Florian Herzog 1993658fb2 Add SEC fund prospectus -> RDF triple dataset pipeline

Builds a relationship-rich finance dataset for text-to-RDF-triple extraction
from SEC fund disclosures, the dataset for the thesis 'Magical RDF Triples and
how to synthetize them'.

- build_rdf_dataset.py: gold (N-CEN graphs), fetch (EDGAR prospectus prose,
  all books per trust), samples (per-fund segmentation, marker + plain
  serializations), split (trust-level 80/10/10, no leakage)
- score_baseline.py: no-model string-match baseline + strong-model scorer
- dataset_description.{tex,pdf}: scientific description of the dataset
- data/rdf_poc/gold_graphs.jsonl: structured gold knowledge graph (2025Q3)
- Large prose/sample files and raw SEC downloads are gitignored (reproducible)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

2026-06-03 10:31:35 +02:00

18 KiB

Raw Blame History

SEC Reference Data Fields vs. openfunds Data Model

1. Overview of SEC Structured Data Sources

The SEC provides four distinct structured data sources that contain reference data for US-registered funds. Each covers different aspects:

Source	Form	Content	Format	Granularity
Series/Class CSV	—	Identity & identifiers	CSV/XML	Trust → Series → Class
XBRL Risk/Return	N-1A (485BPOS, 497K)	Prospectus-derived structured data	XBRL → flat files	Series & Class level
N-PORT Data Sets	NPORT-P	Portfolio holdings & fund financials	XML → flat files	Series & Holding level
Submissions API	—	Filing history metadata	JSON	Entity (CIK) level

2. Complete Field Inventory by SEC Source

2.1 Series/Class Reference CSV

This is the identity backbone — maps the hierarchy of trust → fund → share class.

Field	Description	openfunds Equivalent
`Reporting File Number`	811-XXXXX Investment Co. Act number	— (no direct equivalent)
`CIK Number`	10-digit SEC entity identifier	— (SEC-specific)
`Entity Name`	Trust/investment company name	OFST005010 Umbrella Name
`Entity Org Type`	Organization type code	OFST160100 Legal Form
`Series ID`	S###### fund series identifier	— (SEC-specific)
`Series Name`	Fund name	OFST010110 Legal Fund Name Only
`Class ID`	C###### share class identifier	— (SEC-specific)
`Class Name`	Share class name (e.g. "Admiral Shares")	OFST020060 Full Share Class Name
`Class Ticker`	Exchange ticker symbol	OFST020020 Bloomberg Code (partial)
`Address_1, Address_2, City, State, Zip Code`	Registrant address	—

Coverage: ~15,000+ investment company trusts, ~50,000+ series, ~100,000+ classes.

2.2 XBRL Risk/Return Summary (from Prospectus — the richest source)

This dataset is extracted from prospectus XBRL filings and is the closest to what openfunds covers. It contains the structured data that prospectuses specify.

A. Fund Identity & Structure

XBRL Element	Description	Data Type	openfunds Equivalent
`RiskReturnHeading`	Prospectus section heading	Text	—
`ObjectiveHeading`	Heading of objectives section	Text	—
`ObjectivePrimaryTextBlock`	Investment objective narrative	Text Block	OFST010300 Investment Objective
`ObjectiveSecondaryTextBlock`	Additional objective detail	Text Block	OFST010300 Investment Objective
`StrategyHeading`	Heading of strategy section	Text	—
`StrategyNarrativeTextBlock`	Principal investment strategies	Text Block	— (no single openfunds equivalent)

