• American stocks
• financial economics
• Kelly criterion
• Kernel density estimation
• Machine Learning
• MATLAB
• non-parametric methods
• Portfolio optimization
• sequential investment strategies
• universal portfolios

Maximization of capital growth over time seems a more suitable goal for practitioners such as portfolio and fund managers than mean variance. This links to log-optimal portfolios which offer the best performance ex-post. The concept of universal portfolios has generated promising results even for ex-ante strategies. Nevertheless these results are asymptotic and often outcomes of complex methodologies. This work tries to analyze and test a non-parametric sequential investment strategy from the literature with very general assumptions, using kernel density estimation and machine learning, claiming not only great asymptotical results but also good finite-horizon performance. The test is performed using MATLAB software after some necessary adjustments on datasets made of American companies daily returns from CRSP till the end of 2018. The aim is to contribute this promising literature which however has very few practical examples of this kind, always with older data. Results are really encouraging displaying a very good annualized growth of capital compared to a set of benchmarks portfolios. However many variables and techniques are involved, therefore some important improvements research can make are outlined in addition to further uses of this methodology outside investments.