MODULE DEVELOPMENT – 2020/2021

MODULE: ECONOMETRICS
MODULE CODE: ECS3706

PART I: INTRODUCTION TO REGRESSION ANALYSIS

LESSON 1: AN OVERVIEW OF REGRESSION ANALYSIS

1.1 What is Econometrics?

To explain what econometrics is, we can look at its goal, its method and its uses.
✓ The goal of econometrics is the estimation of economic relationships.
✓ Its method is mainly regression analysis, using actual data.
✓ Econometric models are mainly used for describing economic reality, hypothesis testing,
simulation and forecasting.
Econometrics makes use of the following inputs/disciplines:

• economic theory
• economic data
• statistics

Economic theory

In previous economics modules, you encountered the following two economic relationships:
1. The demand curve in microeconomics: 𝑃 = 𝑎 + 𝑏𝑄𝑑 where (P: price, Qd: quantity
demanded).
 2. The consumption function in macroeconomics: 𝐶 = 𝐶̅ + 𝑐𝑌 where (C: private
consumption, Y: income, c: marginal propensity to consume).
In the previous modules, both were treated in a theoretical fashion. The coefficients of these
equations (for example a and b in the demand curve) were simply assumed to have some
predetermined values. To be of real use, however, you require accurate estimates thereof.
Econometrics provides methods to estimate these coefficients, using actual data.

The theory of economics is important in econometrics as it provides knowledge about the

nature of economic relationships. It helps us to choose the variables as well as the functional
form to be used, for example y = a+bX or log(Y) = a + bX. You will learn more about these
functional forms in lesson 8. In this module we assume that you are familiar with intermediate
microeconomic and macroeconomic models and their specifications.

The process of converting economic theory into a mathematical form is called the specification
of a model. The specification of an equation involves the selection of the dependent variable
(the Y variable); the variable/s that cause/causes the effect (the X variables) as well as the
functional form. The specification of a model can be simple in some cases and more difficult in
other cases.

Let us turn to a monetary economics example:

Monetary theory provides us with a simple specification problem. For many years, in the field of
monetary theory, the relationship between income (Y) and the money stock (M) was a hotly
debated issue, where Y is the real level of economic activity and M reflects the money stock. Of
course, changes in M arise mainly because of the number of net new loans created by banks. The
main issue was: Does M → Y, or does Y → M? This, of course, affects the way in which the
model is specified. The two schools of thought were the monetarists and the post-Keynesians.
According to the monetarists
Y=f(M) where the direction of causality runs from M → Y and where M is controlled by the
central bank. The monetarist transmission mechanism is both direct ( ΔM affects the prices
of assets which affects real spending) and indirect (ΔM affects the interest rate which
induces changes in real investment).

According to the post-Keynesians

M=f(Y), which is the current generally accepted view. The direction of causality is Y → M,
which is opposite to the monetarist case. The post-Keynesians believe that M cannot be
controlled. If Y increases, it causes an increase in the demand for M to finance the increased
level of Y.

The examples above show that econometrics depends on economic theory to provide the
variables involved, the direction of causality and the nature of the functional form.

Econometrics cannot resolve the theoretical differences between the two different schools of
thought. Causality depends only on theory. Econometrics can only determine correlation, which
is the strength and nature of a relationship. It cannot say anything about causality. For example,
econometrics will happily estimate both forms Y=a+bM and M=c+dY. If there is a good
correlation between Y and M, then both forms will indicate a good fit.
Economic data

In practice an econometrician must know the sources of economic data and should be aware of
methods of adjusting data to suit the need. Some knowledge of the methods of compiling
economic data is valuable, for example the system of national accounts. The second-year module,
Economic Indicators (ECS2603) is useful but is not a prerequisite for this module.

Data types
Time series data:
The same variable is measured over time (for example the real GDP for the period 1960 to 2021).
A subscript “t” is conventionally used to show time series data.
Cross-sectional data:
Provide a measure of several variables at a point in time (for example population census data for
various provinces as 1 March 2010). The subscript “i” is used to denote cross-sectional data.
Panel data:
A combination of cross-sectional data and time series data. As you could guess, “it” is used as
subscript to denote panel data.

