Forecasting Stock Returns

NOVA SCHOOL OF BUSINESS AND ECONOMICS

FORECASTING STOCK RETURNS

Financial Econometrics Research Project

Group 8 Daniela Gameiro 701 | Diogo Guimarães 638 | Dora Ferreira 700 | Madalena Nascimento 650 | Maria José Prata 677

Teacher: Paulo Rodrigues T.A’s: Francisco Abreu; Miguel Silva

December 27, 2013

Abstract This research project focuses on return forecasting, namely, on the evaluation of models that are used by researchers and practitioners all over the world to accomplish this objective. Firstly, a brief literature review of the existing material tied to this issue was performed. Secondly, returns for the PSI20 (Portuguese stock composite index) were computed using three different approaches: the linear model, the sum-of-parts approach and the ARCH and GARCH-in-the-Mean models. Although some of the variables computed using these models are statistically significant, almost all the models presented show a low R-squared and a poor out of sample performance. These results may be partially influenced by the challenges that were faced in the collection of statistically relevant data for this project. Nevertheless, they also support the premise that although there have been great efforts, especially in the past 40 years, to find new and more efficient approaches to forecast stock returns, there still has not been created a model that fully accomplishes this goal.

Contents

1. 2. i. ii. iii. iv. 3. 4. Introduction ............................................................................................................................................... 2 Literature Review ...................................................................................................................................... 2 Economically motivated model restrictions .......................................................................................... 4 Forecast Combination...