Multifractal approaches in econometrics and fractal-inspired robust regression

0546143 - ÚI 2022 RIV CZ eng C - Konferenční příspěvek (zahraniční konf.)
Kalina, Jan
Multifractal approaches in econometrics and fractal-inspired robust regression.
MME 2021, 39th International Conference on Mathematical Methods in Economics. Conference Proceedings. Prague: Faculty of Economics and Management, Czech University of Life Sciences Prague, 2021 - (Hlavatý, R.), s. 238-243. ISBN 978-80-213-3126-6.
[MME 2021: International Conference on Mathematical Methods in Economics /39./. Prague (CZ), 08.09.2021-10.09.2021]
Institucionální podpora: RVO:67985807
Klíčová slova: chaos in economics * fractal market hypothesis * reciprocal weights * robust regression * prediction
Obor OECD: Applied Economics, Econometrics
https://mme2021.v2.czu.cz/dl/99363?lang=en

While the mainstream economic theory is based on the concept of general economic equilibrium, the economies throughout the world have recently been facing serious transformations and challenges. Thus, instead of a convergence to equilibrium, the economies can be regarded as unstable, turbulent or chaotic with properties characteristic for fractal or multifractal processes. This paper starts with a discussion of recent data analysis tools inspired by fractal or multifractal concepts. We pay special attention to available data analysis tools based on reciprocal weights assigned to individual observations - these are inspired by an assumed fractal structure of multivariate data. As an extension, we consider here a novel version of the least weighted squares estimator of parameters for the linear regression model, which exploits reciprocal weights. Finally, we perform a statistical analysis of 31 datasets with economic motivation and compare the performance of the least weighted squares estimator with various weights. It turns out that the reciprocal weights, inspired by the fractal theory, are not superior to other choices of weights. In fact, the best prediction results are obtained with trimmed linear weights.
Trvalý link: http://hdl.handle.net/11104/0322694