A multiplied by the final demand vector
F
X ∼ F + AF + A2 F + ... + An F (6)
which provides the basis for quick computation of X given F in virtu-
ally any programming language.2 In words, this expression says: to 2
The classic reference, ?, is still the
produce F one needs also to produce inputs for F and then inputs to best since it shows the close connection
between economic theory and the I-O
produce the inputs for F and so on. The gross output, X, is then the framework.
sum of these inputs plus the quantity of F itself.
To see how an infrastructure project that might treat one or more
rivers that deliver phosphorous to Lake Champlain consider the
following experiment. Raise the second coefficient in the vector F by
an amount ∆F2 , the planned expenditure during the period3 3
A numerical example of precisely this
infrastructure shock is given below.
" #
F1
F2 + ∆F2
This says the project direct inputs will be drawn from the industrial
sector. Indirect inputs will, of course, be drawn from all sectors.
Recall that F is a compact way of writing the more familiar national
income and product accounting equation
F = C + I + G + Nx
where C is consumption (including imports), I is the sum of struc-
tures, equipment, residential construction and change in inventories
undertaken by both private and public sectors, G is current govern-
ment consumption and Nx is net exports, exports minus imports.
When the infrastructure is built, it will enter final demand as invest-
ment (assuming it takes less than or equal to one year).4 It will then 4
If not, the stimulus is spread over
begin to pull on its intermediate inputs, which in turn pulls on theirs the entire project period as seen in the
example below.
in an infinite but asymptotic chain.

Employment in I-O models

With data in hand on the wage bill for the infrastructure project (sub-
ject to the caveats mentioned above about its composition) one can
 check your understanding: input-output models 5

perform an employment impact assessment of the project.5 Without 5

The concepts explained in this section
the aid of the conceptual framework developed so far, one might be will apply to Social Accounting Ma-
trices, SAMs, both static and dynamic,
tempted to write the employment total, L, as that are discussed in the next sections.

n
L = LF = ∑ li Fi
i =1

where L = [l1 , l2 , ..., ln ] is a row-vector of labor coefficients or ratios of

the employment to the level of gross outputs of each sector, derived
from the base SAM in which wages and prices are conventionally set
to one. This would, however, be incorrect since it omits the employ-
ment generated by the production of the inputs for F. It also omits
the inputs to produce the inputs and so on, as just discussed. The
correct expression for total employment is
n n
L = LX = ∑ li Xi > ∑ li Fi .
i =1 i =1

The employment impact in the case of the infrastructure, ∆Ld , can be

written as
" # " # " #2 " #n  " #
h i 1 0 a 11 a 12 a 11 a 12 a 11 a 12 0
∆ L d = l1 l2  + + +···+ 
0 1 a21 a22 a21 a22 a21 a22 ∆F2

The change in direct employment due to the infrastructure project is

∆Ld0 , is
" #
h i 0
∆ L d 0 = l1 l2
∆F2
whereas the change in direct plus indirect employment is ∆Ld .

Closed and open I-O models

Closed and open I-O models are usefully distinguished (?), primarily
because the former reveal the inner workings of the multiplier. The
model above is open but a closed model would include coefficients
for consumption as well as intermediates. These are known as labor-
feeding coefficients, cij ,
 
c11 c12 ... c1n
 c21 c22 ... c2n 
 
c= .
 
 .. .. .. 
 . . 
cn1 cn2 ... cnn

which measure the amount of good i labor must consume to produce

one unit of labor. In some sense this is an odd concept, but again
 check your understanding: input-output models 6

the coarse-grain nature of I-O modeling allows for a large amount

of detail to be built in, ex post. Total consumption demand for this
economy is
C = cLX
Here equation 5 above, the material balance equation, can be altered
to include the labor-feeding coefficients by replacing A with
   
a11 a12 ... a1n c11 l1 c12 l2 ... c1n ln
 a21 a22 ... a2n   c21 l1 c22 l2 ... c2n ln 
   
A+ = 
 .. .. .. + .
 .. .. 
 . . .   .. . .


an1 an2 ... ann cn1 l1 cn2 l2 ... cnn ln
which can be compactly written as
Figure 3: Vector diagram of final
demand in an I-O model
X = A+ X + F
Industry

where A+ = A + CL. The system can then be solved for a given level
F3=3Final3demand X
of final demand, F.
Inputs
X ∼ F + A+ F + ( A+ )2 F + ... + ( A+ )n F AX

The rest of the employment analysis follows in the wake of this 0utput

rather fundamental change in the nature of the model. The labor-

feeding coefficients, cil , are not easy to estimate and embody an unre-
alistic assumption, namely, that workers do not adjust their consump-
tion bundle in response to changes in the macro economy, whether
incomes or prices. So far, of course, there are no relative prices in the Agriculture

