BRAIN Sentiment - Data Quality Report

1. Executive Summary

Completeness

B+

Accuracy

A

Timeliness

A-

Stability

B

Overall

B+

Data Quality Assessment

✅ SUITABLE FOR PRODUCTION

✅ Strengths

100% coverage for earnings call and 10-K sentiment
News coverage improved from 64% to 93% over time
Low outlier rates (<2% across all years)
Stable signal distribution (mean near zero)
8+ years of consistent daily data
Point-in-time data with no backfill bias

⚠️ Limitations

Early years (2017-2018) have lower news coverage (~65-70%)
Signal volatility decreased over time (regime change)
Small-cap stocks may have sparse news data
2020 shows elevated signal std (COVID volatility)
Universe composition changes ~1% annually

Recommended Use Cases

✓ Alpha Research
Signal quality supports factor research

✓ Risk Models
Consistent coverage for sentiment risk

⚠ Small-Cap
Use caution for micro-cap analysis

2. Dataset Overview & Lineage

7.1M

Total Records

5,076

Unique Tickers

2,136

Trading Days

8+ yrs

Time Span

Data Source

VendorBRAIN (braincompany.co)

Dataset NameBrain Combined Sentiment (BCS)

Coverage Start2017-01-02

Coverage End2025-03-10

FrequencyDaily (EOD)

DeliveryFTP / AWS S3

Data Lineage

News SourcesMultiple news aggregators

EC TranscriptsEarnings call transcripts

SEC Filings10-K full documents

NLP ProcessingBRAIN proprietary models

AggregationEqual-weight average

Update CadenceDaily by market close

Data Flow

Raw Sources (News, Transcripts, Filings) → NLP Processing (Sentiment extraction) → Normalization (Cross-sectional z-score) → Aggregation (Combined signal) → Delivery (Daily files)

3. Universe Definition & Coverage

85.9%

Avg News Coverage

100%

EC Coverage

100%

10-K Coverage

0.8%

Annual Churn

Coverage by Year

Year	Tickers	News Coverage	EC Coverage	10-K Coverage	Status
2017	3,494	64.2%	100%	100%	✗ Low
2018	3,503	70.8%	100%	100%	⚠ Fair
2019	3,460	82.6%	100%	100%	⚠ Fair
2020	3,503	92.7%	100%	100%	✓ Good
2021	3,628	93.5%	100%	100%	✓ Good
2022	3,776	92.2%	100%	100%	✓ Good
2023	3,673	92.3%	100%	100%	✓ Good
2024	3,521	91.8%	100%	100%	✓ Good
2025	3,121	92.6%	100%	100%	✓ Good

Key Finding: News coverage improved significantly from 64% (2017) to 93% (2024), a +28pp improvement. EC and 10-K coverage remain at 100% throughout.

4. Completeness & Missing Data Analysis

14.8%

Avg Missing News

42.0%

Max Missing News

0%

Missing EC

0%

Missing 10-K

Missing Data Patterns

Imputation Policy

Field	Missing Handling	Max Gap	Notes
NEWS_SENTIMENT	Forward fill	90 days	Excluded after 90d gap
EC_SENTIMENT	Carry forward	Until next EC	~90 day refresh cycle
CF_10K_SENTIMENT	Carry forward	Until next filing	Annual refresh
COMBINED	Calculated	N/A	Equal-weight average

Note: Early years (2017-2018) have higher missing rates for news (~35-40%). Consider excluding or weighting down this period for sensitive analyses.

5. Accuracy & Validation Checks

0.83%

Avg Outlier Rate

0.14%

Min Outlier Rate

1.60%

Max Outlier Rate

Pass

Range Check

Validation Checks

Check	Criteria	Result	Status
Range Check	Signal in [-1, 1]	All values within range	✓ Pass
Null Check	No unexpected nulls	Only expected missing (news)	✓ Pass
Duplicate Check	No duplicate ticker-date pairs	No duplicates found	✓ Pass
Date Continuity	No gaps in trading days	All trading days present	✓ Pass
Outlier Check	<2% outliers per year	Max 1.6% (2017)	✓ Pass
Distribution Check	Mean near zero	Mean: -0.0005	✓ Pass

Summary: All validation checks pass. Outlier rates are consistently below 2% and declining over time (1.6% in 2017 → 0.14% in 2025).

6. Timeliness & Latency

T+1

Availability Lag

EOD

Timestamp

None

Revisions

Daily

Update Freq

Latency Specifications

Component	Typical Latency	Max Latency	Notes
News Sentiment	T+1 (EOD)	T+1	30-day rolling aggregation
EC Sentiment	T+1 after call	T+2	Processed after transcript available
10-K Sentiment	T+1 after filing	T+3	Full document processing
Combined Signal	T+1	T+1	Available by market open

Timestamp Reliability

✓ Point-in-Time: All data is timestamped as of EOD on the observation date. No backfilling or revisions occur. Signal is available T+1 for trading at next open.

7. Stability & Structural Breaks

16.1%

Std CV

0.131

Avg Std

-0.34

5th Pctl

0.21

95th Pctl

Signal Distribution Over Time

Signal Volatility (Stability Check)

Detected Regime Changes

Period	Observation	Impact	Recommendation
2017-2018	Lower news coverage (64-71%)	Sparse signal	Weight down or exclude
Mar 2020	Elevated volatility (COVID)	Signal std spike	Expected, no action needed
2020-2021	Signal std declining trend	Regime shift	Monitor for continued decline
2023-2025	Stable, lower volatility	New steady state	Adjust models if needed

Note: Signal standard deviation has declined from ~0.16 (2017) to ~0.10 (2024-2025). This may indicate reduced cross-sectional dispersion in sentiment or improved data quality.

