Abstract
Nowadays, more and more investment techniques are incorporating model-based techniques to facilitate portfolio management process; however, techniques that could predict future stock expected returns are relatively scarce. This study mainly focuses on using model-based methods to evaluate stocks in the TMT (Technology, Media, Telecom) sector based on historical data for the last four years. LSTM (Long Short-Term Memory) neural network and Fama-French three factor models are employed in this study to predict the future expected return of the stocks and further, evaluate whether the stock could come into the optimal portfolio. Efficient frontiers are drawn using the variance of expected return against the mean of future expected return. Then investment utility function is set up along with the efficient frontier to make optimizer to get the best weight of the optimal portfolio. To justify if the model-based approach is robust, comparative study is done between basic qualitative approach and model-based approach. Two parallel approaches will use different methods as well as metrics to evaluate the same set of competitor stocks and generate the optimal portfolio. The results have shown that both approaches have the same optimal portfolio and the model-based approach is justified. Thus, this quantitative model-based approach is robust and applicable for investment since it could generate consistent result as the basic qualitative approach and it is more explicit in data.