A New Strategy for Stock Pricing Based on MachineLearning Algorithms

Stock price prediction is a complicated and interestingproblem. Noisy trends make stock pricing sensitive andcomplicated while the economical motivation behind, keeps itinteresting for researchers and investors. In this paper we are tooutline two novel ideas for combining well-known machinelearning based methods to provide an improved prediction. Thefirst idea is based on using the best algorithm between the selectedchoices until its performance decrease during short periods. Thesecond idea is to assign an index to the best method in each dayand interpolate the index of the method as a polynomial functionof the day index. The interpolation might be used to estimate theindex of the best method for each test date. We also test eachsuggested algorithm for predicting the price of ۲ stocks fromnational market. To show the efficiency of our proposedalgorithm, we compare the predicted prices with real values overa test set. Moreover, a backtest analysis is performed to verify thecloseness of annual returns based on real and predicted prices.

Keywords:

LSTM , Linear Regression , Moving Average , Incremental Regression