B. Fee & Expense Data (Shareholder Fees — paid directly by investor)

XBRL Element	Description	Data Type	openfunds Equivalent
`MaximumSalesChargeImposedOnPurchasesOverOfferingPrice`	Front-end load	Ratio	OFST451320 Max Subscription Fee In Favour Of Distributor
`MaximumDeferredSalesChargeOverOfferingPrice`	Back-end load (CDSC)	Ratio	OFST451391 Contingent Deferred Sales Charge Exit Fee
`MaximumDeferredSalesChargeOverOther`	CDSC on other basis	Ratio	OFST451392 Contingent Deferred Sales Charge Upfront Fee
`MaximumSalesChargeOnReinvestedDividendsAndDistributions`	Load on reinvested dividends	Ratio	—
`RedemptionFeeOverRedemption`	Redemption fee (% of amount)	Ratio	OFST451440 Max Redemption Fee In Favour Of Fund
`RedemptionFee`	Redemption fee (flat $)	Monetary	OFST451439 Min Redemption Fee In Favour Of Fund
`ExchangeFeeOverRedemption`	Exchange fee (% of amount)	Ratio	—
`ExchangeFee`	Exchange fee (flat $)	Monetary	—
`MaximumAccountFeeOverAssets`	Account maintenance fee (%)	Ratio	—
`MaximumAccountFee`	Account maintenance fee ($)	Monetary	—
`MaximumCumulativeSalesChargeOverOfferingPrice`	Cumulative max sales charge	Ratio	—

C. Annual Fund Operating Expenses (ongoing costs deducted from fund assets)

XBRL Element	Description	Data Type	openfunds Equivalent
`ManagementFeesOverAssets`	Management fee	Ratio	OFST452010 Management Fee Maximum
`DistributionAndService12b1FeesOverAssets`	12b-1 distribution fee	Ratio	OFST454165 Distribution Fee Maximum
`Component1OtherExpensesOverAssets`	Other expense component 1	Ratio	—
`Component2OtherExpensesOverAssets`	Other expense component 2	Ratio	—
`Component3OtherExpensesOverAssets`	Other expense component 3	Ratio	—
`OtherExpensesOverAssets`	Total other expenses	Ratio	—
`AcquiredFundFeesAndExpensesOverAssets`	Acquired fund fees (fund-of-funds)	Ratio	—
`ExpensesOverAssets`	Total Annual Fund Operating Expenses	Ratio	OFST452100 TER Excluding Performance Fee
`FeeWaiverOrReimbursementOverAssets`	Fee waiver/reimbursement	Ratio	—
`NetExpensesOverAssets`	Net expenses after waivers	Ratio	OFST452200 Ongoing Charges

D. Expense Example (hypothetical cost projections)

XBRL Element	Description	Data Type	openfunds Equivalent
`ExpenseExampleYear01`	Cost for $10K after 1 year	Monetary	—
`ExpenseExampleYear03`	Cost for $10K after 3 years	Monetary	—
`ExpenseExampleYear05`	Cost for $10K after 5 years	Monetary	—
`ExpenseExampleYear10`	Cost for $10K after 10 years	Monetary	—
`ExpenseExampleNoRedemptionYear01`	Cost if no redemption, 1 year	Monetary	—
`ExpenseExampleNoRedemptionYear03`	Cost if no redemption, 3 years	Monetary	—
`ExpenseExampleNoRedemptionYear05`	Cost if no redemption, 5 years	Monetary	—
`ExpenseExampleNoRedemptionYear10`	Cost if no redemption, 10 years	Monetary	—

E. Performance Data

XBRL Element	Description	Data Type	openfunds Equivalent
`AnnualReturn[YYYY]`	Annual return for calendar year	Ratio	OFDY025000-range (Performance data)
`BarChartHighestQuarterlyReturn`	Best quarter return	Ratio	—
`BarChartLowestQuarterlyReturn`	Worst quarter return	Ratio	—
`BarChartHighestQuarterlyReturnDate`	Date of best quarter	Date	—
`BarChartLowestQuarterlyReturnDate`	Date of worst quarter	Date	—
`AverageAnnualReturnYear01`	Average annual return, 1 year	Ratio	OFDY025000-range
`AverageAnnualReturnYear05`	Average annual return, 5 years	Ratio	OFDY025000-range
`AverageAnnualReturnYear10`	Average annual return, 10 years	Ratio	OFDY025000-range
`AverageAnnualReturnSinceInception`	Return since inception	Ratio	—
`AverageAnnualReturnInceptionDate`	Inception date	Date	OFST020560 Share Class Launch Date
Performance dimensions: Before Taxes, After Taxes on Distributions, After Taxes on Distributions and Sales			—

