Thematic ETFs have become a standard building block in modern portfolios. They offer clean narratives: artificial intelligence, energy transition, cloud computing, cybersecurity. For allocators, they are efficient instruments to express a view. For analysts, they create a structural challenge. An ETF labeled “AI” is not an exposure in itself. It is a container of individual securities, each with its own sector classification, factor profile, and risk contribution. If we stop at the wrapper, we are analyzing product labels rather than the actual portfolio.
Consider a multi-asset portfolio that holds several thematic ETFs alongside direct equities. The manager wants to understand effective exposure to AI or clean energy. Summing ETF weights provides a capital allocation number, but it does not reveal how much of the underlying book sits in semiconductors versus software, or how concentrated the exposure is across a handful of mega-cap names. Overlap quickly becomes an issue, as many thematic ETFs share the same large-cap growth stocks. Counting each theme independently can lead to double counting at the security level. On top of that, ETF holdings drift over time. Constituents change, weights shift, and static mappings become outdated. Manual tagging is possible, but it does not scale.
The lookthrough principle is straightforward in theory. Instead of treating an ETF as a single line item, we decompose it into its constituents and re-aggregate exposures at the security level. If a portfolio holds 5% in a thematic ETF and that ETF holds 10% in Company X, the effective exposure to Company X is 0.5%. Repeating this across all ETF constituents and merging the result with directly held positions produces a unified security-level book. From there, any standard attribution framework can operate as usual, whether the focus is sector allocation, security selection, or factor tilts. The idea is simple. The implementation is not.
At Kiski, we built an automated pipeline to handle this process in a systematic way. ETF-level positions are parsed and matched with third-party constituent data. Holdings are mapped to our internal security master and enriched with economically meaningful classifications, including sector, style, and thematic groupings. The system expands ETF exposures into security-level weights, merges them with direct holdings, and feeds the combined dataset into downstream attribution models and dashboards. The emphasis was practical: avoid daily manual tagging, avoid static definitions, and preserve consistency with the existing attribution logic. When constituent data updates, the changes flow through automatically.
This matters particularly in thematic investing, where narratives can obscure structural concentration. Once we look through the wrappers, patterns become measurable. A portfolio diversified across multiple innovation ETFs may still concentrate heavily in a small set of technology names. Energy transition funds may tilt more toward industrial supply chains than utilities. AI exposure may be dominated by semiconductor equipment rather than software platforms. None of this is surprising, but without systematic lookthrough, it remains implicit rather than quantified.
Traditional Brinson-style attribution decomposes returns into allocation and selection effects relative to a benchmark. The same logic applies after lookthrough. Once ETFs are decomposed into their constituents, indirect exposures are no longer hidden. Allocation effects reflect overweight or underweight positions at sector or theme levels, and selection effects capture security-level contributions within those groupings. A manager holding a thematic ETF is still making underlying sector and factor bets, even if they are expressed indirectly. The attribution framework simply makes those bets visible.
Operationally, the challenge is data hygiene. Constituent files must be timely and version-controlled. Security identifiers need consistent reconciliation. Corporate actions and weight changes must be incorporated without breaking historical continuity. For this reason, we integrated lookthrough directly into the regular analytics pipeline rather than treating it as a special analysis performed only during periods of stress. The goal is not perfection. It is a repeatable process that improves transparency over time.
Thematic ETFs will likely continue to grow in number and specialization. As wrappers multiply, the distance between label and underlying exposure increases. Automating lookthrough attribution does not change the economics of the portfolio. It simply makes them visible. Once the container is opened and exposures are decomposed, portfolio discussions become more grounded. Themes remain useful as organizing ideas, but they become decomposable risk exposures rather than abstract narratives.
