Do You Need a Data Vendor or a Research-Dataset Provider?

Do You Need a Data Vendor or a Research-Dataset Provider?

Most stalled backtests are not short of price data. They are short of structure. Here is the two-layer model that explains why.

Two Layers, Not One

When researchers compare data providers, they usually compare within one layer: which price API is deepest, which fundamentals feed is cheapest, which terminal is broadest. That framing hides a more useful distinction. There is a raw data layer, made of prices, quotes, fundamentals, and volatility surfaces, and there is a research-dataset layer, made of point-in-time universes, dated corporate events, and regime classifications that are already structured for a backtest. They are different products solving different problems.

A data vendor sells the first layer. A research-dataset provider sells the second. Confusing the two is why a researcher can pay for excellent price data and still be unable to run a valid backtest.

What a Data Vendor Gives You

Data vendors deliver the building blocks. Polygon.io (Massive), Databento, FinancialModelingPrep, and the enterprise terminals all live here, and they do this job well. They give you prices, fundamentals, and surfaces, at varying depth, cost, and coverage. Our roundup of the best market data APIs compares them, and the right one depends on your asset classes and budget.

What a data vendor does not give you is structure aligned to a point-in-time research process. The data describes the market as it is or was, in raw form, and the work of making it reproducible and backtest-ready is left to you.

What a Research-Dataset Provider Gives You

A research-dataset provider sells the second layer: data that has already been processed into strategy-ready inputs. Instead of raw prices, you get point-in-time universe membership, size aligned to each trading day, and corporate events parsed from filings and stamped with the date they became public. The discipline behind all of it is point-in-time correctness, explained in our explainer on point-in-time market data.

This is the layer that prevents lookahead bias and universe leakage, the structural errors that quietly invalidate backtests, and it is exactly what data vendors leave out by design.

Head-to-Head

Why Most Stacks Need Both

The two layers are complementary, not competing. You need a data vendor for prices and fundamentals, and a research-dataset provider for the structure that makes those inputs usable in a reproducible backtest. A worked example makes it concrete: to backtest a strategy on small caps after dilutive offerings, you need price data from a vendor and a dated dilution-event feed plus point-in-time size from a research-dataset provider. Each handles a different part of the data sources a systematic strategy depends on.

Alphanume operates on the second layer. The historical market cap dataset delivers point-in-time size, the dilution events feed turns SEC filings into machine-readable events, and the optionable tickers dataset provides point-in-time options eligibility. These sit on top of whatever data vendor you choose.

How to Decide What You Need

Ask what is actually blocking you. If you cannot get prices, fundamentals, or surfaces at the depth and price you want, you need a data vendor, and the comparison is within that layer. If you have the raw data and your backtests still feel fragile, biased, or impossible to reproduce, you do not need another price feed. You need the research-dataset layer that supplies point-in-time universes and dated events.

Most serious workflows end up paying for both, because raw market data and strategy-ready structure are genuinely different products. Knowing which layer your problem lives in is what turns a stalled research project into a working one.