F. Risk Disclosures

XBRL Element	Description	Data Type	openfunds Equivalent
`RiskHeading`	Risk section heading	Text	—
`RiskNarrativeTextBlock`	Principal risks narrative	Text Block	—
`RiskLoseMoney`	"You may lose money" statement	String	—
`RiskMoneyMarketFundMayImposeFeesOrSuspendSales`	MMF gate/fee risk	Boolean	—
`RiskMoneyMarketFundPriceFluctuates`	MMF NAV fluctuation risk	Boolean	—
`BarChartAndPerformanceTableHeading`	Performance section heading	Text	—
`PerformanceNarrativeTextBlock`	Performance context narrative	Text Block	—

G. Portfolio Turnover

XBRL Element	Description	Data Type	openfunds Equivalent
`PortfolioTurnoverHeading`	Section heading	Text	—
`PortfolioTurnoverTextBlock`	Turnover narrative	Text Block	—
`PortfolioTurnoverRate`	Turnover rate (%)	Ratio	OFRE000025-range (Fund Ratios)

2.3 N-PORT Data Sets (Portfolio Holdings — quarterly)

This provides dynamic portfolio data not typically in openfunds static fields.

A. Fund-Level Information (`FUND_REPORTED_INFO`)

Field	Description	openfunds Equivalent
`SERIES_NAME`	Fund name	OFST010110 Legal Fund Name Only
`SERIES_ID`	SEC series identifier	—
`SERIES_LEI`	LEI of the fund series	OFST010030 LEI Of Fund
`TOTAL_ASSETS`	Total assets (USD)	OFDY000010-range (AuM/TNA)
`TOTAL_LIABILITIES`	Total liabilities	—
`NET_ASSETS`	Net assets (TNA)	OFDY000010-range
`SALES_FLOW_MON1/2/3`	Monthly inflows	—
`REDEMPTION_FLOW_MON1/2/3`	Monthly outflows	—
Credit spread sensitivities (3m,1y,5y,10y,30y)	Risk measures	—

B. Interest Rate Risk (`INTEREST_RATE_RISK`)

Field	Description	openfunds Equivalent
`CURRENCY_CODE`	Currency of exposure	OFST010410 Fund Currency
`INTRST_RATE_CHANGE_*_DV01`	DV01 by maturity bucket	—
`INTRST_RATE_CHANGE_*_DV100`	Impact of 100bp shift	—

C. Monthly Returns (`MONTHLY_TOTAL_RETURN`)

Field	Description	openfunds Equivalent
`CLASS_ID`	Share class identifier	—
`MONTHLY_TOTAL_RETURN1/2/3`	Monthly returns per class	OFDY025000-range

D. Portfolio Holdings (`FUND_REPORTED_HOLDING`)

Field	Description	openfunds Equivalent
`ISSUER_NAME`	Holding issuer name	OFPH-range (Portfolio Holdings)
`ISSUER_LEI`	LEI of issuer	OFPH-range
`ISSUER_TITLE`	Security title/description	OFPH-range
`ISSUER_CUSIP`	CUSIP of holding	OFPH-range
`BALANCE`	Position size	OFPH-range
`UNIT`	Shares/principal/other	OFPH-range
`CURRENCY_CODE`	Currency of holding	OFPH-range
`CURRENCY_VALUE`	Value in reporting currency	OFPH-range
`EXCHANGE_RATE`	FX rate applied	—
`PERCENTAGE`	% of net assets	OFPH-range
`PAYOFF_PROFILE`	Long/Short/N/A	OFPH-range
`ASSET_CAT`	Asset type classification	OFST350000 MiFID Securities Classification (concept)
`ISSUER_TYPE`	Corporate/Government/etc.	—
`INVESTMENT_COUNTRY`	Country of issuer (ISO)	OFPH-range
`IS_RESTRICTED_SECURITY`	Restricted security flag	—
`FAIR_VALUE_LEVEL`	Fair value hierarchy (1/2/3)	—

