A Decade of D20 Constituent Changes: Who Was In, Who Dropped Out

An index is only useful if it still describes the thing it claims to track. The D20 index has existed for roughly a decade now, and in that time the autonomous-vehicle industry has not stood still long enough for any fixed list of twenty companies to stay accurate. Programs launched. Some merged. A few got acquired outright, and others shut down entirely, sometimes within a year or two of looking like leaders. None of that turnover is unusual for a young, capital-intensive technology sector; what would be unusual is expecting an index to hold still while the sector underneath it does not.

The categories of change are not interchangeable, even though all of them eventually show up as the same line in a constituent-changes table. An acquisition folds one company's autonomous-vehicle work into a much larger parent's balance sheet, sometimes preserving the original brand as a subsidiary and sometimes not. A shutdown removes a name from the list because the underlying program stopped operating, not because it changed hands. A merger between two constituents collapses two index slots into one without either company disappearing outright. Treating all three as the same kind of event, a company that simply "left the index," flattens distinctions that matter to anyone trying to understand what actually happened to a given firm rather than just noticing that its name is gone from the list.

That churn is not a flaw in how the index was built. It is a description of the industry itself. A company that looked central to autonomous vehicles in year one of tracking might be a division of a larger firm by year five, or gone altogether by year eight. Freeze the constituent list at inception and refuse to touch it, and within a few years the index stops representing anything real. It becomes a historical snapshot wearing the label of a current index, which is worse than no index at all because it still looks authoritative to anyone who doesn't check the fine print.

Periodic reconstitution is the only way around that problem, and it is not a cosmetic exercise. Deciding when a company has drifted out of relevance, when a merger collapses two constituents into one, or when a shutdown removes a name entirely requires judgment calls that need to be made consistently over time rather than reactively whenever something in the news forces the issue. Consistency, more than any single call, is what keeps a decade of reconstitution decisions from looking arbitrary in hindsight.

There's a real tension underneath all of this that a decade of reconstitution doesn't resolve so much as manage. An index that changes its constituent list too readily loses the comparability that makes tracking it worthwhile in the first place; a reader trying to compare this year's performance against five years ago wants roughly the same set of companies represented in both periods, not a list substantially rebuilt in between. An index that changes too reluctantly, on the other hand, ends up carrying dead weight: a constituent that no longer does the work the index exists to measure, kept in place only because removing it would mean admitting the list needs updating more often than anyone would like. Every reconstitution decision sits somewhere between those two failure modes, and getting the balance right has less to do with any individual call than with applying the same threshold consistently across all of them.

This site's ten-year retrospective documents the specific reconstitution decisions made across that span, along with the rules behind them. Some of those decisions were straightforward. Others involved companies that didn't disappear so much as fade into ambiguity, still technically operating but no longer doing anything that resembled the autonomous-vehicle work that got them included in the first place.