E CONOMETR ÍA II: S ERIES DE T IEMPO

W EEK 3.2

Juan R. Hernández1

Licenciatura en Economı́a, Spring 2025

1
I will draw heavily on the contents of the book of reference and some slides in its website.
1 / 18
Outline

1. MLRM: More Diagnostics and Refinements

2. Time Series Regression: Dynamic Models

2 / 18
1. Diagnostics and Refinements: Omitted Variables
Some immediate questions arise when, for example, some variables
are not statistically significant, but the F-test suggests that not all
estimates are zero.

This can be seen from the perspective of omitting an important

variable or including an irrelevant variable.
Omission of an Important Variable:
▶ Consequence: The estimated coefficients on all the other
variables will be biased and inconsistent unless the excluded
variable is uncorrelated with all the included variables.
▶ Even if this condition is satisfied, the estimate of the coefficient
on the intercept term will be biased.
▶ The standard errors will also be biased.
Inclusion of an Irrelevant Variable:
▶ Coefficient estimates will still be consistent and unbiased, but the
estimators will be inefficient.
3 / 18
1. Diagnostics and Refinements: Stability

So far, we have estimated regressions such as

yt = β1 + β2 x2t + β3 x3t + ut

▶ We have implicitly assumed that the parameters (β1 , β2 and β3 )

are constant for the entire sample period.
▶ We can test this implicit assumption using parameter stability
tests.
– The idea is essentially to split the data into sub-periods and then
to estimate up to three models, for each of the sub-parts and for
all the data and then to “compare” the RSS of the models.
▶ There are two types of test we can look at:
– Chow test (analysis of variance test)
– Predictive failure tests

4 / 18
1. Diagnostics and Refinements: Stability
To conduct the Chow Test, follow these steps:
(i) Split the data into two sub-periods. Estimate the regression over the
whole period (this is now the restricted regression) and then for the two
sub-periods separately (3 regressions). Obtain the RSS for each
regression.
(ii) Compute an F-test to assess the difference between the RSS’s:

RSS − (RSS1 + RSS2 ) n − 2K

test statistic = ×
RSS1 + RSS2 K

where:RSS = RSS for whole sample; RSS1 = RSS for sub-sample 1;

RSS2 = RSS for sub-sample 2; n = number of observations; 2K =
number of regressors in the “unrestricted” regression (since it comes in
two parts); K = number of regressors in (each part of the)
“unrestricted” regression.
(iii) Perform the test. If the value of the test statistic is greater than the
critical value from the F-distribution, which is an F(K, n-2K), then
reject the null hypothesis that the parameters are stable over time.
5 / 18
1. Diagnostics and Refinements: Stability
A Chow Test example:
▶ Consider the following regression for the CAPM Beta (β2 ) for
the returns on Glaxo. We are interested in estimating Beta for
monthly data from 1981-1992. The model for each sub-period is
▶ 1981M1–1987M10

r̂gt = 0.24 + 1.2rMt n = 82 RSS1 = 0.03555

▶ 1987M11–1992M12

r̂gt = 0.68 + 1.53rMt n = 62 RSS2 = 0.00336

▶ 1981M1–1992M12

r̂gt = 0.39 + 1.37rMt n = 144 RSS = 0.0434

6 / 18
1. Diagnostics and Refinements: Stability
The results of the Chow test are:
▶ The null hypothesis is
(1) (2) (1) (2)
H0 : β1 = β1 and β2 = β2

▶ The unrestricted model is the model where this restriction is not

imposed

0.0434 − (0.0355 + 0.00336) 144 − 4

test statistic = ×
0.0355 + 0.00336 2
= 7.698

▶ Compare with 5% F(2,140) = 3.06

▶ We reject H0 at the 5% level and say that we reject the restriction

that the coefficients are the same in the two periods.

7 / 18
1. Diagnostics and Refinements: Stability
Problem with the Chow test is that we need to have enough data to do the
regression on both sub-samples. An alternative is the predictive failure test:
▶ Estimate the regression over a “long” sub-period (i.e. most of the data)
and then predict values for the rest of the sample and compare the two.2
To calculate the test statistic:
– Run the regression for the whole period (the restricted regression) and
obtain the RSS.
– Run the regression for the “long” sub-period and obtain the RSS
(called RSS1 ).3
RSS − RSS1 n1 − K
test statistic = ×
RSS1 n2
where n2 = number of observations that the model is attempting to
‘predict’. The test statistic will follow an F(n2 , n1 − K).
2
There are 2 types of predictive failure tests: Forward predictive failure tests: keep the last few observations for forecast
testing (e.g. If we have observations for 1970Q1-1994Q4, estimate the model over 1970Q1-1993Q4 and predict
1994Q1-1994Q4). Backward predictive failure tests, the aim is to “back-cast” the first few observations (e.g. estimate the model
over 1971Q1-1994Q4 and backcast 1970Q1-1970Q4).
3
Note the label for the number of observations n1 in the “long” subperiod (even though it may come second).
8 / 18
1. Diagnostics and Refinements: Stability
An example of the Predictive Failure Tests. We have the following
models estimated for the CAPM Beta (β2 ) on Glaxo:
▶ 1981M1–1992M12 (whole sample)

r̂gt = 0.39 + 1.37rMt n = 144 RSS = 0.0434

▶ 1981M1–1990M12 (‘long sub-sample’)

r̂gt = 0.32 + 1.31rMt n = 120 RSS1 = 0.0420

Can this regression adequately ‘forecast’ the values for the last two
years? The test statistic would be given by
0.0434 − 0.0420 120 − 2
test statistic = × = 0.164
0.0420 24
Compare the test statistic with an F(24,118) = 1.66 at the 5% level.
So we fail to reject the null hypothesis that the model can adequately
predict the last few observations.
9 / 18
1. Diagnostics and Refinements: Stability
Ok...but how do we decide the sub-parts to use?
▶ As a rule of thumb, we could use all or some of the following:
– Plot the dependent variable, y, over time and split the data
accordingly to any obvious structural changes in the series:
1400

1200

1000

800
yt

600

400

200

0
1

129

161

193

321

417
33

65

97

225

257

289

353

385

449
Observation number

Figure: Source: Brooks (2019).

