Information Technology Reference
In-Depth Information
0.74
0.66
KGHM
07
PEKAO
07
PKNORLEN
07
TPSA
07
PKOBP
07
KGHM
08
PEKAO
08
PKNORLEN
08
TPSA
08
PKOBP
08
0.72
0.64
0.7
0.62
0.68
0.6
0.66
0.64
0.58
0.62
0.56
0.6
0.54
0.58
0.52
1
1.5
2
2.5
3
3.5
4
4.5
5
1
1.5
2
2.5
3
3.5
4
4.5
5
m
m
Fig. 6.
The correctness
Ψ
as a function of the memory length
m
, for 5 companies with
the biggest impact on index WIG20 for two different years 2007 (left) and 2008 (right).
case, as good as an ordinary GCMG or even as good as a linear filter. Indeed, we
checked that one of the strategies permanently outperforms others. Interestingly,
this strategy represents the mean-reverting approach, i.e. after history
μ
=1it
suggests
a
=
1 it suggests
a
= 1. Accordingly, the opposite
strategy representing a trend-follower approach, is the worst one. The results
are consistent with autocorrelation analysis where for
τ
= 1 the coecient is
negative. We checked that the results are general for all five examined firms.
The results for Polish market cannot be easily generalized to the London Stock
Exchange market. For example, companies like Vodafone or Astrazeneca are not
characterized by only one strategy being permanently better than others. The
more so, for Astrazeneca, the best results with the correctness equal to 0
.
6are
achieved for
m
= 2. It is dicult to explain why such fundamental differences
between stock markets exist. Their existence remains an open question.
The above analysis also shows that it is dicult to build a profitable investing
system if one would capture patterns using only agents' strategies. If a sign of a
next increment is known to the investor with encouraging probability, then the
investor has to put the order. But placing the order introduces a perturbation
to the forecast. If the order is executed, the system would move to the next
time step and the transaction would be considered as entailing the positive or
negative increment. The investor would predict only its own transactions what,
of course, does not assure any profit. Given this, the prediction for at least two
steps is required.
−
1andafter
μ
=
−
7Con lu on
We applied the minority game as a predictor of an exogenous process. We found
and explained that the degenerated game with only single agent and all strategies
from FSS is the most ecient configuration. If using the FSS is computationally
impossible, then the RSS is recommended. Considering the quality of prediction,
minority and majority games are equivalent. Considering non-stationary signals,
the parameter
λ
is introduced in order to speed up the model's convergence.