I-O model and so one could presumably sleep soundly, even after
making this assumption. More realistically, side calculations could be
made by hand on the time-path of the cij coefficients as relative prices
do in fact evolve.
What workers require to live and keep working admittedly sug-
gests an excessively rigid vision of the economy, but there is one
attractive feature of the closed model that could in principle be used
to establish an upper bound for the impact of an investment project
shock. Figure 4 shows that F, A and A+ X are all collinear and this
corresponds to the maximal rate of growth of the closed system and
therefore the maximal rate of growth of employment.6 Employment 6
The eigenvector associated with
in the economy is L = LX. the maximal eigenvalue gives the
proportions for balanced growth along
this von Neumann turnpike. The
surplus over and above the technical
and labor-feeding coefficients, S =
( I − A+ ) X is itself a vector of goods
that may be consumed, invested or
exported. Only if all the surplus is
reinvested will a closed economy grow
at its fastest sustainable rate.
 check your understanding: input-output models 7

Figure 4: Vector diagram of balanced

growth with maximal employment
Prices and dual of the material balance growth

Since the I-O model is a matrix, it can be viewed either as a “stack"

of rows or a “rack" of columns. Viewing the problem from a row
perspective gives a material balance, a balance between supply and
demand. The price level for each row is given by

PX = PAX + PC + PI + PE − eP∗ M (7)

where P = [ p1 , p2 , ..., pn ], A is the I-O coefficient matrix, C =

0 0 /Users/wgibson/ILO/graphics/IO_4.pdf
[C1 , C2 , ..., Cn ] is the consumption column vector, I = [ I1 , I2 , ..., In ]
0
is the level of investment7 , E = [ E1 , E2 , ..., En ] is exports and M =
0
[ M1 , M2 , ..., Mn ] is the level of imports. The nominal exchange rate
is given by e and the price of imports is P∗ . Here PAX is interme-
diate demand and PC + PI + PE − eP∗ M is final demand. Factor
demand for labor is denoted L = [ L1 , L2 , ..., Ln ] and capital is given by
K = [K1 , K2 , ..., Kn ]. Value added, V, is then
7
The apostrophe indicates vector
V = PX − PAX = wL + rK transpose.

where w is the wage rate and r is the rate of return to capital. Value
added must equal the value of final demand since

PF = PX − PAX

or V = PF. It should be clear from this discussion that I-O is less of

an economic model than it is a framework to record the data of an
economy; in other words, a highly structured data-base.
Columns of the matrix require prices since, reading down the
columns of the I-O matrix, the goods are heterogeneous and must be
aggregated by the price vector. To this nominal value is added the
return to labor, wL, and the return to capital, rK, which are both mea-
sured in nominal terms. This suggests that the entire presentation of
the I-O framework must be in nominal terms. If a given I-O system is
also compiled for a base year, the values in the structured data base
are both real and nominal. As noted above, it is convenient to normal-
ize all prices to one for the base year, along with the base year wage
rate. The units of X are then in millions of LCUs and if prices were to
rise by, for example, 15 per cent, it could be said that PX is the cost of
what could have been purchased in the base year with one-million LCUs
of the base year. This is a useful convention in I-O accounting and
widely adopted.
In table 1 there are two sectors, industry and agriculture. Here
the GDP, computed as the sum of value added in both sectors, is
48+70 = 118. Assume, unrealistically, that the infrastructure can be
 check your understanding: input-output models 8

Table 1: I-O framework

Agriculture Industry Consumption Investment Government Exports Total

Agriculture 10 8 30 5 15 10 78
Industry 20 12 40 8 15 -5 90
Value Added 48 70 - - - - -
labor 30 42 - - - - -
Capital 18 28 - - - - -

Total 78 90 70 13 30 5 -

- = N/A
Millions of LCUs.
Source: Authors’ computations based on illustrative data.
Table 2: An infrastructure project
Agriculture Industry Cons Investment Govt Exports Total

Agriculture 102 8.12 30 5 15 10 78.1

Industry 204 12.19 40 9.18 15 -5 91.4
Value added 48.09 71.09
Labor 306 42.66
Capital 18.03 28.44

Total 78.1 91.4 70 14.2 30 5 -

Millions of LCUs.
Source: Authors’ computations based on illustrative data.