E. Holding Identifiers (`IDENTIFIERS`)

Field	Description	openfunds Equivalent
`IDENTIFIER_ISIN`	ISIN	OFST020000 ISIN
`IDENTIFIER_TICKER`	Ticker	—
`OTHER_IDENTIFIER`	SEDOL, etc.	OFST020040 SEDOL

3. Mapping: What openfunds Fields CAN Be Found in SEC Data?

Fully Available (structured, machine-readable)

openfunds Category	openfunds OF-ID	openfunds Field	SEC Source	SEC Field
Key Fact: Company	OFST001000	Fund Group Name	Series/Class CSV	Entity Name
Key Fact: Umbrella	OFST005010	Umbrella Name	Series/Class CSV	Entity Name
Key Fact: Fund	OFST010030	LEI Of Fund	N-PORT	SERIES_LEI
	OFST010110	Legal Fund Name Only	Series/Class CSV + N-PORT	Series Name
	OFST010300	Investment Objective	XBRL R/R	ObjectivePrimaryTextBlock
	OFST010410	Fund Currency	N-PORT	CURRENCY_CODE (inferred)
Key Fact: Share Class	OFST020000	ISIN	N-PORT Holdings	IDENTIFIER_ISIN
	OFST020005	CUSIP	N-PORT Holdings	ISSUER_CUSIP
	OFST020040	SEDOL	N-PORT Holdings	OTHER_IDENTIFIER
	OFST020060	Full Share Class Name	Series/Class CSV	Class Name
Classification	OFST350000	Securities Classification	N-PORT	ASSET_CAT
Fees	OFST451320	Max Subscription Fee (Distributor)	XBRL R/R	MaximumSalesChargeImposedOnPurchasesOverOfferingPrice
	OFST451391	CDSC Exit Fee	XBRL R/R	MaximumDeferredSalesChargeOverOfferingPrice
	OFST451440	Max Redemption Fee	XBRL R/R	RedemptionFeeOverRedemption
	OFST452010	Management Fee Maximum	XBRL R/R	ManagementFeesOverAssets
	OFST452100	TER Excl. Performance Fee	XBRL R/R	ExpensesOverAssets
	OFST452200	Ongoing Charges	XBRL R/R	NetExpensesOverAssets
	OFST454165	Distribution Fee Maximum	XBRL R/R	DistributionAndService12b1FeesOverAssets
Performance	OFDY025xxx	Return periods	XBRL R/R	AverageAnnualReturnYear01/05/10
Dynamic: AuM	OFDY000xxx	TNA / AuM	N-PORT	NET_ASSETS, TOTAL_ASSETS

Partially Available (derivable from prospectus text, not structured)

These fields exist in the text of the prospectus but are NOT in the SEC structured datasets. They would need to be extracted by an LLM — which is exactly the use case:

openfunds OF-ID	openfunds Field	Where in Prospectus
OFST010420	Open-ended Or Closed-ended Fund Structure	Registration form type implies this (N-1A = open-end)
OFST010440	Fiscal Year End	Mentioned in prospectus text, in Submissions JSON
OFST010500	Is Fund Of Funds	Inferred from `AcquiredFundFeesAndExpensesOverAssets > 0`
OFST010580	Is ETF	Inferred from form type or share class structure
OFST010720	Is Passive Fund	Strategy narrative mentions "index" tracking
OFST010730	Management Approach Type	Strategy narrative (active/passive/enhanced)
OFST020300	Valuation Frequency	Prospectus "Pricing of Fund Shares" section
OFST020400	Distribution Policy	Prospectus "Dividends and Distributions" section
OFST020540	Share Class Currency	Inferred; US funds typically USD
OFST020558	Subscription Period Start Date	Only for closed-end or interval funds
OFST400xxx	Minimum Investment	Prospectus "Purchase and Sale of Fund Shares"
OFST451027	Has Performance Fee	Prospectus fee table
OFST451100	Hurdle Rate	Prospectus fee table
OFST013000	Prospectus Date	Filing date in submissions API

