Survivorship-Bias-Free Fundamental Indicators for S&P 500 Companies, 2009–2026

A quarterly panel of computed fundamental indicators for 692 companiesassociated with the S&P 500 index between 2009 and 2026, comprising 38,302company-quarter observations. The dataset addresses two biases that commonly invalidate backtests builtfrom readily available sources. SURVIVORSHIP BIAS. Most freely available S&P 500 datasets contain onlycurrent index members. This one additionally includes 195 companies that leftthe index during the period — retrieved separately from SEC EDGAR, since acompany's filings remain available after its ticker disappears. Each suchcompany carries the date it left the index and the reason (marketcapitalisation decline, acquisition, spin-off, or other), allowing users toreconstruct the index membership at a given point in time and to analyseexits by cause. LOOK-AHEAD BIAS. Every observation carries two dates: the end of thereporting period and the date the report first became public, taken from the"filed" field in SEC company facts. A quarter ending 31 March is not publicknowledge on 31 March — the median filing lag in this dataset is 35 days, andthe 90th percentile is 54 days. Analyses that filter on period end ratherthan publication date use reports that did not yet exist. CONTENTS. Indicators are computed from primary filings rather than copiedfrom a data vendor: net margin, return on assets, asset turnover, earningsquality (operating cash flow over net income), year-over-year revenue growth,and a partial Piotroski score. Annual rows are built from trailingtwelve-month figures, so they cover full twelve-month spans regardless of acompany's fiscal calendar. METHOD. Financial statements are taken from SEC EDGAR company facts in XBRLform. Indicators for each quarter are computed exclusively from dataavailable up to that quarter; later reports cannot influence earlier rows.Historical index membership is reconstructed backwards from the currentconstituent list using a public log of index changes. LIMITATIONS, STATED EXPLICITLY. (1) The index-change log extends back toJanuary 2016, so point-in-time membership can be reconstructed for a decade,while fundamentals reach back to 2009. (2) The dataset contains no price orreturn data; delisted tickers are largely unavailable from free pricesources. (3) The Piotroski score uses four of the nine original criteria —the remainder require balance-sheet items not present for all companies —and consequently lacks discriminating power. (4) Seventeen removed companiescould not be retrieved, mostly entities that ceased reporting before XBRLadoption or changed identity through merger. The dataset accompanies a Polish-language public website which documents themethodology in full and presents the same data per company.CHANGES IN 1.1.0Added two columns that were described in version 1.0.0 but missing from thefiles: "Poza indeksem od" (the date the company left the S&P 500; an emptyvalue means it is still a member) and "Powód wyjścia" (the reason: market-caprebalance, M&A trigger, spin-off, committee discretion, or other). Withoutthem the survivorship-bias claim could not be verified from the data itself,which was the point of the dataset. Indicators and their computation areunchanged.