B. Quality Analysis of Solution Found
In order to estimate the quality of the GA tracking portfolio
found, we performed several experiments. First of all, we
measured the above-described out-of-sample performance.
Figure 1 shows the out-of-sample performance compared to
the performance of the true AEX-index. We observe that
initially, both performances are almost equal while gradually
small differences emerge due to, among other things, changes
in the weights of the shares composing the AEX-index.
We further note that the up and down movements of both
portfolios are quite similar.Second, we calculated the out-of-sample tracking error as
defined by equation (5), respectively equation (6). The error
values found are presented in the first column of Table 4.In trying to understand the quality of these error values, we
compared them with the performance of randomly selected
portfolios. To do so, 50 portfolios consisting of 10 randomly
selected stocks were constructed and the optimal weights of
these stocks were calculated by solving minimization problem
(1). Next, we calculated the out-of-sample performance
of the randomly selected portfolios according to equations (5)
and (6), the averages of which are shown in the second
column of table 4. We conclude that, roughly spoken, the
optimal GA tracker, on average, performs almost four times
better than a randomly selected portfolio with optimized
stock weights.
We further selected the 10 highest, respectively lowest
capitalized stocks in the AEX-index using equal weights,
the tracking errors of which are shown in the fourth column
of table 4 and the tracking behaviors of which are shown
in figure 2. Actually, the tracking errors of the portfolio
consisting of the 10 highest capitalized stocks appear to be
quite small although the performance in terms of the value
of this portfolio is slightly worse than that of the index. In
addition we observe that the tracking errors of the lowest
capitalized stocks in the AEX-index appear to be higher
while the performance in terms of the value of this stock
is much better than the AEX-index. Finally we applied full
replication of the AEX-index, the performances of which are
shown in the last column. As expected, these performances
are very well although not 100% due to (small) intermediate
changes of the composition of the AEX-index in the out-ofsample
test data.