This paper presents a new, objective data-driven approach to market sectorization based on company fundamentals. The authors analyze existing sectorization heuristics and find that they are not entirely quantitatively driven, but rather appear to be subjective and rooted in dogma. They then examine alternative approaches to market sectorization and find that returns-based methods are flawed due to bias in the existing classifications. The authors propose using corporation fundamentals as descriptors of the economic operating domain of a company.
To develop the sector classification heuristic, we use fundamentals data from Form 10-K for 362 companies in the S&P 500. We developed a hierarchical clustering algorithm to generate a set of candidate sector universes and use a backtest-driven sector universe evaluation tool to rank them. The tool backtests portfolios of synthetic ETFs based on the specifications of a candidate sector universe. We found that their learned sector universe portfolio outperforms the benchmark with respect to absolute portfolio value and the risk-adjusted return of the portfolio. In conclusion, our approach provides a superior risk-diversification profile compared to the existing classification heuristic.