Fama-French Data Library

The Fama-French Data Library publishes documented factor, portfolio, breakpoint, and research-return datasets for asset-pricing analysis.

The Fama-French Data Library is Kenneth French’s public collection of factor returns, research portfolios, breakpoints, and related asset-pricing datasets. Researchers and analysts use it to estimate factor exposures, reproduce academic tests, and compare portfolio returns with documented benchmarks. The files are research data, not investment recommendations or immutable market records.

Key Takeaways

  • Select a dataset by region, factor definition, frequency, currency, and portfolio construction rather than by name alone.
  • Read the linked details page before using a return series; similarly named U.S. and international factors can use different breakpoints and inputs.
  • Check units, date format, return convention, missing-value codes, and whether a file contains more than one table.
  • Library histories can change when source data and construction methods are revised.
  • Reproducible work should preserve the raw download, access date, documentation, transformations, and code version.

What the Library Contains

The exact catalog changes over time. Major groups include:

Dataset groupExamplesTypical analytical use
Research factorsMarket excess return, SMB, HML, profitability, investment, momentumFactor regression and return attribution
Sorted portfoliosPortfolios formed on size, book-to-market, profitability, investment, or prior returnsTesting cross-sectional patterns and model fit
Multi-dimensional sorts2 x 3, 5 x 5, and other combinationsExamining interactions between characteristics
BreakpointsMarket equity, book-to-market, profitability, investment, and prior-return cutoffsReproducing portfolio assignment rules
Industry portfoliosPortfolios grouped by industry classificationsIndustry controls, benchmarking, and research tests
International dataDeveloped, regional, and selected emerging-market factors and portfoliosAnalysis outside the U.S. dataset
Historical archivesPrior data cuts for selected U.S. factors and construction portfoliosMeasuring the effect of later data revisions

Availability, history, and frequency differ by dataset. The main catalog and each details page are the controlling references.

Reading the Standard Three-Factor File

A standard monthly three-factor dataset commonly includes:

  • a date field in (YYYYMM) form
  • (Mkt-RF), the market return minus the risk-free return
  • (SMB), the small-minus-big factor return
  • (HML), the high-minus-low book-to-market factor return
  • (RF), the risk-free return

The source documentation should be checked for units and conventions before calculation. Factor files commonly present returns as percentages, so a displayed value of 2.10 is normally converted to 0.0210 for software expecting decimal returns.

Hypothetical Data-Row Example

Consider this invented monthly row, which is not an observation from the library:

DateMkt-RFSMBHMLRF
2025012.10-0.401.200.35

If the values are percentages, the broad market return represented by the row is:

$$ R_M=(R_M-R_F)+R_F=2.10\%+0.35\%=2.45\% $$

For a regression package that expects decimal returns, the input values would be 0.0210, -0.0040, 0.0120, and 0.0035. Mixing percentage and decimal units creates coefficients and intercepts off by a factor of 100.

The Fama-French Three-Factor Model explains how these series enter a return regression.

How the U.S. Three Factors Are Constructed

The official details page describes the U.S. stock factors as follows:

  • The market factor is a value-weighted market return for the eligible U.S. CRSP universe minus the one-month Treasury bill return used by the library.
  • SMB and HML are constructed from six value-weighted portfolios formed on size and book-to-market.
  • SMB averages the three small portfolios and subtracts the average of the three big portfolios.
  • HML averages the two high-book-to-market portfolios and subtracts the average of the two low-book-to-market portfolios.
  • Portfolio formation uses specified market-equity and positive book-equity data and a July-to-June holding cycle.

These are construction rules, not plain-language labels. SMB is not simply the return of a small-cap index, and HML is not simply the return of a value-stock fund.

Choosing the Correct Dataset

Region and Currency

Do not substitute a U.S. factor file for a developed-market or regional analysis without a stated reason. International factor documentation should be checked for the return currency, regional breakpoints, and market universe.

Frequency

Match daily, weekly, or monthly factors to the asset-return frequency. Converting one frequency to another requires correct compounding and date alignment; repeating a monthly return across daily rows is not valid.

Factor Family

The three-factor, five-factor, and momentum datasets answer different model questions. Adding every available series can create multicollinearity and data-mining problems. Choose factors before inspecting the desired result and justify each inclusion.

Portfolio Type

Factor returns, sorted portfolio returns, and breakpoint files are different products. A factor is generally a return difference between constructed portfolios. A sorted portfolio is a funded return series. A breakpoint is a classification threshold, not a return.

Total Return vs. Ex-Dividend Return

Some portfolio datasets have versions excluding dividends. Confirm which version matches the research question and do not compare ex-dividend returns with total-return benchmarks without adjustment.

