The publication related to this project 
The project aimed to develop a new market sectorization heuristic, with the goal of improving the risk-diversification profile of portfolios. The project built upon existing research on the capital structure irrelevance principle and the Modigliani-Miller theoretic universe conditions, postulating that corporation fundamentals - particularly those components specific to the Modigliani-Miller universe conditions - would be optimal descriptors of the true economic domain of operation of a company.
Market sectors play a critical role in the efficient flow of capital through the global economy. They provide a framework for investors to compare companies within the same industry and identify potential investment opportunities. Market sectors also impact the calculation of rates of return for industries, which in turn affect the pricing of products such as insurance. In addition, market sectors help to identify trends and shifts in the economy, allowing investors and analysts to make informed decisions about where to allocate capital. Overall, market sectors are an essential tool for understanding and navigating the complex landscape of the global economy.
Good market sectors are important for portfolio construction because they provide a clear and logical framework for grouping together companies with similar characteristics. This allows investors to easily identify and compare companies within a sector, and make informed decisions about how to allocate their capital. By grouping companies together based on fundamental characteristics, investors can create well-diversified portfolios that are less susceptible to market shocks and other risks. This can ultimately lead to higher risk-adjusted returns, as investors are able to take advantage of the unique opportunities and challenges within each sector. Good market sectors also enable investors to make more accurate predictions about the performance of different industries and companies, which can help them to make better investment decisions.
Motivating the Learned Sectors
The motivation behind the development of a new sectorization heuristic was the observation that existing sectorization heuristics, such as the GICS and the NAICS, were not entirely quantitatively driven. Instead, these heuristics appeared to be highly subjective and rooted in dogma. The goal of the "Learned Sectors" project was to develop a sectorization heuristic that was more fandamentals-driven, and therefore more objective and less susceptible to subjective biases. To that end, corporation financial data were used to create clusters of companies that were related, as indicated by their fundamentals.
The technical development of the Learned Sectors project involved the use of a hierarchical clustering algorithm to generate a set of potential candidate learned sector universes. The algorithm was varied in terms of the linkage method used and the number of resulting sectors derived from the model, resulting in a total of 60 candidate learned sector universes. A backtest-driven sector universe evaluation research tool called reIndexer was then used to rank the candidate sector universes produced by the learned sector classification heuristic.
Architecture of reIndexer order execution engine
The Learned Sectors project demonstrated that a fundamentals-driven sectorization heuristic can provide a superior risk-diversification profile than existing sectorization heuristics. The project showed that using corporation fundamentals as descriptors of the true economic domain of operation of a company can lead to a more objective sectorization scheme, with more correlated risk-adjusted performance when compared to GICS sectorization.
Backtesting the Portfolios
The Learned Sectors project also highlighted the value of backtesting as a tool for evaluating and ranking different sectorization heuristics. The use of the reIndexer tool allowed for the objective comparison of different sector universes, enabling the identification of the optimal learned sector universe. This was done by evaluating the performance of optimal minimum variance portfolios created with each of the candidate sectorization heuristics, and picking the best one.
Return of efficient minimum variance portfolios generated with each of the 60 sectorization heuristics
One potential application of the Learned Sectors project is in the development of investment portfolios. By using a fundamentals-driven sectorization heuristic, it may be possible to construct portfolios that are better diversified and have a more favorable risk-return profile. This could be of particular value to investors who are seeking to reduce the impact of subjective biases on their portfolio construction decisions.
Generated sectorization heuristic (with Complete Linkage, 17 sectors) outperforming the GICS sectors in absolute portfolio value and risk-adjusted return
By demonstrating the value of using corporation fundamentals as descriptors of the true economic domain of operation of a company, the project has shown that it is possible to develop more objective, and less subjective sectorization schemes. This work has the potential to improve the risk-diversification profile of investment portfolios, and to support more effective decision-making in the capital markets.