NOT Available in SEC Data

These openfunds fields are European/international-specific or distribution-channel-specific and have no SEC equivalent:

openfunds Category	Examples
UCITS/AIFMD fields	OFST160100 Legal Form (SICAV/FCP), OFST011200 Is UCITS With Leveraged Benchmark
European regulatory	OFST350100 EFAMA EFC Category, OFST010075 CSSF Code
Distribution-specific	OFST453151 Is Trailer Fee Clean, OFST451305 Applied Subscription Fee
MiFID/PRIIPs/KID	OFEM-range (MiFID Template), OFEP-range (PRIIPs Template)
ESG/Sustainability	OFST820xxx, OFEE-range (EU sustainability regulation specific)
Country registrations	OFST6000XX (country-specific registration fields)
Solvency II	OFST500xxx
Swiss/German/UK specific	OFST700xxx

4. Summary: Is Asset Class, Currencies, Fees, Risk Data in the SEC Dataset?

Data Category	In SEC Structured Data?	Source	Notes
Asset Class	YES	N-PORT `ASSET_CAT` field	Values: equity, debt, derivative, etc.
Currencies	YES	N-PORT `CURRENCY_CODE` per holding; interest rate risk by currency	Per-holding currency + fund-level
Fees (sales loads)	YES	XBRL R/R	Front-end load, back-end load, redemption fee
Fees (operating expenses)	YES	XBRL R/R	Management fee, 12b-1, TER, net expense ratio
Risk data (narrative)	YES	XBRL R/R	Principal risks text block
Risk data (quantitative)	YES	N-PORT	DV01, credit spread sensitivity, VaR
Performance	YES	XBRL R/R + N-PORT	Annual returns, avg annual returns, monthly returns
Investment Objective	YES	XBRL R/R	Full text of objective
Strategy	YES	XBRL R/R	Full text of principal strategies
Portfolio Turnover	YES	XBRL R/R	Turnover rate
Portfolio Holdings	YES	N-PORT	Security-level: name, CUSIP, ISIN, country, asset type, value
Country of Issuer	YES	N-PORT `INVESTMENT_COUNTRY`	ISO country code per holding
Minimum Investment	PARTIAL	In prospectus text, not structured	LLM extraction target
Distribution Policy	PARTIAL	In prospectus text, not structured	LLM extraction target
ESG/Sustainability	NO	Not in SEC structured data	European regulation specific
UCITS Classification	NO	N/A for US funds	European regulation specific

5. Implication for LLM Training Dataset

The SEC provides an excellent foundation for your LLM training dataset:

Ground Truth (structured) — available directly from SEC:

Fee tables (management fee, expense ratio, loads, 12b-1)
Performance data (1yr, 5yr, 10yr returns)
Investment objective text
Principal risks text
Portfolio turnover rate
Total net assets
Fund/class identifiers (CIK, Series ID, Class ID, Ticker, CUSIP)

Extraction Targets (in prospectus text, to be derived by LLM):

Minimum initial investment amounts
Distribution frequency and policy
Share class currency
Open/closed-end structure
Active vs. passive management
Benchmark index name
Tax status information
Purchase/redemption cut-off times
Settlement cycle

This creates a natural supervised learning setup: the XBRL structured data serves as labels/ground truth, and the prospectus HTML/text serves as input, enabling the LLM to learn the mapping from legal language to structured reference data.

18 KiB Raw Blame History