Chasco Yrigoyen Choir, Department of Applied Economics, UAM.

EXAMPLE OF ECONOMETRIC MODELS

See Case 9 (pp. 505 and following) of the book by A. Pulido and A. López (1999), "Prediction and Simulation"
applied to the economy and business management." Ed. Pyramid. It presents "some
basic macroeconomic relationships for the Spanish economy: national private consumption,
private investment, exports and imports of goods and services, interest payments on the
AAPP, credits to the private sector and interest rates.

1. UNIVARIATE TIME SERIES MODELS

Example 1: Suppose the following model:

Vt= a+ bPRt+ cPBt

being V: sales of a company

PR: prices
PB: budget in advertising

Example 2: In order to predict the unemployment volume in Spain, the following could be elaborated.
model:
TPAt= a 0+ a1 P14t + a 2UCP+ Ut t

being TPA: unemployment rate, defined as the unemployed population divided by the active population.
P14: population aged 14 and over
PROD: productivity, defined as GDP divided by the employed population.

Another model with PPA (unemployment population) could also have been specified as a variable.
endogenous
PPA=
t
a 0+ a1 P14t + a 2UCP+ Utt

Example 3: Suppose that the consumption of a certain product depends on income.

familiar and whether or not they have property housing, the following model can be established:

C i= a0+ a1 Ri+ a 2F+ UI i

being: Cithe consumption of the product
Rithe household income
Fia fictitious variable that will take the value 1 if the family owns a home and
0 otherwise.
Chasco Yrigoyen Choir, Department of Applied Economics, UAM.

Example 4: The production function calculated by Cobb and Douglas, for the set of the
The economy of the United States in the period 1900-1922 was estimated as follows:

Pt = 1.10L0t,75C t0.25

being: P: total production index per year

L: work 'input' index
C: capital 'inputs' index

This model is already estimated (the value of the parameters is known), so it does not
includes the term of random disturbance: we will go deeper into this later.
issue.

When, as in this and many other cases, one starts from a potential specification, that is,
linear in the logarithms of the variables, then the parameters are directly the
elasticities: elasticity of P with respect to L and C. In this case, the parameters will indicate, in
sufficiently approximate form for small percentage increments of L or C, the
percentage increase corresponding to P. For example, in the USA, in the period 1900-1922,
a 1% percentage variation in the labor inputs (L) resulted in an increase in the
total production of 0.75%, while increases of 1% in capital inputs C
they produced an increase in production of only 0.25%. That is, the increases in inputs of
Work, in relation to capital, produced increases 3 times greater in total production.

Example 5: Let R be a model of income (R) as a function of educational level (E):

R t = β1+ β2 E t+ u t

This model forgets that most people have higher incomes when they are older than when they are younger.
young people, regardless of their education. That is why the parameterβ2will be overestimating the
impact of education on income. Therefore, a more appropriate specification should include
the age variable (A) in the model:

R t= β1+ β2 E t+ β3 A t+ u t

However, it is a well-established fact that income tends to grow less than proportionally in
the last years of working life than in the first ones. Therefore, the model should be specified:

R t= β1+ β2 E t + β3 A t+ β4 A 2t+ u t

In this case, it is expected thatβ3have a positive sign and thatβ4be negative.

Chasco Yrigoyen Choir, Department of Applied Economics, UAM.

2. MULTIEQUATIONAL TIME SERIES MODELS

Example 6: A trading company intends to know the evolution of its gross profit (GP),
taking into account the invested capital (SK), the number of workers (L), the selling price
(PRE) and an economic activity index (ACT). To estimate the behavior of these
Initially, an equation is proposed with the following specification:

BENt = β1+ β2SK t+ β3L t+ β 4 PREt+ β5 ACT+ u t

After an initial estimate, it is concluded that the model is not very good due to
that the benefit has 2 clearly differentiated elements whose behavior should be
analyzed separately: income (ING) and expenses (GAS), so the model looks like this:

INGt = β1+ β2SK t + β3L t+ β4 PREt+ β5 ACT+ u t

BENt= INGt− GASt

When making predictions, the company faces the difficulty of determining the
future spending level, as it will be directly conditioned by the evolution of
the income, as well as by the number of employees, confirming the need to endogenize
the expense variable to simultaneously determine the level of future income and expenses.
Likewise, the income variable not only depends on environmental variables, but also on itself.
volume of purchases made, so GAS will be included as an explanatory of ING:

INGt= β1+ β2GASt+ β3SK t + β4 L t + β5 PREt+ β6 ACT+ u1,tt

GASt= γ1+ γ 2 INGt + γ 3L t+ u 2,t
BENt= INGt− GASt

Finally, once the need to develop a multiequational model has been established, we
it is appropriate that, for proper management of the company, it is necessary to model.
In addition to the benefits, the evolution of the capital stock, fulfilling the dual objective of
analyze the benefits and their reinvestment for the increase of the capital stock:

INGt = β1+ β2GASt+ β3SK t+ β4 L t + β5 PREt + β6 ACT+ u1,tt

GASt= γ1+ γ 2 INGt+ γ 3L t+ u 2,t
SK t = λ1+ λ 2 BENt + u 3,t
BENt = INGt− GASt

As can be seen, we have moved from a uniecuational model (1st) to another model.
recursive multiequational (2nd), passing through a block-recursive multiequational model (3rd) and
ending with a multiequational model of simultaneous equations (4th).
Chasco Yrigoyen Choir, Department of Applied Economics, UAM.

