Related papers:

Earnings forecasts play a major role in several decision contexts such as equity valuation, acquisitions or divestitures, evaluation of accounting method changes, design of executive compensation plans, and performance evaluation of management. Previous research indicates that accurate forecasts of earnings could not be obtained using time-series techniques. This is mainly attributed to estimation biases and sampling errors due to small sizes of the samples used, and to the presence of outliers and structural changes in time series of earnings that adversely affect the model identification tools.
   Dr. Aksu's interests are in modeling and forecasting accounting earnings and other financial time-series using new procedures and software that have been developed in recent years for detecting and adjusting for outliers and structural changes.

Related papers:

In the context of forecasting, the relevant measure of performance to select among forecasting or forecast combination models is the efficiency (accuracy) of the ex-ante forecasts. Dr. Aksu's research interests in this area involve: i) evaluation of the desirable properties of forecast error measures, and ii) design and development of a methodology to create new error measures specially for evaluating combined forecasts.
   The focus on combination error measures is based on the arguments that such error metrics should be more sensitive, less outlier-independent, less stable, and less "typical" than original forecast error measures, and should possess an additional "conservatism" property.

Related papers:

A linear combination of forecasts obtained from models which may use different functional forms and/or information sets usually outperform individual forecasts. Although the superiority of nonnegativity restricted least squares (NRLS) combinations to ordinary least squares and equality restricted least squares combinations are demonstrated theoretically in Dr. Aksu's papers, there is only limited supporting empirical evidence.
   Dr. Aksu is interested in i) providing convincing empirical evidence of the superiority of NRLS combination models, ii) providing combined forecasts which outperform analysts' forecasts of accounting and financial time series, and iii) developing new robust regression-based heuristic combination methods which preserve the characteristics of the NRLS technique while minimizing its computational burden.

Related papers: