Carl Thornberg present his master's thesis
	
Portfolio optimization using hidden Markov models

Abstract

The goal of tactical asset allocation is to increase the short-term
return of a portfolio while restricting the risk. Investors often have a
long-term strategic asset allocation to fulfill requirements from
share-holders such as return on investment or not crossing some risk
level. This long-term strategic allocation could be given as percentages
of invested money to be held in different assets such as stock, bonds or
infrastructure, but these percentages are given as an interval of a few
percentages. The investor is thereby free to move inside these intervals
to increase the return of the portfolio by over-weighting assets that
performs well at the time.

This study uses Hidden Markov Models to predict future movements of
global indexes. Predictions of future movements generated by a
statistical model can be used to optimize the tactical asset allocation
and to increase portfolio return. External variables such as price
movements and credit spreads are used to incorporate the global business
and macroeconomic climate into the model.

Results show that the Hidden Markov Model used and examined in this
study performs poorly in the sense of predicting index movements 30 and
60 days ahead. The phases found by the ML-method has means gathered
around zero but different variance which gives information of the
riskiness of the market but it makes the model unable to the predict the
direction of future return. However, the results also show that the
model can separate volatile states from less volatile states, which is
useful when investors are interested in avoiding unnecessary risk exposure.