This section should generally answer the question what model type corresponds best to the given data. Since the task is to fit a full VAR model the consideration is restricted to the question whether the given data set fits well in a full VAR framework. For this we note that the inference we want to make in Sections 17.3 and 17.4 requires data generated by a stable process. Stability implies mean and variance stationarity of the data. These features will be of interest in the following preliminary analysis.
It is good practice to start time series investigation by just visual inspection of the data graphs. We can view all time series in one chart or separate charts. Since we deal with multiple time series analysis we choose option one. This gives the following picture:
It is common to handle linear trend by differencing the data. Exponential growth can be transformed by applying the natural logarithm. Exactly these two transformations are supported. If both transformations are chosen the logarithmic transformation is automatically performed first.
Further transformations may be performed with
XploRe
before the data matrix
is given to
domulti
.
Here we choose both transformations for the series and
.
Since we deal with seasonally adjusted data we use the default differencing
lag of 1.
Now it is reasonable to assume mean and variance stationarity of
the and
series. However, at the beginning of
both series we still observe a period of high fluctuations
compared with the end. We might keep this feature in mind
for later steps of the analysis.