Creating New ETF Indexes

[pmcgover]

Greetings,
Exchange Traded Funds (ETF) are interesting investment tools for a variety of reasons. As a programmer/analyst it would seem possible to create a process that could generate and test ETFs for use in real time markets. Below is my high level concept of the ingredients for this system.

If this is worth doing (see last question), then please answer the other questions. If it is not worth doing, please explain why...

Work flow steps for an ETF generating system:

1) Obtain a historical list of companies with their performance details. The data set should include performance details like stock price, earnings, revenue, dividends, book value, etc. The data will be used to back test all the ETF test scenarios.

Questions: What are the best sources for this data? What are the most affordable sources?
What time intervals should the data be in, weekly, quarterly? How old are the newest records (1 week)?

2) Load the data into a database for retrieval/processing.

3) Create a population of sample ETF indexes: Use programming algorithms that will create many sample ETF collections with each set having a different mix of companies and proportions. The formula for each ETF index must be understandable and repeatable by humans and describe the proportion of each stock. The population sets could target broad categories or specific sectors.

4) Back test the sample ETFs against historical data: Use the index formula to create index sets at a historical point, test them and re-balance them on a periodic basis throughout the test. Compare the results with other benchmark indexes and ETFs.

5) Select a small population of top performing sample ETFs. Include them in future back testing. Review them periodically and cull out the low performers.

6) Forward test the best sample ETFs: Include the best sample ETFs in an ongoing forward test. Review them periodically and cull out the low performers.

Question: How long should a good sample ETF be forward tested before they could be considered "ready for market"?

7) Analyse the overall results: Use statistical analysis to determine the significance of the trends and patterns. Look for correlations that could be used to improve the algorithms. Consider refactoring the algorithm if needed. Consider adding new algorithm modules to diversify the population.

8) Repeat the above process: Update the historical list of target companies, create new sample ETFs for a variety of market categories/sectors. Promote the best sample ETFs and cull the rest...

9) Market the best ETFs:
Questions: How do you make money with this model? What is the criteria for a marketable ETF? Do you develop a marketing campaign and try to sell the best ETF formulas or negotiate for a percent of the operating expense? Do you try to sell the entire algorithm process? How many companies could be potential buyers? Is the market saturated with well tested but new ETFs?

10) Is this worth doing?

Thanks!
Pat

[pmcgover]

It would be interesting to see the scope and coverage that Wisdom Tree's patents have for this process. If they can legally claim a process at a HIGH LEVEL like I described above then they have the market cornered.

However, if it covers a specific, granular technique to generate new models, then it should not be difficult from a programming standpoint because there are many different ways to optimize a problem.
P