Advanced Data Mining and Applications

Volume 7713 of the series Lecture Notes in Computer Science pp 516-526

Stock Trend Extraction via Matrix Factorization

  • Jie WangAffiliated withComputer Information Systems, Indiana University Northwest

* Final gross prices may vary according to local VAT.

Get Access


A diversified stock portfolio can reduce investment losses in the stock market. Matrix factorization is applied to extract underlying trends and group stocks into families based on their association with these trends. A variant of nonnegative matrix factorization SSMF is derived after incorporating sum-to-one and smoothness constraints. Two numeric measures are introduced for an evaluation of the trend extraction. Experimental analysis of historical prices of US blue chip stocks shows that SSMF generates more disjointed trends than agglomerative clustering and the sum-to-one constraint influences trend deviation more significantly than the smoothness constraint. The knowledge gained from the factorization can contribute to our understanding of stock properties as well as asset allocations in portfolio construction.


portfolio construction matrix factorization trend extraction clustering