constructed in one year and that its construction requires an expendi-

ture of one per cent of GDP, the minimum the framework is capable
of seeing. This is 1.18 million LCUs and is added to the level of in-
vestment demand for industry.8 In the I-O framework this rise in 8
Keep in mind that the investment
aggregate demand gives direct and indirect effects on the both sectors above is investment by origin and not
destination. The former is a component
as seen in table 2. of aggregate demand whereas the latter
From table 2 it is evident that investment has risen from 8 to 9.18, is an increment in the capital stock.
or one per cent of GDP. This change causes direct and indirect effects
on both industry and agriculture precisely in the fashion described in
the mathematical model above. This is the simplest possible model
inasmuch as consumption remains fixed and only intermediate de-
mand rises in response to the infrastructure shock. In particular, the
capital and labor components of total value added rise proportion-
 check your understanding: input-output models 9

ately. The demand for labor increases according to labor coefficients

calculated from the base (30/78 for agriculture and 42/90 for indus-
try). The same occurs with capital in this entirely linear model.
The example identifies the limitations of the linear I-O model as a
framework for analyzing the impact of infrastructures on the demand
for labor. It is unrealistic to think that the cost of capital will rise
proportionately with output in the short run. It might be better to
have the capital fixed in the short run so that the cost of capital itself
remains fixed. In the short run, output increases in both sectors, but
employing more workers with the same amount of capital is satisfied
by the rise in demand.
How does the demand for labor actually change when an infras-
tructure is introduced? In the short run, capital is fixed, as is the
nominal wage rate. The assumption is that the project managers will
employ labor until the value of its marginal product is just equal to
the real wage. Any other assumption would violate the most basic
principles of profit maximization, implicitly allowing firms to employ
more or less labor than they need.
The literature on project analysis suggests that “satellite" models
can be constructed to evaluate the impact of project investment on
the economy, they mean that one of the most crucial assumptions
for the construction of a satellite model is that there should be a
“representative firm" for the satellite that shows how it reacts to the
infrastructure shock. This will allow for a more realistic non-linear
response to the increase in demand. The satellite model must be
consistent with the value added in the I-O model.
Assume first that the industrial sector production function is given
by
n
β j 1− β j
p j x j − ∑ pi aij x j = A j K j L j (8)
i =1
where A j is an arbitrary calibration constant that can be used to
model technological change or spending on environmental protec-
tion. Here K j is the capital employed by the industrial firm and L j is
the labor. Nominal value added, v j is given by
2
v j = p j x j − ∑ pi aij x j
i =1

Since this is a Cobb-Douglas production function, the marginal prod-

ucts for capital and labor, respectively are

β j v j /K j = r
(1 − β j )v j /L j = w (9)

where for the moment, w and r are common to all sectors. Under the
standard neo-classical assumption that a rational firm would employ
 check your understanding: input-output models 10

labor until the value of the marginal product of labor is equal to the
wage and the value of the marginal product of capital is equal to the
cost of capital, these two equations can be rearranged to yield

β j = rK j /v j
(1 − β j ) = wL j /v j

The right-hand side in both cases is the share of the factor in the
value of total production, precisely what is shown in the last two
rows of the I-O matrix in table 1. In this case

β 1 = 18/(18 + 30) = 0.375

β 2 = 28/(28 + 42) = 0.4

Once the share of capital, β j is estimated, the share of labor can be

deduced from the assumption of constant returns to scale (that the
exponents in the production function add to one). The labor shares
are

1 − β 1 = 0.625
1 − β 2 = 0.6

where it is seen that the factor shares are similar between the two
branches of production.
Why is the Cobb-Douglas production function used here? Observe
that the “adding-up” feature of the Cobb-Douglas is fully consistent
with the I-O framework and can be easily calibrated from data of
the matrix. Here the structure of the data determines the production
function. There is no real reason why the exponents would necessar-
ily add to one, but if the Cobb-Douglas is to be consistent with the
I-O structure, it must.
This is one of many examples in which the structure rather than
the data determines, or limits, what can be said using a model. One
of the reasons that I-O is so popular in the developing world is that it
imposes strict constraints on what the data can say. This can be seen
as both a benefit and a cost of the method.
If the percentage is given in the short-run, the year for which
the data is collected, but labor is variable and determined by the
marginal productivity condition, the firm is on its rising marginal
cost curve. Any increase in output then must be accompanied by a
rise in demand for labor. As output xi rises so too does vi . Equation
9 shows that the marginal product of labor and capital rise propor-
tionately. In the case of labor, employment rises and the wage rate is
 check your understanding: input-output models 11

fixed. For capital, however, the quantity is fixed and thus the rate of
return rises above the market rate. The assumption of the short run
drives this result but as capital invested changes in the next period,
the model is in a new “short-run” for which the same logic applies.
Equation shows how the new values of X can be computed. The
base level is given by
" # "" # " # " # " # " ## " #
78 1 0 0.128 089 039 023 01 01 003 002 60
∼ + + + +
90 0 1 0.256 0.133 067 041 02 01 005 003 58