Reproducible Download Workflow

  1. Open the official Data Library catalog.
  2. Choose the exact region, model, frequency, and portfolio type.
  3. Save the raw TXT or CSV download without editing it.
  4. Save or link the dataset’s details page and record the access date.
  5. Inspect headers, footnotes, table boundaries, date coverage, units, and missing-value conventions.
  6. Parse the required table without silently treating explanatory lines or annual summaries as monthly observations.
  7. Convert return units once and document the transformation.
  8. Align dates with the test portfolio and investigate unmatched observations.
  9. Validate summary statistics and several rows against the raw file.
  10. Preserve the code, package versions, and a checksum or immutable copy of the raw download.

This process is more defensible than downloading a current file later and assuming it is identical to the one used in the original analysis.

Data Revisions and the 2025 Format Change

The library states that U.S. research returns can change when CRSP revises its source database and that the full return history is reconstructed during updates. It provides historical archives for selected monthly factor and construction-portfolio files so users can compare prior data cuts.

The library also documents a significant source-format transition. CRSP’s legacy Stock and Indexes format was discontinued after the December 2024 release, and U.S. research returns beginning with the January 2025 release use CRSP’s Flat File Format 2.0. The library notes that monthly return construction differs between the old and new formats, including the timing of dividend reinvestment.

This does not make either history unusable. It means a reproduced result should identify the data vintage, and an unexplained difference may arise from a source revision rather than an error in the regression code.

Missing Values and Parsing Controls

The library states that missing data can be represented by values such as -99.99 or -999. These are sentinels, not extreme returns. Convert them to missing values before calculating means, compounding returns, or estimating a model.

Other controls include:

  • verify that dates are unique within the selected table
  • confirm chronological ordering after parsing
  • check for gaps rather than automatically forward-filling returns
  • distinguish daily dates from monthly (YYYYMM) identifiers
  • ensure negative signs survive spreadsheet import
  • compare observation counts before and after merging datasets
  • document any winsorization, deletion, or imputation

Worked Research Example

Suppose an analyst wants to estimate whether a U.S. equity fund had market, size, and value exposure over a five-year monthly window. A defensible setup would:

  1. Obtain the fund’s monthly total returns and confirm whether they are net or gross of fees.
  2. Download the U.S. monthly three-factor file and retain the raw archive.
  3. Convert both return series to the same units.
  4. Subtract the library’s monthly (RF) from the fund return to create the dependent excess return.
  5. Join on month and review every dropped observation.
  6. Regress fund excess return on (Mkt-RF), (SMB), and (HML).
  7. Report coefficients with uncertainty, sample dates, observation count, diagnostics, and data vintage.
  8. Compare model-implied exposures with the fund’s holdings and stated mandate.

The regression output remains descriptive. A positive historical HML loading does not guarantee future value exposure or return.

Common Mistakes

  • Treating displayed percentages as decimals or converting them twice.
  • Subtracting (RF) from a return series that is already an excess return.
  • Joining a month-end portfolio return to a daily factor row.
  • Using the U.S. file for a foreign-currency portfolio without addressing region and currency differences.
  • Interpreting missing-value sentinels as actual losses.
  • Combining sections of a text file into one rectangular dataset.
  • Using a current revised history without recording that the original study used an earlier data cut.
  • Assuming a public research factor is directly investable without costs or constraints.
  • Treating a high regression fit as proof of causation or future performance.

Risks and Limitations

The library is an important research resource, but its series inherit limitations from source data, accounting definitions, portfolio formation rules, and historical revisions. Results can be sensitive to sample dates, frequency, test assets, factor selection, and statistical method.

Public availability does not eliminate licensing or redistribution questions for downstream products. Users should review the library’s notices and the terms governing any underlying data before redistributing files or building a commercial data service.

Authoritative Resources

  • Factor Models: Models that use common drivers to explain return or risk.
  • Factor Investing: Portfolio implementation that targets selected factor characteristics or exposures.
  • Backtesting: Historical testing that requires point-in-time data and controlled assumptions.
  • Survivorship Bias: Bias from excluding securities or funds that disappeared from the sample.
  • Regression Analysis: A common method for estimating factor loadings and alpha.

FAQs

Is the Fama-French Data Library free to access?

The website provides public downloads. Public access does not automatically grant unrestricted redistribution of every underlying dataset or permit every commercial use, so applicable notices and source-data terms should be reviewed.

Can historical Fama-French data change?

Yes. The library reconstructs histories as source data and methods are updated. Preserve the data vintage and use historical archives where reproducibility matters.

Are Fama-French factors investable indexes?

Not necessarily. Research factors are constructed return differences used for analysis. A fund or portfolio targeting similar characteristics faces funding, long-only or shorting constraints, fees, turnover, liquidity, and tracking differences.

This article provides general financial education and research-data guidance. It does not recommend a dataset, model, fund, factor allocation, security, or trading strategy.

Browse Investing