AI-Powered Direct Indexing: Exploring Thematic Universes for Enhanced Risk-Adjusted Returns

Abstract

This paper explores direct indexing (DI) in the stock market using FINDALL. FINDALL is our self-engineered Transformers-based search engine which detects thematically relevant stock tickers in websites and PDFs. We demonstrate creating thematic indices in minutes, compiling strategies from passive to leveraged momentum positions with external trading signals. We study thematic indices such as Bionic, Defense, Energy, and Luxury.

We benchmark our returns against flagship thematic exchange traded funds (ETFs) from major issuers. Using OLS regression, we test for index outperformance over its thematic ETF counterpart.

First, our key finding is that the rule-based FINDALL approach is more accurate in selecting thematically relevant stocks compared to portfolio managercurated ETF stock selection.

Additionally, our expense ratios are in all cases at least 66% lower than the industry average for thematic ETFs. Secondly, the Sharpe ratio indicators for all indices are higher than the ETF benchmarks showing enhanced risk-adjusted returns.

Keywords: Thematic Investing, Passive and Active Investment, Portfolio Optimization, Leveraged Positions, NLP, Direct Indexing

JEL Classification: C58, G11, G12