Financial Development and Sectoral Output Growth in 19th Century Germany

2 Description of the data and preliminary analysis

Table 2: Results of Cointegration Tests

Note: ** and * indicate significance at 5% and 1% level by employing critical values from Osterwald-Lenum.
^{◦ ◦} and ^◦ indicate significance at 5% and 1% level for critical values from Cheung and Lai (1993). For Engle
and Granger (1987), ** and * indicate significance at 5% and 1% level using critical values from MacKinnon
(1991).

We start our empirical analysis, by testing the unit root properties of our time series. We
first apply the conventional Augmented Dickey Fuller test. In table 1 that reports the results
for our main variables, we can see that all of our time series are nonstationary in levels, but
stationary in first differences. The optimal lag length in the test specifications were chosen
by the Schwarz information criterion.

In the following sections of the paper we will estimate the causal linkages among our main
variables by using a vector autoregression. In this VAR our variables enter in logged levels and
we therefore need to check the cointegration properties of our data set as second preliminary
exercise (see table 2).

Overall, there is substantial evidence on cointegration among our time series, although in
some cases the evidence is mixed, when using different techniques of estimation. Using the
Engle and Granger (1987) approach, we find evidence of cointegration among all pairs of
time series that later enter the VAR analysis, except services and bank lending. We cannot
generally confirm cointegration with using the Johansen (1991) test, however. In particular
the three variable system of net domestic product, investment and bank lending as well as
some bivariate combinations do not appear cointegrated in this second approach.

Although there is only mixed evidence on cointegration we continue with the VAR speci-

<<   5   6   7   8   9   10   11   >>

More intriguing information