CMStatistics 2022: Start Registration
View Submission - CFE
Title: Projecting long-run compound returns: The limits of data-driven inference Authors:  Tamas Kiss - Orebro University, School of Business (Sweden) [presenting]
Erik Hjalmarsson - University of Gothenburg (Sweden)
Andreas Dzemski - University of Gothenburg (Sweden)
Adam Farago - University of Gothenburg (Sweden)
Abstract: What is the unconditional forecast, or projection, for the payoff of a stock investment over very long horizons, such as 30 years? Empirically, the question is inherently difficult to answer, since we, at most, observe a handful of actual 30-year returns. Some possible solutions have recently been proposed to deal with this problem (in particular, using bootstrap-resampling schemes to construct a large number of long-run returns), but little is known about the statistical properties of these methods. In particular, whereas some recent work has dealt with obtaining empirical point estimates of long-run stock return distributions, the central question of sampling uncertainty has been left mostly unanswered. We analyze the properties of different empirical methods used for projecting long-run return distributions. The aim is to provide an understanding of the limits of what we might feasibly be able to say, with some meaningful precision, regarding the distribution of long-run returns. We provide formal theoretical results on the properties of recently proposed bootstrap methods and contrast these with alternative parametric methods. Initial findings highlight that confidence bands around long-run distributions are very wide and often dwarf potential differences between empirical distributions under different distributional assumptions and inference methods.