Stock exchanges are one of the major areas of investment because of the possibility of high returns and big winners. They are affected by a variety of factors making it difficult to get consistent returns and accurate predictions when using systematic forecasting techniques.
We consider a portfolio formation problem by comparison of the trend strengths of multiple assets. The trend strength determined by the slope and errors from the regression line provides a useful method for crosssectional comparison of stocks.
We use weekly and monthly data from 1965 to 2018 from the CRSP US Stocks Database to test the performance of these factors when used to predict the direction of movement for an asset in the future. We investigate the feasibility of this two factor model and various methods of combination to determine the optimal stock trend forecasting model.
EMD2FNN: A strategy combining empirical mode decomposition and factorization machine based neural network for stock market trend prediction