The construction of optimized portfolios usually involves estimating optimization inputs from an historical sample of returns, repeated for every rebalancing frequency. The error between the estimate and what transpired, is disregarded at every iteration. The Galton-algorithm exploits predictability in these errors over time, to generate superior optimization inputs. Originally developed on monthly data for US stocks, I employ the method on monthly, weekly, and daily data for the Norwegian stock market. I find that the strategy produces portfolios that outperform not only other optimal portfolios but also naïve equal- and valueweightingschemes. As more information is fed to the algorithm by increasing the frequency of datapoints, more accurate estimates are made, increasing the Sharpe ratio. These portfolios have the exciting feature of predictable portfolio variance ex-ante, making it possible for portfolio managers to manage risk levels in real-time.
Date of Award | 8 Jul 2021 |
---|
Original language | English |
---|
Awarding Institution | - Universidade Católica Portuguesa
|
---|
Supervisor | Pedro Barroso (Supervisor) |
---|
- Portfolio optimization
- Risk management
- Machine learning
Learning from out-of-sample errors in Norway´s stock market
Slensvik, J. U. (Student). 8 Jul 2021
Student thesis: Master's Thesis