8. Bias & Distortion Analysis

Bias Assessment

Bias Type	Present?	Severity	Mitigation
Survivorship Bias	✓ No	None	Point-in-time data includes delisted securities
Look-Ahead Bias	✓ No	None	T+1 lag enforced, trade at next open
Backfill Bias	✓ No	None	No historical revisions
Selection Bias	✓ No	None	Universe defined independently
Coverage Bias	⚠ Yes	Low-Med	Large caps have better news coverage
Temporal Bias	⚠ Yes	Low	Early years have lower coverage

Signal Mean by Year (Bias Check)

Coverage Bias Detail: Large-cap stocks typically have 95%+ news coverage while small-caps may have 60-70%. Consider filtering universe by market cap or applying coverage-weighted analysis.

9. Corporate Actions & Event Handling

Corporate Action Treatment

Event Type	Handling	Historical Consistency
Ticker Changes	COMPOSITE_FIGI maintains continuity	✓ Consistent
Mergers/Acquisitions	Acquirer inherits target sentiment history	✓ Consistent
Spin-offs	New FIGI assigned, fresh history	✓ Consistent
Delistings	Data ends at delist date	✓ Consistent
IPOs	Data begins at IPO date	✓ Consistent
Stock Splits	No impact (sentiment, not price)	✓ N/A

Earnings Call Handling

TimingSentiment updated day after call

Carry ForwardPrevious EC sentiment until next call

Missing CallsRare (<1%), uses prior value

Summary: Corporate actions are handled consistently using COMPOSITE_FIGI as the primary identifier, ensuring continuity across ticker changes and corporate events.

10. Field-Level Documentation & Metadata

Data Dictionary

Field	Type	Range	Description	Known Limitations
DATE	Date	2017-01-02 to present	Trading day (EOD timestamp)	None
COMPOSITE_FIGI	String	BBG format	Bloomberg FIGI identifier	Some legacy tickers may lack FIGI
NEWS_SENTIMENT	Float	[-1, 1]	30-day news sentiment score	Sparse for small-caps
NEWS_VOLUME	Int	[0, ∞)	Article count in lookback	None
EC_LAST_CALL_SENTIMENT	Float	[-1, 1]	Most recent EC sentiment	~90 day stale for some
CF_LAST_10K_SENTIMENT	Float	[-1, 1]	Most recent 10-K sentiment	Annual update only
COMBINED_SENTIMENT	Float	Normalized	Equal-weight average	Dominated by news for active names
NEWS_DAYS	Int	30	News aggregation window	Fixed parameter
EC_DAYS	Int	[0, ~120]	Days since last EC	None
CF_DAYS	Int	[0, ~400]	Days since last 10-K	None

11. Controls, QA Process & Monitoring

Quality Assurance Checks

Check	Frequency	Owner	Action on Failure
File delivery	Daily	Data Ops	Alert, investigate vendor
Record count	Daily	Automated	Alert if >5% deviation
Schema validation	Daily	Automated	Reject file, alert
Range checks	Daily	Automated	Flag outliers for review
Coverage check	Weekly	Data Ops	Investigate drops >5%
Distribution check	Monthly	Quant Team	Review regime changes

Incident Handling

Process: Data issues are logged, investigated within 24h, and communicated to downstream users. Historical corrections are applied as new files (not in-place updates).

12. Known Issues & Limitations

Active Issues

Issue	Severity	Affected Period	Workaround
Low news coverage in 2017-2018	Medium	2017-01 to 2018-12	Weight down or exclude period
Signal std regime change	Low	2020 onwards	Re-normalize if needed
Sparse small-cap coverage	Low	Ongoing	Filter by market cap or news volume

Resolved Issues

Issue	Period	Resolution	Date Fixed
Coverage gap expansion	2017-2019	News sources expanded	2020-01
High outlier rates	2017	Improved NLP models	2018-06

Known Constraints

EC sentiment may be stale (up to 120 days between calls)
10-K sentiment updates annually only
Combined signal dominated by news for high-volume names
International stocks not covered (US equities only)

13. Compliance & Licensing Considerations

Data Rights

Aspect	Status	Notes
Internal Use	✓ Permitted	Research, trading, risk
Model Development	✓ Permitted	Alpha signals, factors
Redistribution	✗ Not Permitted	Requires separate license
Client Reports	✓ Permitted	Aggregated/derived only
Regulatory Filing	✓ Permitted	Audit trail available

Regulatory Considerations

MNPI: Sentiment is derived from public sources only (news, public filings, public earnings calls). No MNPI concerns identified.

Audit Trail: Full lineage maintained. Historical files retained for 7 years per regulatory requirements.

14. Appendices

A. Monthly Coverage Patterns

B. Yearly Ticker Counts

C. Statistical Summary

Metric	Value
Total Records	7,116,723
Trading Days	2,136
Unique Tickers (Max)	3,776
Date Range	2017-01-02 to 2025-03-10
Average News Coverage	85.9%
Signal Mean (Overall)	-0.0005
Signal Std (Overall)	0.1313
Average Outlier Rate	0.83%

D. Methodology Notes

Signal Construction: COMBINED_SENTIMENT is calculated as the equal-weighted average of NEWS_SENTIMENT, EC_LAST_CALL_SENTIMENT, and CF_LAST_10K_SENTIMENT after cross-sectional z-score normalization within each date. Outliers are winsorized at the 1st and 99th percentiles.