12.2 The Data
- z =
panstats(z{, T})
- computes summary statistics for the variables
|
Before we can estimate equation (12.3) we have to load the
XploRe quantlib metrics
for panel data analysis. This is
easily be done by typing
library("metrics")
The complete XploRe code needed for the subsequent analysis is
included in the quantlet
pantlet.xpl
. Next we load the
data set to XploRe. In general, a panel data set is assumed to be
ordered by the cross-section units. (If the data are ordered
according to the time index, the
pansort
quantlet can be
used to reorganize the data.) That is, the complete data of the
first individual is given in the first
rows, then the data
of the second individuals in the rows
and
so on. If the data set is unbalanced, the first two columns must
provide the identification number of the cross-section unit and
the time period. If the data are in balanced panel form, it is
sufficient to provide the common number of time periods
to
assign the data to the cross-section units.
The data matrix must be organized in the following form:
1 |
2 |
3 |
4 |
 |
3+ |
1 |
1 |
 |
 |
 |
 |
&vellip#vdots; |
|
|
|
|
&vellip#vdots; |
1 |
 |
 |
 |
 |
 |
2 |
 |
 |
 |
 |
 |
 |
|
|
|
|
 |
The first and the second column provide the cross-section
identification number and the time period respectively.
in the third column is the dependent variable, whereas the next
columns contain the explanatory variables. For more details on
the panel data set structure needed by XploRe see the XploRe
Learning Guide (Section 12.4.1).
Next we load the panel data from the data file uippanel.dat:
z=read("uippanel.dat")
This example data set has already the appropriate format for the
following analysis. To clarify the structure of a XploRe panel
data set we have reproduced rows 20 to 25 from the UIP data:
[ 20,] 1 92 95.217 -1.3578 2.1699
[ 21,] 1 93 97.091 -1.3401 1.3319
[ 22,] 1 94 96.962 -2.7027 -0.51471
[ 23,] 1 95 100 -1.0097 -0.20261
[ 24,] 2 73 87.671 1.2726 -0.07492
[ 25,] 2 74 89.041 0.51274 0.3582
The first column is the index of a specific country, e.g. 1
corresponds to Austria and 2 to Belgium (see Table 12.1 for all
country codes). The second column contains the year of the
observations and hence is the time index. Real effective exchange
rate index is given in column three. This index measures the
purchasing power relative to the OECD average. The base year of
this index is 1995.
In the fourth column of the data set we find the long-term
interest rate differential defined as the difference between the
long-term domestic real interest rate and its foreign counterpart.
The foreign interest rate corresponds to a weighted average of
long-term real interest rates from 17 industrial countries. More
precisely, this average is computed from interest rates of the
fifteen European Union members, Canada, Japan and the United
States left out the country the spread is computed for.
Finally, the fourth column is the short-term real interest spread,
i.e. the difference between the domestic short-term interest rate
and the foreign short term rate. Again, the foreign interest is
constructed by the procedure characterized above.
To detect missing values or potential problems with the data we
highly recommend to first compute summary statistics for the
panel. In XploRe this is conveniently done by typing
panstats(z)
where z is the data set. For the UIP data the output turns
out to be:
[1,]
[2,] N*T: 378, N: 16, Min T(i): 17, Max T(i): 26
[3,]----------------------------------------------------------
[4,] Minimum Maximum Mean Within Var.% Std.Error
[5,]----------------------------------------------------------
[6,] 30.96 638.9 117.1 54.74 67.37
[7,] -13.77 9.852 0.3543 70.82 3.082
[8,] -15.13 8.324 0.2948 79.63 2.978
[9,]
The summary table gives an informative overview of the data set
properties. First, we can easily see that there are 16 countries
included. Next and perhaps more interesting is the fact that the
UIP panel is unbalanced. The data set includes at most 26
observations from 1973 to 1998, however, for some countries the
interest rate spread is not available for the whole period. These
missings have been deleted prior to this analysis such that the
resulting data set is unbalanced.
The column "within Var.%" gives the fraction of variance
due to the within-group deviations. A zero in this column implies
that the respective variable is constant over time. This is
an important information for the estimation of fixed effects
models. Not surprisingly, in our example macroeconomic data set
none of the variables is constant over time. Since there are no
missings and no other problems with the data, we proceed with a
fixed effects model.