Over-optimization or a Possible Strategy?

General
1

I currently invest in 12 countries in Europe, as well as Canada and the USA. I am generally very satisfied with my strategies for both regions. However, in constructing the ranking systems for these two regions, I have based them on the entire universe of available options—one system for the USA and Canada, and another for all of Europe.

Has anyone attempted to optimize their ranking systems to the individual country level and then merge them in some way afterward?

Country-Specific Optimization

Is this too much over-optimization, or is it a method for capturing the significant differences in the stock market landscape across various countries? For example, Norway is heavily weighted towards commodities, sharing similarities with Canada, while Sweden—a neighboring country—has a greater emphasis on technology and consumer goods.

Regardless, has anyone tried country-specific optimization, or what about sector optimization? What were the results, and how did you merge or utilize them after such optimization?

1 Like
2

This is another very good question.

We prewiously discussed this and the conclusion was: one system for the US, and another for Canada/Europe. However, my current setup is more granular. I’ve developed six separate systems:

US market (60%)

  1. US micro caps (60%)
  2. US + ADR utilities (portfolio hedge) (15%)
  3. US + ADR defensive non-cyclicals (portfolio hedge) (15%)
  4. US large caps, high dividend (alpha diversifier) (10%)

Non-US Small caps (50%)

  1. Europe/Canada ex-Poland(60%)
  2. Poland (40%) (not many stocks but well diversified economy)

Interestingly, my utilities and non-cyclicals systems perform very well. Yes, it is possible to beat the industry benchmark by a wide margin — but the factor structure is very different from what works in micro caps.

With ~1.5 years of live performance, I may eventually package this into a Designer Model.

3

Sector/industry “optimization” should honestly be standard practice given that different metrics imply different things for different industries. Some industries like banking do not even use some metrics like say gross margin/income. Also if you are reviewing a bank balance sheet deposits show as current liabilities so it will come up as risky if you use current assets to current liabilities so over-generalizing is objectively subobtimal. Of course part of whether the specific change is good depends on the process used

In terms of mechanics I use conditional nodes and eval and isna functions. I also separate ranks for drastically different industries like banks or REITs sometimes

1 Like