Nasdaq-100 Historical Constituents:
Every Change, 2007–2026

If you are looking for the historical membership of the Nasdaq-100, here is the shape of the record. Between February 2007 and August 2026, 273 distinct tickers held membership in the index. 103 are members today (the count exceeds 100 because of dual share classes and recent spin-offs); 170 have left — acquired, delisted, transferred to the NYSE, or dropped for falling below weight requirements. Across those 19.6 years the index recorded 228 constituent changes, an average of just under 12 per year, with individual years ranging from 7 changes (2010, 2014, 2024) to 21 (2015).

The turnover mechanism itself changed in June 2026: after decades of annual December reconstitution plus ad-hoc replacements, Nasdaq moved the index to a regular quarterly rebalancing schedule. The June 2026 rebalance replaced five names at once. If your mental model of NDX turnover is "one December shuffle per year," it is now out of date.

Every Constituent Change, 2025–2026

The table below is the complete change record for 2025 through 2026-08, taken directly from the dataset this site maintains. Additions are green, removals red.

Two details in this table repay attention. Versant (VSNT) was spun off from Comcast on January 5, 2026, entered the index automatically as a spin-off of a member, and was removed on January 9 — six trading days of membership. Insmed (INSM) was added in the December 2025 reconstitution and removed in the June 2026 one — in and out within six months. SpaceX (SPCX) was fast-tracked in on July 1, 2026 under new IPO-inclusion rules. Any backtest whose membership data is sampled monthly, or reconstructed loosely from announcement articles, gets these wrong.

How Companies Enter and Leave

Across the 228 recorded changes, the reasons break down into three broad groups:

The scheduled reconstitutions are the easy part — dated, announced, documented. The other ~45% of changes are the reason maintaining an accurate membership record is a continuous job rather than a one-time scrape: they happen on arbitrary dates, sometimes reverse within a week, and in the M&A cases they take the company's price history with them.

Why the Full Historical List Is Harder Than It Looks

The change events themselves are public record. If a list of dates and tickers were the whole problem, this article would end here. In practice, three things break naive reconstructions.

Tickers get reused by different companies

SNDK was removed from the index in December 2008 — that was SanDisk, the flash-memory maker, later acquired by Western Digital. SNDK was added to the index in April 2026 — that is Sandisk again, but a different legal entity, spun back out of Western Digital eighteen years later. A reconstruction keyed on ticker symbols alone will happily splice these two companies into one continuous price series. Similar traps exist wherever a symbol was recycled.

Companies leave and come back

Ross Stores was removed in the December 2007 reconstitution and re-added in December 2008. Seagate was added in 2008, left, and returned in the December 2025 reconstitution. A membership table that stores one entry date and one exit date per ticker cannot represent this. Correct point-in-time membership is interval-based: a company can have multiple disjoint membership periods, and about a dozen NDX names do.

The shortest tenures are shorter than your sampling frequency

Comverse Technology's final stint in 2007 lasted 9 trading days. The Liberty Media Braves tracking shares (BATRA/BATRK) lasted 19 trading days in 2016. Solstice lasted 31, Versant 6. If your membership snapshots are monthly — which is what most free reconstructions provide — these companies either never appear or appear for a month they were not actually tradable as index members. Daily granularity is not a nice-to-have; it is the difference between a record and an approximation.

How Point-in-Time Membership Should Be Structured

The representation that makes backtests correct by construction is a boolean matrix: trading dates as rows, all 273 historical tickers as columns, True where the ticker was an index member on that date. Universe selection then becomes a single lookup with no room for look-ahead:

import pandas as pd

pit = pd.read_parquet("ndx_pit_daily.parquet")   # 4,938 days × 273 tickers, bool

# The universe on any historical date — exactly what was tradable that day
universe = pit.columns[pit.loc["2013-05-20"]].tolist()
# ['AAPL', 'ADBE', ..., 'YHOO']   ← includes members that later left

# Interval-based membership survives re-entries (ROST, STX) automatically:
# each date is answered independently, no entry/exit pair assumptions

This structure handles every pathology above. Re-entries are just separate runs of True. Six-day members occupy six rows. Reused tickers are disambiguated by date. And because the universe for each backtest date comes from that date's row, it is structurally impossible to leak today's constituent list into a 2013 trading decision — the error that inflates Nasdaq-100 backtest returns by roughly 216 bps per year.

What a Complete Constituent Dataset Contains

Whether you buy it or build it, a historical constituent dataset you can actually backtest against needs four layers, and most free sources stop after the first: