CEO Tracker
Every CEO at every S&P 500 company since 2005, sourced from SEC filings.
Coverage
500 companies. Every CEO since 2005, with start and end dates pulled from the actual SEC filings that announced them. Founders or incumbents already serving when the window opens get anchored to the earliest filing that names them, which is usually that year's annual proxy. Nothing earlier than that is reconstructed.
Validated against the Gentry et al. CEO turnover dataset, the standard reference in management research (S&P 1500, 2000-2018). 408 of 500 companies fall inside Gentry's window. 806 CEO records match. Of the 445 cases where both sides have a precise date, 70% match to the day and another 13% are 1-day shifts (Dec 31 vs Jan 1 conventions). Around 5% disagree by more than a month. On the ones I spot-checked against the SEC filing, the SEC-sourced date was usually right. Where Gentry lists CEOs not in this dataset, most are either entity rebrands (different legal company under a new CIK) or attribution mistakes in their CSV, not missed transitions. Gentry stops in 2018, this runs to today.
How the data was built
Filings are filtered in stages, cheap to expensive:
- Pull every 8-K, 10-K, and DEF 14A filed by each ticker since 2005
- Sentence-level keyword and regex filter
- spaCy dependency parsing to confirm the named person is grammatically the subject of the transition verb
- Claude Haiku extracts structured records (name, role, date, predecessor, successor) from the survivors
- A tenure builder dedupes raw events into per-CEO stints, handling co-CEOs, interim-to-permanent promotions, mid-window CIK changes (Google -> Alphabet), and pre-window incumbents
- Manual verification on anything still flagged, with durable corrections written back, where necessary.
The web-search step has guardrails against the model writing today's date as an end date for currently serving CEOs. The baseline pipeline is separate because the main pipeline only sees transitions, not whoever was already CEO when the 2005 window opened.
Limitations
- Coverage starts in 2005. Earlier tenures are anchored, not reconstructed.
- Some dates are flagged inferred or tentative when the filing doesn't pin them to a day. Visible on company pages.
- Web verification handles structural issues. Not every individual date is independently fact-checked.
- When a ticker has changed entities over time (mergers, spinoffs, ticker reuse), only the current entity's CEOs are tracked.