Contact

• Jianxin Ma
• jianxin.ma@warwick.ac.uk

Version

• All in Python
• Two pipeline variants live in this repository:
  • chars_ciz/ — current CIZ (CRSP CIZ / v2) pipeline. Use this.
  • chars_siz/ — legacy SIZ (CRSP SIZ / v1) pipeline. Kept for reference only.
• The SAS version is here EquityCharacteristicsSAS
• Extension to China A Share Market
• Extension to Factors and Portfolios in China Market

Academic Background

Empirical asset pricing research often needs firm-level equity characteristics and portfolio-level characteristic summaries. This repository provides a Python toolkit for constructing U.S. equity characteristics from WRDS inputs, with the current implementation centered on the CRSP CIZ data format.

Prerequisite

• Read the listed papers
• WRDS account with subscription to CRSP, Compustat and IBES.
• Python (with polars, pandas, numpy, tqdm, wrds, duckdb, pyarrow)

Files

• Characteristics list

Main Files (chars_ciz/)

• download_data.py — pull raw CRSP CIZ monthly/daily, Compustat funda/fundq, IBES, FF factors from WRDS (via DuckDB) into data/raw/
• functions.py — shared helpers: ttm, ttm4, ttm12, chars_std, industry classifications (ffi49), imputation and standardization utilities, INPUT_PATH / OUTPUT_PATH constants
• accounting.py — builds annual + quarterly accounting characteristics and merges with monthly CRSP (crsp_mom) for monthly-grain chars (me, turn, dolvol, dy, mom*, seas1a, indmom, …). Outputs chars_a_accounting.parquet, chars_q_accounting.parquet
• rolling_chars.py — rolling daily-window characteristics from CRSP daily: beta, baspread, ill, maxret, rvar_capm, rvar_ff3, rvar_mean, std_dolvol, std_turn, zerotrade. Outputs rolling_chars.parquet
• sue.py — unexpected quarterly earnings (SUE)
• abr.py — cumulative abnormal returns around earnings announcement dates
• myre.py — revisions in analysts' earnings forecasts (uses IBES)
• merge_chars.py — merges accounting + rolling + satellite characteristics (sue, abr, myre) with CRSP backfill. Outputs chars_a_raw.parquet, chars_q_raw.parquet
• impute_rank_output.py — reconciles annual/quarterly accounting variables by most-recent datadate, computes ffi49, lags returns one period, and writes the four final outputs (raw / imputed × no-rank / rank)
• iclink_ciz.sas — IBES ↔ CRSP CIZ link table (run on WRDS SAS Studio); output saved to data/raw/iclink_ciz.csv

Documents

• documents/chars_summary.csv — current characteristic acronym, description, author, year, category

How to use (CIZ pipeline)

All commands are run from chars_ciz/. Paths are configured in functions.py (INPUT_PATH, OUTPUT_PATH).

1. python download_data.py — pull raw WRDS tables into data/raw/ (also run iclink_ciz.sas on WRDS to produce iclink_ciz.csv)
2. python accounting.py — build annual/quarterly accounting + monthly chars
3. python rolling_chars.py — daily-window rolling chars
4. python sue.py, python abr.py, python myre.py — satellite chars (can run in parallel; myre.py requires IBES + the iclink table)
5. python merge_chars.py — merge everything into chars_a_raw.parquet and chars_q_raw.parquet
6. python impute_rank_output.py — produce the four final outputs

Outputs

Data

The stock universe is the top three U.S. exchanges (NYSE / AMEX / NASDAQ), filtered to common equity via CRSP CIZ flags (sharetype='NS', securitytype='EQTY', securitysubtype='COM', usincflg='Y', issuertype∈{ACOR,CORP}, conditionaltype='RW', tradingstatusflg='A'). The date range follows the available WRDS coverage.

Returns are shifted one period forward so that characteristics at time $t$ predict the return at $t+1$ (i.e. $ret_{t+1}$ is aligned with $chars_t$).

The four final files (all parquet) are:

1. chars_raw_no_impute.parquet — raw characteristic levels, missing values preserved
2. chars_raw_imputed.parquet — same as above with industry-median / industry-mean imputation
3. chars_rank_no_impute.parquet — cross-sectional rank-standardized characteristics (no imputation)
4. chars_rank_imputed.parquet — cross-sectional rank-standardized characteristics (imputed)

Information Variables

• stock identifier: gvkey, permno, ticker, conm, comnam
• time: datadate (accounting), date (return date)
• industry: sic, ffi49
• price / size: prc, shrout, me, log_me, lag_me
• return: ret (delisting-adjusted via CRSP CIZ mthret)

Method

Equity Characteristics

The equity-characteristic definitions follow the empirical anomaly and replication literature, especially Green, Hand, and Zhang and Hou, Xue, and Zhang. The current implementation documents 99 U.S. equity characteristics in documents/chars_summary.csv and provides audited formula notes under documents/formula_docs/.

Portfolio Characteristics

Portfolio characteristics can be constructed as equal-weighted or value-weighted averages of firm-level characteristics for equities in each portfolio.

Common portfolio applications include:

• Characteristics-sorted Portfolio, see the listed papers and also Deep Learning in Characteristics-Sorted Factor Models
• DGTW Benchmark, see DGTW 1997 JF
• Industry portfolio

Reference

Papers

• Dissecting Anomalies with a Five-Factor Model by Fama and French 2015 RFS

  • Define the characteristics of a portfolio as the value-weight averages (market-cap weights) of the variables for the firms in the portfolio
  • French's Data Library

• The Characteristics that Provide Independent Information about Average U.S. Monthly Stock Returns by Green Hand Zhang 2017 RFS

  • sas code from Green's website

• Replicating Anomalies by Hou Xue Zhang 2018 RFS

  • Anomaly Portfolios by Zhang's website

Codes

• Legacy SAS calculations mainly refer to SAS code by Green Hand Zhang.
• Portfolio characteristics mainly refer to WRDS Financial Ratios Suite and Variable Definition

All comments are welcome.