– Split the data according to any known important historical events

(e.g. stock market crash, new government elected).
– Use all but the last few observations and do a predictive failure
test on those.
10 / 18
2. Time Series Regression

We now know how to estimate (through OLS, LS, GLS, MLE, or

GMM) the parameters in a model.

We know how to conduct hypothesis testing. But we need to know

how to determine the following key aspects:
(i) Is the model fitting the data well? That is, is the independent
variables are able to explain the dependent variable?
(ii) Is the inference I am conducting sound? That is, are all
assumptions A1-A5 satisfied or modelled explicitly?
If the answer to either one of these is “No”, we need to re-think our
modelling approach.

11 / 18
2. Time Series Regression

Time Series assumptions on the LRM:

A1 The variable x does not contain a single repeated value (and if there
were any other x′ s they would be linearly independent).
A2 The Data Generating Process (DGP) is given by the stochastic process:

ut = yt − f (xt ), E(ut ) = 0, for all t = 1, 2, . . . , n.

A3mi The variable x is mean independent from u, E(u|x) = 0.

A4Ω The error process variance E(uu′ |x) is given by Ω. Allows for
heteroskedasticity and autocorrelation.
A5 The error process u is Gaussian: u ∼ N(0, Ω).

12 / 18
2. Time Series Regression

So what to do if we want to model Autocorrelation?

▶ If the form of the autocorrelation is known, we could use a GLS
procedure with elements in Ω of the form:

ut = ρ1 ut−1 + ρ2 ut−2 + ρ3 ut−3 + · · · + ρr ut−r + vt

▶ But GLS assumes that we know the form of the autocorrelation.

▶ However, it is unlikely to be the case that the form of the

autocorrelation is known, and a more “modern” view is that
residual autocorrelation presents an opportunity to modify the
regression...
▶ The London School of Economics (LSE) view: Estimate a
dynamic model.

13 / 18
2. Dynamic Models

We now introduce dynamic models. These models are a direct way to

address autocorrelation, but they are interesting in their own right.
They also happen to be an intuitive way to test theories and model
time series.

In dynamic models we extend the analysis to the case where the

current value of y depends (in addition to contemporaneous x′ s) on
previous values of y or the x’s. Call this an autorregressive distributed
lag model (ARDL). For example:

yt = β1 + β2 x2t + · · · + βk xkt + γ1 yt−1 + γ2 x2t−1

+ · · · + γk xkt−1 + ut

We could extend the model even further by adding extra lags, e.g.
x2t−2 , yt−3 .

14 / 18
2. Dynamic Models

There are many reasons why we might want (or need) to include lags
in a regression:
▶ Inertia of the dependent variable (e.g. inflation).

▶ Over-reactions in the market (e.g. after a policy surprise).

▶ Measuring time series as overlapping moving averages (e.g.

cumulated stock returns on a given period).
▶ Measuring the effects of a measure that took place in the past
(e.g. lags of the effect of changes in monetary policy on the
economy).

15 / 18
2. Dynamic Models

Another way to dynamic models can be expressed is in first

differences. This representation also helps to model explicitly
autocorrelation.
▶ Denote the first difference of yt , i.e. yt − yt−1 as ∆yt ; similarly
for the x-variables, ∆x2t = x2t − x2t−1 etc.
▶ The model would now be

∆yt = β1 + β2 ∆x2t + · · · βk ∆xkt + ut

Sometimes the change in y is purported to depend on previous values

of y or xt as well as changes in x:

∆yt = β1 + β2 ∆x2t + β3 ∆x2t−1 + β4 yt−1 + ut

16 / 18
2. Dynamic Models

Dynamic models may have a long run static equilibrium solution,

which is useful to test theories and discipline short-run dynamics.
▶ ‘Equilibrium’ implies that the variables have reached some
steady state and are no longer changing, i.e. if y and x are in
equilibrium, we can say

yt = yt+1 = . . . = y and xt = xt+1 = . . . = xt , and so on.

Consequently, ∆yt = yt − yt−1 = y − y = 0

▶ So the way to obtain a long run static solution is:
(i) Remove all time subscripts from variables.
(ii) Set error terms equal to their expected values, E(ut ) = 0.
(iii) Remove first difference terms altogether.
(iv) Gather terms in x together and gather terms in y together.

17 / 18
2. Dynamic Models

As always, there are trade-offs:

▶ We need to relax the assumption that the independent variables
are non-stochastic. (Not too complicated)
▶ Is (a little) harder to interpret: what does an equation with a large
number of lags actually mean?
▶ Note that if there is still autocorrelation in the residuals of a
model including lags, then the OLS estimators will not even be
consistent.

18 / 18