- author = "Jordan G. Milev",
- title = "Genetic Programming Use in Structural Modeling Applied to the Earnings-Returns Relation",
- school = "Yale",
- year = "2004",
- address = "USA",
- month = dec,
- keywords = "genetic algorithms, genetic programming",
-
URL = "
http://search.proquest.com/docview/305110188/AD4457348014431PQ/1?accountid=14511",
- size = "136 pages",
- abstract = "The selection of appropriate functional form for describing the relation between two economic variables has profound implications about the consistency and significance of estimated model parameters and about the predictions obtained from such a model. Until recently, nonparametric approaches have been the only solution to problems of model identification when the parametric form of the function is unknown. In the first part of the dissertation we develop an implementation of the algorithmic model selection technique of genetic programming (GP). We illustrate how it works and offer a brief comparison with nonparametric estimation methods. In the second part of the dissertation we specifically address a recent issue in the GP literature about overfitting and illustrate how it can be controlled. We also examine GP's ability to recognize a spurious regression and devise an illustrate a metric measuring the predictability of a data set using GP. In the third part of the dissertation we use GP to model how stock prices react to unanticipated accounting earnings. The result is a nonlinear parametric specification of the reaction of excess current-period stock price returns to the unexpected component of quarterly earnings. We confirm the existence of a nonlinear earnings response model that has superior in-sample and out-of-sample predictive power over the traditionally employed linear earnings regression. Our results have several implications: 1) it is important to incorporate forecast revisions in the earnings-returns specification; 2) when the earnings-returns relation is nonlinear, a nonsymmetric response to earnings announcements can be achieved even when the earnings response function itself is symmetric; 3) firms can affect the size of their earnings response coefficient by pre-announcing earnings; 4) appropriately accounting for the nonlinear form of the earnings-returns relation decreases the abnormal returns associated with earnings surprises. Our approach suggests an alternative to the linear earnings-returns relation which may provide suitable framework for future empirical work.",
-
notes = "http://www.genealogy.ams.org/id.php?id=118746
Supervisor: Peter Charles Bonest
Phillips
http://korora.econ.yale.edu/phillips/info/pcbvita-0908.pdf
UMI Microform 3152961",
