6 – M2L6 11 Clustering Stocks V3

Let’s see how we would find cointegrated pairs from a universe of stocks. For example, we might look at companies listed in the S&P 500. If we were to compare every pair of stocks this would take a long time. More importantly, we would end up with spurious false positives just by chance. It’s important to partition our universe of stocks into meaningful groups. One way is to group stocks by sector or industry. However, since grouping by sector is quite popular, we want to find other relationships that are not as obvious but still meaningful. One way to find meaningful but less obvious groups is to use a class of unsupervised machine learning algorithms called clustering. The inputs to the clustering algorithm are the time series themselves. More similar time series get grouped together. Note that after clustering we still want to analyze the fundamental relationships between pairs of stocks that exhibit co integration.

%d 블로거가 이것을 좋아합니다: