Page 165 - Contributed Paper Session (CPS) - Volume 8

CPS2225 Retno Subekti et al.
excute the portfolio. Then, we apply updating views to renew the portfolio in
t and we observe the result. Comparing both Fig. 4 and 5 show that updating
view through MA-3 as a portfolio 2 give the best estimate and result rather
than a portfolio without renewing weight. The portfolio return is
underperformed for almost entire investment horizon time, it only
outperforms in short term which is t+11 until t+14 according to Fig. 2 and Fig
5. This is relevant with (6) when view is predicted with RBFNN it yields a best
performance based views for 10 period ahead. We cannot hold with views for
long term periods because the error will be increase.
Let say, |−1 as predicted mean conditional to the posterior return in t-1
and treat it as a renew Q. Then we described it as a procedure in building the
portfolio BLM with updating views. The following step by step is a procedure
incorporating updating weight in BLM:
1. Create an initial view as the difference between predicted mean ̂ +1
from time series method, such as MA and the return at t, . We

denote as = Δ ̂ +1,
0
2. Determine the t for execute the portfolio when it reaches the
expected return. When it does not gain the expected then hold the
portfolio without revision of the Q. Let say t as a time to execute then
we create new prediction based on |−1
3. Determine the posterior return as a new result on portfolio return

5. Conclusion
Based on the result, we investigate only two periods in evaluation and
show that investor’s view is no longer to be put in BL portfolio for the entire
investment horizon time. We demonstrate how to renew the weight based on
views updating in BLM. The research is still limited on practical for the building
new procedure and we ignore the component of error to be listed in this stage.
We will continue for the further research to be matched with mathematical
modelling.

154 | I S I W S C 2 0 1 9