Data are often adjusted to enhance their use. Examples of adjusted data are the following:
• A price index time series is adjusted relative to the price of a base year, for example the
consumer price index is expressed as 2005=100.
• Use of annual percentage change of a variable rather than its level – the inflation rate
calculated from the consumer price index.
• Adjustment of time series data to remove the seasonal effect/influences.
To remove the effect of price increases, values are often deflated (calculated at constant prices,.
for example, the use of real Gross Domestic Product (GDP) allows for the comparison of the
levels of real output of different years, without being affected by a change in the price level.
•
In contrast to the physical sciences, virtually all economic data are of the non-experimental type.
In the physical sciences, experimental data may be collected under controlled conditions. The
researcher may isolate the effects of a specific variable of interest because all other factors may be
held constant. In econometrics controlled experiments are almost impossible. The data observed
are of the non-experimental type, which reflects the combined impact of many variables
simultaneously. It is left to the econometrician to suggest causality; that is, cause-and-effect
relationships between variables.

Statistics

Econometrics makes extensive use of statistical techniques. Examples are the regression analysis
and hypothesis testing techniques.

Econometricians must be familiar with statistical concepts such as a sampling distribution, the
normal distribution, t-tests, the expected value of a sample estimate, standard errors and more.
These matters are discussed in chapter 17. The focus of this module is on teaching students how
to interpret and use statistical concepts. The focus is not as much on deriving estimators from
first principles as is often the case in econometrics courses.

Because of the unique nature of economic data and/or models, special statistical techniques have
been developed to cope with these difficulties. The latter part of the module focuses on the issues
of multicollinearity, serial correlation and heteroscedasticity.

1.2 Uses of Econometrics

Econometrics is used mainly for structural analysis, forecasting and policy evaluation.

Econometric models are used at different levels and degrees of complexity.

1. Most central banks have large, complex models of their national economies.
These are mostly simultaneous equation models which are used to direct monetary and
fiscal policy. The South African Reserve Bank uses an econometric model to forecast
inflation. The current inflation targeting monetary policy framework is inherently
forward looking and relies heavily on the forecasts provided by this model.
2. The Department of Finance uses an econometric model to forecast tax income
and to simulate the effects of alternative policy options.
 3. Commercial banks use econometric models to better understand how different
economic sectors and industries may react to shocks on the economy.
4. Basic models may be used by business and industry for forecasting and planning. A firm
may, for example, use a single equation model to determine the impact of its advertising on
sales.

Discussion point:
Identify other examples where econometric models are used.

✓ Structural analysis
Structural analysis relates to the quantification of economic relationships. Going back to the
consumption function: 𝐶 = 𝐶̅ + 𝑐𝑌 , where 𝐶=consumption expenditure, 𝐶̅ =autonomous
consumption, Y=disposable income and the coefficient c reflects the marginal propensity to
consume.
The value of c is a structural coefficient which indicates the tendency of consumers to spend additional
income on consumer goods and services (the marginal propensity to consume). Structural
analysis means that we gain quantitative knowledge about relationships between economic
variables.

✓ Policy evaluation

Economic models can be used by government to compare the effects of policy measures.
Alternative policy instruments are quantified and fed into an econometric model. The model is
solved to provide a quantitative outcome for each policy option.

✓ Forecasting

Forecasting entails a forward simulation of an econometric model. Assumptions are made about
the exogenous variables (the level of government expenditure, the gold price, the growth of
overseas economies, etc) and these variables are fed into the model, which then provides forecasts
of the endogenous variables (income, private consumption, etc).

Although many other means of forecasting and simulation do exist (and can be successful), there
are advantages of using an econometric model. A good econometric model represents a logically
consistent structure. The critically important variables have already been identified and the
relationships between the variables have been quantified. Using this model has the advantage
that the results are internally consistent, meaning that no important factors are ignored and that
proper account has been taken of the interrelationships between variables. The econometric
approach to forecasting is particularly useful in the medium to longer term when structural
relationships are more dominant than short-term or random effects.

1.3 What is Regression Analysis?

Assume that you have two data sets, a series of private consumption expenditure
(CONS) and a series of gross domestic product (GDP) data for, say the period 1980 to
2020. You believe that there is a macro relationship between CONS and GDP (the consumption
function) specified as follows:
 𝐶𝑂𝑁𝑆 = 𝑎 + 𝑏𝐺𝐷𝑃

where a and b are coefficients of the equation. Regression analysis is simply a technique to estimate
the coefficients of the equation from the data.

