A.8.3 Commercial Providers

Basic Prerequisites: 02 · SEC EDGAR & XBRL.

---

1. Intuition & Practical Objective

Everything a commercial provider sells you is EDGAR plus a standardization decision. The filer tags RevenueFromContractWithCustomerExcludingAssessedTax; Compustat calls it SALE; Sharadar calls it REVENUE; the vendor has decided how to treat discontinued operations, minority interest, SBC, operating leases, and fiscal-year alignment - and different vendors decided differently. Standardization is the product, and standardization is also the risk. The job of this page is to make you a literate consumer of that product: to know what each vendor actually gives you, and to reconcile them when they disagree rather than pretending they don’t.

Four provider families, four different jobs:

• Sharadar SF1/SEP (via Nasdaq Data Link) - the affordable, retail-accessible standardized US fundamentals + prices, with ~20 years of history. This is the backbone of the club’s own screening blueprint and of the company-tracker architecture in 06.
• Nasdaq Data Link (formerly Quandl) - the platform that hosts Sharadar and many other feeds behind one API; not a data vendor itself.
• WRDS Compustat (+ CRSP link) - the academic standard, the database behind virtually every published factor study. Crucially, it comes in two flavours: the default (restated) and the Point-in-Time edition.
• Bloomberg / FactSet - the institutional layer: global coverage, estimates, ownership, corporate actions, and (for backtests) genuine as-of fields. Best-in-class and effectively un-reproducible outside a licence.

There is also the free scraped aggregator (Yahoo quoteSummary): fast, universal, and not point-in-time - useful for a sanity check, dangerous as a signal source.

---

2. Mathematical Ground Truth & Derivations

The standardization map. A provider is a function from filer tags to a fixed contract:

$$\text{provider}:\{\text{us-gaap tags}\}⟶\{\text{provider fields}\},{\Phi }_{\text{Sharadar}}\ne {\Phi }_{\text{Compustat}}\ne {\Phi }_{\text{Bloomberg}}.$$

Because the maps differ, the same economically identical firm-period yields different displayed values. The honest response is triangulation: report a central value plus a spread, and document the basis of each contributor.

The point-in-time distinction, stated as a database property. Let $\mathcal{D}(t)$ be the vendor’s database as it stood on date $t$:

$$\text{PIT}⟺\mathcal{D}(t)\text{is retrievable for historical}t\text{vs.}\text{“restated”}⟺\text{only }\mathcal{D}(t_{\text{now}}).$$

The default Compustat product (and Yahoo, and Damodaran’s snapshots) exposes only $\mathcal{D}(t_{\text{now}})$: today’s restated values. Backtesting a screen on that is backtesting on information that did not exist at the time - the inflation is not hypothetical (§3 and 05 both measure it).

The cost/hygiene frontier. Providers trade along two axes, and the naive assumption - “more expensive is more point-in-time” - is false:

Provider	~Cost/yr	PIT?	Note
SEC EDGAR + XBRL	$0	✅ by construction	authoritative, but build-it-yourself
Sharadar SF1 (Nasdaq Data Link)	~$500	✅ (ARQ/ART as-reported tables)	the value pick for a retail pipeline
WRDS Compustat	academic seat	❌ default	restated; the reason PIT exists as a separate product
WRDS Compustat PIT	academic seat	✅	purpose-built to remove look-ahead/survivorship
Bloomberg / FactSet	$$$$$	✅	best coverage; licence-bound, not reproducible

The sharp point: the most expensive default products are not automatically the most honest. Compustat’s standard file is restated; the PIT edition is a distinct purchase. Paying more buys coverage and features - not, by itself, point-in-time hygiene.

---

3. Computational Implementation - reconciling three providers to one canonical EBITDA

Offline and fully runnable (stdlib only). It maps three providers’ native field names onto one canonical schema, then quantifies the disagreement - the number you must carry whenever you mix vendors.

A 5.7% spread on the same firm-year is not a bug in any one provider - it is the standardization choice showing through. Notice the practical consequence: the EV/EBITDA a screen computes can move from 9.24 to 9.78 without the company doing anything. Multiply by a few hundred names and provider choice alone reshuffles a screen.

Provider selection cheat-sheet.

Need	Use	Why
Free, deep, honest history of US fundamentals	EDGAR	authoritative, PIT by construction (02)
20+ yr standardized fundamentals + prices, cheap	Sharadar SF1/SEP	the retail value pick; as-reported tables
Reproduce published academic factor results	Compustat + CRSP link	what the papers used
Backtest without look-ahead	Compustat PIT (or a self-built EDGAR panel)	restated default is not enough
Global coverage, estimates, ownership, live	Bloomberg / FactSet	the institutional standard
Quick eyeball of a ratio today	Yahoo quoteSummary	free - but no as-of date, never a signal

---

4. Failure Modes & First-Principles Breakdowns

1. Mixing vendors without a reconciliation layer (first principle: each map $\Phi$ differs). Joining a Sharadar revenue to a Compustat EBITDA produces a margin that belongs to no consistent schema. Rule: one canonical schema, one documented mapper per source, and always carry the provider tag alongside the value.
2. Assuming “expensive ⇒ point-in-time” (first principle: PIT is a product feature, not a price point). Default Compustat is restated; the PIT edition is bought separately. Even Bloomberg needs its as-of flag switched on - the live snapshot has already been restated.
3. The Yahoo trap (first principle: an undocumented endpoint has no contract). quoteSummary gives latest-restated fundamentals with no vintage, can change shape without notice, and is explicitly not licensed for redistribution. Fine for “what’s the P/E today,” fatal for “what did the screen see in 2015.”
4. Fiscal-period alignment (first principle: fiscal ≠ calendar). A vendor may or may not snap a June-FYE retailer to a calendar year; two vendors that disagree about this will disagree about which period a value belongs to - a source of spurious “surprises.”
5. Vendor lock-in cost (first principle: switching costs are real). Once every downstream number is keyed to a vendor’s field names and fiscal conventions, changing providers invalidates the entire history silently. Mitigate with a canonical schema and a provider-provenance column from day one.
6. Restated-by-default ≠ wrong - but unfalsifiable. Restated Compustat is better for describing what happened; it is unusable for asking what was knowable. Use the restated file for description and a PIT file for backtests, and never confuse the two.

---

5. References

• WRDS - Compustat and Compustat Point-in-Time documentation and the Compustat User’s Guide. The canonical statement of the restated-vs-PIT distinction; cross-referenced from the Corpus’s data-sources-and-corporate-data section.
• Sharadar (Nasdaq Data Link) dataset documentation
• Kenneth R. French, Data Library
• Fama & French, “The Cross-Section of Expected Stock Returns” (JF, 1992) and “A Five-Factor Asset Pricing Model” (JFE, 2015)
• Fridson & Alvarez, Financial Statement Analysis: A Practitioner’s Guide

---

6. Connected Graph Bridges

• Back: 02 · SEC EDGAR & XBRL · Index Hub
• Forward: 04 · Insider & Ownership · 05 · Failure Modes
• Application: Quantitative Fundamental Investing (the screens these feeds power) · 06 · Advanced Extensions (Sharadar 20-yr panel for the company tracker)