The driverless stock index after a decade: performance and constituents reviewed

When the driverless stock index was introduced in October 2014, the consensus timeline had driverless vehicles in general commercial use by the early 2020s. Its twenty driverless stock index constituents spanned compute, sensing, mapping, vehicle manufacturing and tier-one supply, equally weighted so that no single share price could stand in for the whole thesis.

A decade later the thesis has been partly vindicated and the instrument has been thoroughly embarrassed. Judged on performance alone the driverless stock index looks like a success; judged on whether that performance came from the theme it was built to track, it does not. Driverless vehicles do carry paying passengers on public roads in several countries. They do it inside tightly bounded operating areas, roughly a decade later than the optimistic forecasts, and the companies doing it are largely not the ones the 2015 market favoured.

The driverless stock index constituents did not hold their shape

The most concrete finding has nothing to do with returns or performance. Of the nineteen recovered driverless stock index constituents, a majority went through a corporate event that an index has to have a written rule for. One was acquired outright and returned to public markets five years later under its own ticker. One divided in two, with only the autonomy half inheriting the slot and the remainder sold into a different industry. Two were renamed following reorganisations that involved no market transaction at all. One acquired, separately capitalised, publicly deployed, suspended and then reabsorbed an entire robotaxi subsidiary without ever leaving the index.

This is why the reconstitution rules are published in full rather than summarised. A thematic index whose handling of splits and renames is undocumented is not reproducible, and over a ten-year window in a consolidating sector, those events are not edge cases — they are most of what happens.

What performance the index actually recorded over the decade

The driverless stock index constituents were equal-weighted specifically to stop one of them dominating. One dominated anyway. The compute constituent's rise over the period dwarfed everything else on the list, and the overwhelming majority of that rise came from demand that had nothing to do with vehicles. A thematic index is structurally unable to separate the theme from a coincidence that happens to run through the same company, and no weighting scheme fixes that.

A decade of forecast error, not just delay

The sector did not simply arrive late. It arrived in a different shape. The 2014 framing assumed automation would be a feature of privately owned vehicles, which is the assumption underpinning the household-demand arithmetic in the UMTRI work of the period. What actually shipped first was a metered service in a handful of metropolitan areas, operated by companies that own their fleets. That inverts the ownership question the index was implicitly built around.

The safety argument aged better. The road-safety paper from the same programme argued that mixed automated and human traffic would produce failure modes belonging to neither, and that general driving was harder than the crash-rate comparison implied. Every operator that reached genuine driverless service did so by bounding its operating domain tightly — which is that argument, conceded in practice.

What a better instrument would have needed

Three changes would have made the D20 measure what it claimed to. It would have needed a rule excluding constituents whose automotive exposure falls below a threshold of revenue, which would have removed the compute constituent well before it began to dominate. It would have needed a mechanism for private-company exposure, since the sector's technical leaders spent most of the decade unlisted or inside subsidiaries that report nothing separately. And it would have needed reconstitution more often than annually, given how frequently the roster's corporate structures changed.

None of those were obvious in 2014, and the last one is the most interesting omission — nobody building a thematic index expects a majority of its constituents to undergo a structural corporate event within ten years.

What the index is good for now

Not as a return series, and not as a performance benchmark. As a record of how a sector reorganises itself under a thesis that takes twice as long as expected to arrive: which structures survived, which were dismantled, and which companies quietly changed what they were while keeping the same name. The full roster and rules are on the index page; what is actually deployed is on the deployment tracker.