Appendix l — Data Card — TRACE (Transaction Reporting and Compliance Engine)

Provider & what it is. TRACE, operated by FINRA and distributed through WRDS, is the transaction-by-transaction record of U.S. corporate (and some other) bond trades: for each reported trade, the price, the par volume, the time, and side flags. Before TRACE, the corporate-bond market was a black box — trades happened over-the-counter with no public tape. TRACE is the microscope that made modern bond-market microstructure and liquidity research possible. The reveal-the-trick point is that TRACE comes in two forms with different sensitivity — a disseminated (public-facing, partly masked) feed and an academic/enhanced version with fields the public feed hides — and the academic version’s extra detail is exactly why the on-GMU-only rule bites hardest here.

Coverage

• Span. Corporate-bond transactions from July 2002 (phased in by bond size/rating through 2005), continuing to the present. The Enhanced/Academic TRACE historical file adds masked-field detail. [CHECK] exact phase-in dates and the academic-file lag (academic data is released with a multi-quarter delay).
• Frequency. Tick-level — one row per reported trade, time-stamped to the second (or finer in recent data).
• Universe. TRACE-eligible debt: corporate bonds, and in later expansions agency debt, securitized products, and Treasuries (a separate reporting regime). The core academic-finance use is corporate bonds.

Key identifiers

• CUSIP (9-character) — identifies the bond issue (the specific security, not the issuer). As with 13F, CUSIP is a licensed identifier: use it to merge, do not redistribute a master.
• bond_sym_id — TRACE’s own bond symbol.
• A trade is otherwise keyed by CUSIP + execution timestamp + a sequence/message field.
• To get issue characteristics (coupon, maturity, rating, covenants) you join CUSIP to Mergent FISD — see that data card; TRACE is trades, FISD is the bond’s biography.

Access path

Same WRDS/wrds pattern; credentials from .pgpass or env (CONVENTIONS §5). TRACE is enormous, so bound tightly by date and CUSIP.

import os, wrds, pandas as pd, pathlib

db = wrds.Connection(wrds_username=os.environ["WRDS_USERNAME"])   # secret from env

raw = pathlib.Path("data/raw/trace_subset_2019.parquet")         # git-ignored
if raw.exists():
    trades = pd.read_parquet(raw)
else:
    # Bound by date AND a CUSIP list -- never an unbounded scan of the trade tape.
    trades = db.raw_sql("""
        SELECT cusip_id, trd_exctn_dt, trd_exctn_tm, rptd_pr,
               entrd_vol_qt, rpt_side_cd, trc_st
        FROM   trace.trace_enhanced
        WHERE  trd_exctn_dt BETWEEN '2019-01-01' AND '2019-12-31'
          AND  cusip_id IN ('037833AK6','594918AC8')   -- illustrative
    """, date_cols=["trd_exctn_dt"])
    trades.to_parquet(raw)                                       # GMU infra only
db.close()

[CHECK] the exact WRDS library/table for Enhanced TRACE (trace.trace_enhanced vs. variants) and the standard cleaning steps — TRACE requires deleting cancellations/corrections/reversals before use (see Dick-Nielsen’s published cleaning procedure). Raw TRACE without that cleaning is not analysis-ready.

License & the “stays on GMU infrastructure” note (CONVENTIONS §5)

TRACE is licensed (FINRA, via WRDS), accessed read-only, and the academic/enhanced version is more sensitive than most WRDS data because it un-masks fields (large-trade volumes, dealer detail) that the public dissemination deliberately caps to protect dealers’ positions. So the §5 rule is non-negotiable here: query only on WRDS Cloud or GMU’s Hopper, cache to a git-ignored data/raw/, and never commit, email off campus, post publicly, or paste into a commercial web service. The un-masked academic fields in particular must not leave GMU infrastructure under any circumstances. Ship the recipe (query + log), not the data; CUSIP masters are not redistributable either.

Gotchas

• Cancellations, corrections, and reversals. A nontrivial share of raw messages are not clean executed trades — they cancel or correct an earlier message. Use the published cleaning procedure (drop trades flagged cancelled/corrected/reversed, handle agency double-counting) before computing anything. This is the single biggest TRACE mistake.
• Disseminated vs. academic volume masking. The disseminated feed caps reported volume above a threshold (e.g., trades over $5M IG / $1M HY shown as “5MM+”); the academic file un-masks it. Mixing the two, or assuming public volumes are exact, biases liquidity measures.
• Agency vs. principal trades & double counting. A single customer trade intermediated by a dealer can generate multiple TRACE reports; net them per the cleaning rules.
• Price is per $100 par, not a dollar price; convert with par.
• No quotes, only trades. TRACE is a trade tape; there is no continuous quote. Liquidity proxies (bid-ask, Amihud) must be estimated from trades.
• Academic-file release lag. The enhanced/academic file is published with a delay, so the most recent quarters are unavailable in academic form. [CHECK] the current lag.

“First 10 rows” schema sketch (illustrative — not real TRACE data)

cusip_id (str)	trd_exctn_dt (date)	trd_exctn_tm (time)	rptd_pr (float, /100 par)	entrd_vol_qt (float, par $)	rpt_side_cd (str)	trc_st (str)
037833AK6	2019-03-14	09:41:07	101.250	2000000	B	T
037833AK6	2019-03-14	10:02:55	101.310	5000000	S	T
037833AK6	2019-03-14	10:03:12	101.310	5000000	B	R
594918AC8	2019-03-14	13:18:44	98.875	250000	S	T
…	…	…	…	…	…	…

(Values invented to show shape and types: row 3 has trc_st='R' — a reversal of row 2’s trade — and must be removed in cleaning, the classic TRACE trap.)

Which chapter/lab/capstone uses it

• Week 7, Ch 7.2 — listed among the core WRDS sources (CRSP/Compustat/IBES/TRACE); the on-GMU-only sensitivity is emphasized.
• A natural data source for any fixed-income capstone (bond liquidity, the credit-spread cross-section), typically merged to Mergent FISD for issue characteristics — the kind of project Priya (insurer credit risk) or a fixed-income-minded student might pursue.
• Bond-market microstructure as an advanced extension of the equity-microstructure ideas in the anomalies/momentum thread.