Filtering techniques for asset allocation using a Discrete Time Micro-structure model: a comparative study
(2017) NEKN01 20171Department of Economics
- Abstract (Swedish)
- This paper is a comparative study of different approaches to using a Discrete
Time Micro-structure model. By using the three filtering techniques Extended
Kalman, Unscented Kalman and Bootstrap Particle, the hidden variables; excess
demand and market liquidity, were estimated and used in an asset allocation
strategy that invested in the asset when the excess demand as estimated as positive,
due to the assumption that positive excess demand would make the price go
up. Two different strategies were used—one based on threshold values of excess
demand and one binary approach simply using the sign of the excess demand—to
try to outperform a passive allocation strategy on 12 different stock indices. They
were then evaluated in terms of... (More) - This paper is a comparative study of different approaches to using a Discrete
Time Micro-structure model. By using the three filtering techniques Extended
Kalman, Unscented Kalman and Bootstrap Particle, the hidden variables; excess
demand and market liquidity, were estimated and used in an asset allocation
strategy that invested in the asset when the excess demand as estimated as positive,
due to the assumption that positive excess demand would make the price go
up. Two different strategies were used—one based on threshold values of excess
demand and one binary approach simply using the sign of the excess demand—to
try to outperform a passive allocation strategy on 12 different stock indices. They
were then evaluated in terms of average daily returns and market timing. The
results showed favourable average daily returns for the Extended and Unscented
Kalman filtering techniques using both kinds of strategies, though none of the results
were statistically significant at the 5% confidence level. The Bootstrap Particle
was deemed generally unreliable. The market timing tests rejected the null
hypothesis of no market ability for most data sets using all three filtering techniques,
with the two Kalman filters yielding the best results. Nothing was concluded
about which filtering technique was superior, though the study indicates
that Kalman filtering techniques can be used successfully in many cases while the
Bootstrap Particle filter as used in this thesis is not reliable. The threshold-based
strategy got slightly worse results in general than those of the binary approach,
but this was tested without taking transaction costs into account. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/8911201
- author
- Zakrisson, Henning ^{LU}
- supervisor
- organization
- course
- NEKN01 20171
- year
- 2017
- type
- H1 - Master's Degree (One Year)
- subject
- keywords
- portfolio management, kalman filter, asset allocation, hidden variable, state space, discrete time micro-structure model
- language
- English
- id
- 8911201
- date added to LUP
- 2017-07-10 13:52:51
- date last changed
- 2017-07-10 13:52:51
@misc{8911201, abstract = {This paper is a comparative study of different approaches to using a Discrete Time Micro-structure model. By using the three filtering techniques Extended Kalman, Unscented Kalman and Bootstrap Particle, the hidden variables; excess demand and market liquidity, were estimated and used in an asset allocation strategy that invested in the asset when the excess demand as estimated as positive, due to the assumption that positive excess demand would make the price go up. Two different strategies were used—one based on threshold values of excess demand and one binary approach simply using the sign of the excess demand—to try to outperform a passive allocation strategy on 12 different stock indices. They were then evaluated in terms of average daily returns and market timing. The results showed favourable average daily returns for the Extended and Unscented Kalman filtering techniques using both kinds of strategies, though none of the results were statistically significant at the 5% confidence level. The Bootstrap Particle was deemed generally unreliable. The market timing tests rejected the null hypothesis of no market ability for most data sets using all three filtering techniques, with the two Kalman filters yielding the best results. Nothing was concluded about which filtering technique was superior, though the study indicates that Kalman filtering techniques can be used successfully in many cases while the Bootstrap Particle filter as used in this thesis is not reliable. The threshold-based strategy got slightly worse results in general than those of the binary approach, but this was tested without taking transaction costs into account.}, author = {Zakrisson, Henning}, keyword = {portfolio management,kalman filter,asset allocation,hidden variable,state space,discrete time micro-structure model}, language = {eng}, note = {Student Paper}, title = {Filtering techniques for asset allocation using a Discrete Time Micro-structure model: a comparative study}, year = {2017}, }