Universität zu Köln

Theising, Etienne ORCID: 0000-0002-0213-3566 (2024). Statistical Methods for Large Financial Data Sets: Essays on Monitoring Cointegration and Distributional Reference Class Forecasting. PhD thesis, Universität zu Köln.

PDF DissertationEtienneTheising.pdf Download (2MB)

Abstract

This doctoral thesis covers two statistical methods with applications to financial data sets. Chapter 1 introduces monitoring statistics for structural changes in systems of cointegrating relationships based on parameter estimation over a calibration period. Different fully modified OLS estimators take into account the effects of error serial correlation and regressor endogeneity while allowing for cross-sectional dependence in homogenous and inhomogenous systems. An empirical application investigates deviations from the arbitrage parity condition for exchange rate triplets including Bitcoin and constructs a portfolio trading strategy based on the breakdates. Chapters 2 and 3 introduce approaches to handle expert or model-based forecasts exposed to (behavioral) bias by finding appropriate outside views. The approaches can also be used to forecast distributions and are illustrated by the case of corporate sales forecasts. The idea is to select reference classes for each analyzed company separately that share similarities to the firm of interest in reference variables. Chapter 2 uses a single co-variate as reference variable and Chapter 3 introduces rank based algorithms using multiple reference variables and an optional preprocessing data dimension reduction via principal components analysis. Reference classes are regarded to be optimal if the forecasted sales distributions match the actual distributions as closely as possible. The forecast quality of different reference variable sets and algorithms is backtested on a data set of 21,808 US firms over the time period 1950-2019. Algorithms using principal component analysis perform particularly well using past sales growth rates and past operating margins changes along with contemporaneous balance sheet and financial market parameters. Comparisions of actual analysts’ estimates to distributional forecasts and of historic distributional forecasts to realized sales growth emphasize the practical relevance of the method.

Item Type:	Thesis (PhD thesis)
Creators:	Creators Email ORCID ORCID Put Code Theising, Etienne etienne_stat@gmx.de orcid.org/0000-0002-0213-3566 UNSPECIFIED
URN:	urn:nbn:de:hbz:38-749083
Date:	2024
Language:	English
Faculty:	Faculty of Management, Economy and Social Sciences
Divisions:	Faculty of Management, Economics and Social Sciences > Economics > Econometrics and Statistics > Professorship for Statistics and Econometrics
Subjects:	General statistics
Uncontrolled Keywords:	Keywords Language hypothesis testing; monitoring; system cointegration; distributional forecast; outside view; prediction; bias correction; reference class English
Date of oral exam:	13 August 2024
Referee:	Name Academic Title Wied, Dominik Prof. Dr. Breitung, Jörg Prof. Dr.
Refereed:	Yes
URI:	http://kups.ub.uni-koeln.de/id/eprint/74908