Produced by Andrei S. Gonçalves and Greg Leonard on 11/21/2022; extended through FF_YEAR 2025 on 08/07/2026

This document provides details for the data in the .csv files "FM_Port_(nport=10)" and "Firm_Level_FM"

If you use the data in these files, please cite the original source:

Gonçalves, A. S., Leonard, G. B., 2023. "The Fundamental-to-Market Ratio and the Value Premium Decline". Journal of Financial Economics 147, 382-405.

########## FM_Port_(nport=10).csv

*** This file contains value-weighted returns for the FM decile portfolios studied in Gonçalves and Leonard (2023)
*** Portfolio returns span the period from July/1973 to June/2026
*** The columns are:
    
    A) YEAR: The year associated with the given portfolio return.
    B) MONTH: The month associated with the given portfolio return.
    C) RANK: The rank (or decile) associated with the given portfolio return (1=decile1; ... ; 10=decile10; 999=10-1 spread).
    D) N: The number of stocks in the given portfolio for the given month.
    E) RET: The portfolio holding period return (in decimal points).

########## Firm_Level_FM.csv

*** This file contains the firm-year FM estimates in Gonçalves and Leonard (2023)
*** The columns are:
    
    A) PERMNO: The CRSP PERMNO identifier for the given security.
    B) FF_YEAR: the year associated with the given FM measure. For example, FF_YEAR=2000 implies FM is publicly available as of June/2000 according to the paper's timing convention. 
    C) FM: The fundamental-to-market measure in Gonçalves and Leonard (2023).

*** FF_YEAR spans the period from 1973 to 2025, which can be used to construct portfolio returns from July/1973 to June/2026
*** Comments:
    The FM variable is not winsorized, but we do make an adjustment for outliers. Footnote 13 of the paper describes our procedure: 
	After estimating FE, we deal with outliers by bounding FE at (1/100)×max(ME, BE) from below and at 100×min(ME, BE) from above. This approach 
	to deal with outliers is analogous to bounding ME/BE at 1/100 from below and at 100 from above. Moreover, bounding FE directly (instead of 
	winsorizing FE/ME and BE/FE separately) assures that the identity in Equation 11 remains valid even at the bounded points.
    Since FM is right skewed, we encourage you to take the log of FM (or some comparable normalization) if you include it in any regression.

########## About this extended vintage

*** The estimates above were produced by re-running the original code on a WRDS pull covering CRSP through 
    June/2026. Three things differ from the 11/21/2022 release, none of them a change to the method:

    1) CRSP now distributes the monthly stock file in the CIZ format (crsp.msf_v2). Share and exchange codes are
       rebuilt from the new security-classification fields, and the cumulative price/share adjustment factors are
       rebuilt from the event-level distribution factors, since CFACPR/CFACSHR are no longer distributed.
    2) In the CIZ format the delisting return is already folded into the monthly return, so it is no longer spliced
       in by hand. This makes delisting months usable that were not before.
    3) In the Legacy (FIZ) format, monthly returns are month to month holding period returns with dividends reinvested 
       at month-end. In the new Flat File Format (CIZ), monthly returns are compounded daily returns with dividends 
       reinvested on their ex-dates.

*** Over the firm-years present in both vintages, the FM estimates here correlate with the 11/21/2022 release at
    0.999 (Pearson) and 0.999 (Spearman), and 94% of firm-years land in the same decile.

*** Decile returns over the overlapping months (July/1973-June/2019) correlate with the released ones at 0.96-0.998,
    depending on the decile. Almost all of that gap is the FM measure moving a few stocks across a breakpoint rather
    than any difference in how the portfolios are formed.