where this calculation, easily done in Excel, is taken only to the 4th
power for ease of exposition.9 The rule of thumb for a minimum 9
The error is less than 0.2 per cent. Any
shock is approximately one per cent of GDP, or 1.18, that would degree of accuracy could be obtained by
extending the power series further.
be added to the second component in the F vector to represent the
increase in final demand for the infrastructure project.10 This gives 10
The model does not “see” small
shocks but is just as blind to large
shocks that induce sufficient structural
" #
78.1 change to do violence to the assumed
X=
91.4 stability of the parameters of produc-
tion and consumption.
assuming no change in the price level. The new value added is calcu-
lated as the difference between the value of output and intermediate
costs at the new level of X, as seen in table 2

v1 = 48.1
v2 = 71.1

then all the increase in value added must appear as added demand
in the labor market and a rise in the rate of return to capital. The
increase in employment is calculated as ∆L = li ( xi − xi0 ) where xi0 is
the base level of output.

∆L1 = 06
∆L2 = 0.66

Note that the rate of return to capital also increases slightly, assuming
that the capital stock remains fixed.11 11
The calculation is tedious but
straightforward. First compute K from
the Cobb-Douglas equation 8 taking
Exercises the wage rate as 1 and L from the base
matrix. The shares β and 1 − β are also
taken from the base matrix. The result
1. Consider the data is K1 = 105.1 and K2 = 150.6. The rate
  of return to capital–or the cost of a unit
0.2 0.3 0.2 of capital–can be then calculated for
A =  0.4

0.1 0.3 
 each sector from the fact that rK can be
read from the base matrix. This gives
0.3 0.5 0.2 r1 = 0.17 and r2 = 0.186. Holding K
constant during the shock produces a
with change in r of 0.172-0.171 = 0.001 and
0.189-0.186 = 0.003 for the two sectors
respectively.
 check your understanding: input-output models 12

 
150
F = 200
 
210

Determine the total demand for industries 1,2 and 3 given the
matrix of technical coefficients and final demand vector F.

Solution: First form

     
1 0 0 0.2 0.3 0.2 0.8 −0.3 −0.2
( I − A) =  0 1 0  −  0.4 0.1 0.3  =  −0.4 0.9 −0.3 
     
0 0 1 0.3 0.5 0.2 −0.3 −0.5 0.8

Next form the adjoint matrix

 
0.57 0.34 0.27
 0.41 0.58 0.32 
 
0.47 0.49 0.6

Finally, divide by the determinant 0.239 to get the Leontief inverse

   
0.57 0.34 0.27 2. 384 9 1. 422 6 1. 129 7
1 
( I − A ) −1 = 0.41 0.58 0.32 = 1. 715 5 2. 426 8 1. 338 9
  
0.239
0.47 0.49 0.6 1. 966 5 2. 050 2 2. 510 5

The last step is to premultiply the F vector by the Leontief inverse

    
2. 384 9 1. 422 6 1. 129 7 150 879. 5
1. 715 5 2. 426 8 1. 338 9 200 = 1023. 9
    
1. 966 5 2. 050 2 2. 510 5 210 1232. 2

2. A massive new public works program is initiated. Compute the

change in total production if final demand increases by 40,20 and
25 respectively.
Solution: We have ∆X = ( I − A)−1 ∆F.
    
2. 384 9 1. 422 6 1. 129 7 40 152. 09
1. 715 5 2. 426 8 1. 338 9 20 = 150. 63
    
1. 966 5 2. 050 2 2. 510 5 25 182. 43

3. Let the direct labor employed per unit of output be given by

h i
L = 0.2 0.15 0.3
 check your understanding: input-output models 13

Compute the employment impact of the new project.

Solution: The change in employment equal to the labor required

by the change in final demand; that is: L∆X = L( I − A)−1 ∆F.
  
h i 2. 384 9 1. 422 6 1. 129 7 40
0.2 0.15 0.3 1. 715 5 2. 426 8 1. 338 9 20 = 107. 74.
  
1. 966 5 2. 050 2 2. 510 5 25

This amounts to an increase of

 
h i 879. 5
0.2 0.15 0.3 1023. 9
 
1232. 2

= 107.74 or 107.74/699. 15 = 0.154 1 or a 15% increase in employ-

ment.
The denominator, 699.15 is given by L(1 − A)−1 F = LX = or
 
h i 879. 5
0.2 0.15 0.3 1023. 9
 
1232. 2

= 699. 15