Hidden clusters: grouping what moves together
Looking at individual cells is exhausting. What you seek is not a cell, but a block.
#Look for a block, not a cell
In the matrix of a twenty-position portfolio there are hundreds of cells. Reading them all one by one is neither possible nor necessary. What you seek is a cluster: groups of assets with high correlation to one another.
#How to find a cluster
By eye and with a simple method:
- Pick a threshold
For example, count above 0.6 as "same cluster." The threshold is arbitrary; try two different ones.
- Mark the pairs
- Chain to merge
- Name the clusters
- Sum the weights
- Rank
Step four is the most valuable: if you can give a cluster a name, you have understood it. If you cannot name it, either the cluster is fake or there is a driver you have not seen yet.
#What does clustering reveal?
Three typical discoveries:
1. A cluster you were unaware of. Assets you picked from different sectors can turn out tied to the same macro driver.
2. Piling into one cluster. Fifteen of your twenty positions may sit in one cluster. That is the real structure of your portfolio.
3. A fake diversifier. An asset bought "to diversify" may actually have been added to an existing cluster.
#Visualize the cluster weights
After finding the clusters, draw your portfolio by cluster. The distribution below is representative:
This view tells you what the position list never will: where the risk is concentrated. (Representative example.)
Yorumlar 0