Introductory Econometrics for Finance

Chris Brooks
Solutions to Review Questions - Chapter 11

1. (a) There are several advantages from using panel data if they are
available:

 We can address a broader range of issues and tackle more complex

problems with panel data than would be possible with pure time-series or
pure cross-sectional data alone.
 It is often of interest to examine how variables, or the relationships
between them, change dynamically (over time). To do this using pure
time-series data would often require a long run of data simply to get a
sufficient number of observations to be able to conduct any meaningful
hypothesis tests. But by combining cross-sectional and time-series data,
one can increase the number of degrees of freedom, and with it the
power of the test, by employing information on the dynamic behaviour of
a large number of entities at the same time. The additional variation
introduced by combining the data in this way can also help to mitigate
problems of multicollinearity that may arise if time-series are modelled
individually.
 By structuring the model in an appropriate way, we can remove the
impact of certain forms of omitted variables bias in regression results.

(b) The seemingly unrelated regression (SUR) framework was initially

proposed by Zellner (1962). This has been used widely in finance where the
requirement is to model several closely related variables over time. A SUR is
so-called because the dependent variables may seem unrelated across the
equations at first sight, but a more careful consideration would allow us to
conclude that they are in fact related after all.

The example given in the book relates to the flow of funds (i.e. net new
money invested) to portfolios (mutual funds) operated by two different
investment banks. The flows could be related since they are, to some extent,
substitutes (if the manager of one fund is performing poorly, investors may
switch to the other). The flows are also related because the total flow of
money into all mutual funds will be affected by a set of common factors (for
example, related to peoples' propensities to save for their retirements).
Although we could entirely separately model the flow of funds for each bank,
we may be able to improve the efficiency of the estimation by capturing at

1
Introductory Econometrics for Finance by Chris Brooks

least part of the common structure in some way. Under the SUR approach,
one would allow for the contemporaneous relationships between the error
terms in the two equations for the flows to the funds in each bank by using a
generalised least squares (GLS) technique. The idea behind SUR is essentially
to transform the model so that the error terms become uncorrelated. If the
correlations between the error terms in the individual equations had been
zero in the first place, then SUR on the system of equations would have been
equivalent to running separate OLS regressions on each equation. This would
also be the case if all of the values of the explanatory variables were the same
in all equations - for example, if the equations for the two funds contained
only macroeconomic variables.

(c) A balanced panel has the same number of time-series observations for
each cross-sectional unit (or equivalently but viewed the other way around,
the same number of cross-sectional units at each point in time), whereas an
unbalanced panel would have some cross-sectional elements with fewer
observations or observations at different times to others.

2. (a) The fixed effects model for some variable yit may be written
yit   xit  i  vit
We can think of i as encapsulating all of the variables that affect yit cross-
sectionally but do not vary over time – for example, the sector that a firm
operates in, a person's gender, or the country where a bank has its
headquarters, etc. Thus we would capture the heterogeneity that is
encapsulated in i by a method that allows for different intercepts for each
cross sectional unit. This model could be estimated using dummy variables,
which would be termed the least squares dummy variable (LSDV) approach.
We would write this as
yit   xit  1 D1i   2 D 2i  3 D3i     N DN i  vit
where D1i is a dummy variable that takes the value 1 for all observations on
the first entity (e.g., the first firm) in the sample and zero otherwise, D2i is a
dummy variable that takes the value 1 for all observations on the second
entity (e.g., the second firm) and zero otherwise, and so on. As it is written in
the second of these two equations, this is now just a standard regression
model and therefore it can be estimated using OLS.

(b) An alternative to the fixed effects model described above is the random
effects model, which is sometimes also known as the error components
model. As with fixed effects, the random effects approach proposes different
intercept terms for each entity and again these intercepts are constant over

2
Introductory Econometrics for Finance by Chris Brooks

time, with the relationships between the explanatory and explained variables
assumed to be the same both cross-sectionally and temporally.