Example 7: Let it be a simplified model for determining consumption C, investment I and the
rent and from a country:

⎧ C t = a+1 a Y+ U
2 t −1 t
⎪ (
I=
⎨ t b+ 1
b 2
Y t − Yt −1) + Vt
⎪Y= C + I + G
⎩ t t t t

The last equation is an accounting identity in a closed economy, which is why it is not included.
the random disturbance, given that the relationship is exact, with no possibility of error.

Example 8: Lawrence R. Klein Model:

⎧ ⎧C=i α + α0 1
(W1+ W2+)I α 2 π1+ u1i
⎪⎪
⎪ ⎨ I=i ρ +0ρ π 1+ iρ π +2 ρi−1K+ u3 i−1 2i
⎪
⎪⎪ W1I= δ 0+ δ1 (Y+ T− W2 ) + δ 2 (Y+ T− W2i−1) + δ 3i+ u 3i
⎩ i
⎨
⎪ ⎧ Y+i
T=i C+ II+ Gi i
⎪
⎪ Y= W+ W+ π
⎪⎨ I 1i 2i i
⎪
⎪⎩ ⎩ ∆K = I
i i

n = 1921-1941 (American economy)

Monetary variables: billions of dollars from 1934 (real terms)
C: private consumption
W1 : salaries paid by the private sector
W2 : salaries paid by the public sector
π: benefits
I: private investment
K: private equity stock
Y: national income
I: corporate tax
G: public spending (except for salaries)
i: tiempo (año de referencia); i = 1, 2, ..., n
u1, u2, u3random disturbances
α, ρ, δmodel parameters.

The first 3 equations are univariate econometric models.

The last 3 equations are deterministic models or accounting identities that must
disappear and be replaced in the other equations.
Chasco Yrigoyen Choir, Department of Applied Economics, UAM.

3. AUTOREGRESSIVE TIME SERIES MODELS

Example 9: Let there be a model of the monthly gasoline consumption series (z), which has been
previously transformed into logarithms and differences (first differences and differences of
order 12) with the aim of eliminating all existing trends from it (trends in
variance and trends in the average, in the regular and seasonal parts). The model finally
specified is the following:

z t= φ1z t −1 + Φ1z t −12+ a t + θa t −1

It is an ARIMA(1,1,1) x SARIMA(1,1,0) model in which the consumption variable of

gasoline is explained in terms of itself in the previous month, as well as twelve months
before, and of the error the model makes in the previous period.

Example 10: Let there be a model of the monthly variable of the CPI of food (p) where this
the variable has previously been transformed into logarithms, in second differences in the part
regular and in first differences in the seasonal part:

p t = Φ1p t −12+ a t + θa t −1

It is an ARIMA(0,2,1) x SARIMA(1,1,0) model where the IPC variable of

Nutrition is explained in terms of whether twelve months earlier and the error that the model makes in
the previous period. The results of the estimation highlight the existence of errors
atypical. Upon delving into the reasons why these values occur, it is discovered that the
climatology (when it has been particularly favorable or unfavorable) has had a special impact
about food prices. Therefore, 3 new dummy variables are included in the model.
that they collect these effects:

p t= α1Z1+
t α 2 Z2 t + α 3 Z3t+ Φ1p t −12+ a t + θa t −1

Z1 takes the value 1 if it is a month of exceptionally bad weather and 0 otherwise.

Z2 takes the value 1 if it is a month of bad weather and 0 otherwise. Z3 takes the
value 1 if it is good weather and 0 otherwise. In this way, it is possible to quantify
the impact of weather conditions (very bad, bad or good) on food prices.

Despite everything, the prediction during the historical period of the CPI still leaves much to be desired, for
what is tested by introducing the incidence that non-food prices (q)
they have about food. The non-food CPI variable is introduced into the model.
transformed (in the same way as p) to avoid the existence of trends in it:

p t= λ1q t+ α1Z1+ αt 2 Z2t + α 3 Z3t+ Φ1p t −12+ a t+ θa t −1

Chasco Yrigoyen Choir, Department of Applied Economics, UAM.

4. CROSS-SECTIONAL DATA MODELS

Example 11: Let it be a model of wage earners from the EU Household Panel 94:

wI= β 0+ β 1 S i+ u I

Where the hourly wage in pesetas and the educational level in years. The estimated coefficient ofβ1
Indicate how many pesetas the salary increases due to increases in the educational level in one year.

If the hourly wage variable is transformed into logarithms:

lnwi= β 0+ β 1 S i+ u i

Is it possible to know by what percentage (in so many per 1) the hourly wage increases with variations?
of educational level in 1 year.

Since work experience (EX) is also an important variable, the previous model
should also consider this variable. Although, in this case, the impact that higher values of
years of experience on the hourly wage is supposed to be produced in a way less than
proportional

lnwi= β 0+ β 1 S i+ β 3EXi+ β 4EX2+ u i

In this case, a negative sign is expected inβ4.