A typical cross section regression equation with more Xs is as follows:

𝑌𝑖 = 𝛽0 + 𝛽1 𝑋1𝑖 + 𝛽2 𝑋2𝑖 + 𝛽3 𝑋3𝑖 + ⋯ + 𝜀𝑖

Make sure you understand the meaning of dependent and independent variables, the meaning of
the coefficients of a regression equation and the meaning of the stochastic error term.

The estimated and true regression equation

Section 1.3 in the textbook focuses on the difference between the true regression
equation and the estimated regression equation. Make sure you understand figure
1.3, which summarises these concepts.

Source: Studenmund (2017:34)

The two basic concepts are:

• The true (population) parameters: 𝛽

• The estimated (sample) parameters: 𝛽̂
The true (population) coefficients in the relationship

𝑌𝑖 = 𝛽0 + 𝛽1 𝑋1𝑖 + 𝛽2 𝑋2𝑖 + 𝛽3 𝑋3𝑖 + ⋯ + 𝜀𝑖

represent the true relationship between the Xs and Y where the 𝜀𝑖 take care of deviations in Y not
explained by the Xs. Note that the 𝛽s are unknown.

The terms“ population” and “sample” originate from an area of statistics called sampling theory. A
sample is a subset drawn from the much larger population. The population is usually too large or
costly to enumerate in full; or it may be a theoretical concept that represents all possible outcomes
which cannot be observed. In this module we always apply regression analysis to sample data.

The stochastic error term versus the residual error term

• the stochastic error term (𝜀𝑖 ) which is the true but unobserved error term as in

𝑌𝑖 = 𝛽0 + 𝛽1 𝑋𝑖 + 𝜀𝑖

• the residual error term (ei) which occurs in the estimated equation

𝑌𝑖 = 𝛽̂0 + 𝛽̂1 𝑋𝑖 + 𝑒𝑖

1.4 A Simple Example of Regression Analysis

This section uses a weight-guessing equation to estimate

weight, given the height.

𝑊𝑒𝑖𝑔ℎ𝑡𝑖 = 103.40 + 6.38(𝐻𝑒𝑖𝑔ℎ𝑡𝑖 ) + 𝜀𝑖

Note that height cannot be a perfect predictor of weight because there is variation in weight caused
by other factors than height. It is, however, still a useful equation as it improves your ability to
predict weight. In economics we follow much the same approach. We try to estimate Y-variables,
given a number of Xs. Although the predictions for Y, given the Xs, are not perfect, we do try
to account for explanatory variables (Xs) which, we believe, are the major determinants of Y.

Discussion point:
Under which practical circumstance could one be interested in understanding factors that
influence weight? Which other factors do you think could be added to explain weight? What
type of data is used in this analysis?

1.5 Using Regression Analysis to Explain House Prices

This section introduces the house price equation (an example of cross-sectional data):
Does the size of a house influence its price? In the following cross-sectional model:
𝑃𝑅𝐼𝐶𝐸𝑖 = 𝛽0 + 𝛽1 𝑆𝐼𝑍𝐸𝑖 + 𝜀𝑖

Where PRICEi=the price (in thousand $) of the ith house

SIZEi= the size (in square feet) of the ith house
𝜀𝑖 = value of the stochastic error term.
The estimated equation is:

𝑃𝑅𝐼𝐶𝐸𝑖 = 40.0 + 0.138𝑆𝐼𝑍𝐸𝑖

Where 40.0 is the constant term and 0.138 is the estimate of the coefficient of SIZE.

Discussion point:
Interpret these results? Do the results make intuitive sense?
The data used in this exercise is available under Additional Resources. Use the data to
replicate the results. Do you think such a model is useful in the South African context?

1.6 Additional Reading (Optional)

1. Botha, B, de Jager, S, Ruch, F & Steinbach, R. (2017). Quarterly Projection Model of
the SARB. Working Paper – WP/17/01. Working Paper– WP/17/01: The Quarterly
Projection Model of the SARB (resbank.co.za)

1.7 Weekly Activity

Bringing the weight/height example to life!
On the blog page, provide your weight and height in the table provided. Do not include your
student number as this is an anonymous exercise. The data will be used to run a regression.
The results will be compared to the textbook example.