However, the difference is that under the random effects model, the
intercepts for each cross-sectional unit are assumed to arise from a common
intercept  (which is the same for all cross-sectional units and over time), plus
a random variable i that varies cross-sectionally but is constant over time. i
measures the random deviation of each entity’s intercept term from the
‘global’ intercept term . We can write the random effects panel model as
yit   xit  it , it  i  vit
where xit is still a vector of explanatory variables, but here the heterogeneity
(variation) in the cross-sectional dimension occurs via the i terms. Note that
this framework requires the assumptions that the new cross-sectional error
term, i, has zero mean, is independent of the individual observation error
term, vit, has constant variance and is independent of the explanatory
variables.

(c) It is often said that the random effects model is more appropriate when
the entities in the sample can be thought of as having been randomly
selected from the population, but a fixed effect model is more plausible when
the entities in the sample effectively constitute the entire population (for
instance, when the sample comprises all of the stocks traded on a particular
exchange).

More technically, the transformation involved in the GLS procedure under the
random effects approach will not remove the explanatory variables that do
not vary over time, and hence their impact on yit can be enumerated. Also,
since there are fewer parameters to be estimated with the random effects
model (no dummy variables or within transform to perform), and therefore
degrees of freedom are saved, the random effects model should produce
more efficient estimation than the fixed effects approach.

However, the random effects approach has a major drawback which arises
from the fact that it is valid only when the composite error term, it, is
uncorrelated with all of the explanatory variables. The assumption is more
stringent than the corresponding one in the fixed effects case, because with
random effects we thus require both i and vit to be independent of all of the
xit. This can also be viewed as a consideration of whether any unobserved
omitted variables (that were allowed for by having different intercepts for
each entity) are uncorrelated with the included explanatory variables. If they
are uncorrelated, a random effects approach can be used; otherwise the fixed

3
Introductory Econometrics for Finance by Chris Brooks

effects model is preferable. A test for whether this assumption is valid for the
random effects estimator is based on a slightly more complex version of the
Hausman test described in Chapter 7. If the assumption does not hold, the
parameter estimates will be biased and inconsistent.

4. (a) The key reason for using a panel unit root test rather than a set of
separate tests on each individual series is the increase in power that can arise
as a result of the increased sample size from a multivariate approach. It is
widely known and documented that conventional single series unit root tests
lack power (they fail to reject the null hypothesis of a unit root even when it is
wrong), especially when sample sizes are small. A panel approach may be
particularly useful when the number of available series, N, is quite large, but
the number of time series observations, T, is quite small.

There are several disadvantages of taking a panel approach, however. The

first is the obvious increase in complexity – the testing framework itself more
sophisticated, with the problem of having to deal with the issue of ‘nuisance
parameters’, which will affect the distribution of the test statistics and require
separate Monte Carlo simulations in each case to obtain the appropriate
critical values. But even the appropriate conclusion to draw is not entirely
clear in the panel unit root context since it may be the case that some of the
series in the panel are unit root processes while others are stationary. The
simpler panel unit root tests assume a common autoregressive parameter
under the alternative hypothesis while the more sophisticated tests allow
each series to have it’s own process under the alterative. The latter are clearly
more general but bring with them an additional layer of complexity.

(b) As alluded to in the answer to part (a), the early unit root tests required all
series in the panel to have the same first order autoregressive parameter
under the alternative hypothesis. Thus the hypotheses are: H0: i   = 0 and
H1:  < 0  i respectively. This approach is used by, for example, Levin, Lin
and Chu (2002) and Breitung (2000). Requiring the autoregressive parameter
to be the same in all cases is, of course an assumption which may or may not
be supported by the data. Im, Pesaran and Shin (2003) developed a more
general framework that allows for heterogeneous autoregressive dynamics.
The null hypothesis is the same as above and under the alternative: H1 : ρi <
0, i = 1, 2, . . . ,N1; ρi = 0, i = N1 + 1,N1 + 2, . . . ,N. So the null hypothesis still
specifies all series in the panel as nonstationary, but under the alternative, a
proportion of the series (N1/N) are stationary, and the remaining proportion
((N − N1)/N) are nonstationary. No restriction where all of the ρ are identical is
imposed.