The hard part of alternative data isn't collecting the data. It's knowing when the data is wrong. For institutional and quantitative data teams, SEC filings are an important source of holdings information. But the raw filing is not the dataset.
Turning filings into a reliable, research-ready holdings dataset requires interpretation, normalization, entity resolution, historical reconciliation, corporate-action handling, and a validation layer capable of identifying anomalies before they propagate downstream.
At Radient, this is a significant part of the engineering work behind our 13F and N-PORT datasets.
The two datasets serve different purposes. Form 13F provides institutional investment manager-level holdings and is reported quarterly, while Form N-PORT provides portfolio holdings at the fund or series level and contains substantially more granular information.
Radient's 13F and N-PORT datasets are designed for enterprise users who need structured holdings data rather than raw filing documents.The datasets are available through API, flat files, and Snowflake, allowing data teams to integrate holdings information directly into their existing research and data infrastructure. Learn more
The 13F dataset also enables security-level analysis of institutional ownership, making it possible to study how institutional ownership changes over time, identify increasingly crowded securities, and analyze which managers are accumulating or reducing positions. For teams building systematic research, portfolio analytics, institutional ownership models or alternative-data pipelines, the underlying quality of the holdings dataset matters as much as the breadth of coverage. The filing is the source. The real work is making the data trustworthy.
A 13F filing may look straightforward: a manager reports securities, shares or principal amounts, and market values. But several characteristics of the filing process can introduce errors if the data is ingested without additional validation.
A manager can file an original 13F-HR and subsequently file a 13F-HR/A amendment. However, an amendment does not always represent the same type of change. In some cases, additional holdings were omitted from the original filing and the amendment provides those additional positions. In other cases, the amendment is a restatement that corrects information previously reported.
These two scenarios require completely different reconstruction logic.
For an additional-holdings filing, the portfolio may need to be reconstructed as:
Original holdings + additional holdings
For a true restatement, the corrected information should replace the previously reported information:
Corrected holdings, rather than original + corrected holdings
Treating every amendment as a separate holdings dataset can therefore result in missing positions, duplicate positions, or inflated portfolio values.
There is another complication: the filing type itself is not always sufficient to determine how the information should be interpreted. In practice, the contents and relationship between filings need to be examined.
At Radient, our pipeline resolves the treatment of amendments at the filing level rather than applying a single generic rule to every 13F-HR/A.
One of the most important and less obvious problems is value denomination.13F filers can report values using different scales, and the stated denomination does not always reconcile cleanly with the actual numbers in the information table. If the denomination is interpreted literally when it should not be, a holding can be off by orders of magnitude.
A position that should be $40 million can effectively become $40 billion in the resulting dataset. The problem is that the row itself may look perfectly valid. It contains a security identifier, a position and a numerical value. There may be no obvious parsing error. The anomaly only becomes apparent when the number is compared with other information.
This is why Radient does not simply accept the reported value.
Our validation algorithms cross-check individual holdings against multiple signals, including:
For example, if a manager's reported value for a security changes by three orders of magnitude while the reported share count and security price do not support that movement, the position is flagged for investigation.
The objective is to identify these order-of-magnitude anomalies before they enter the published dataset.
A filing should rarely be evaluated in isolation.
Historical context provides another powerful way to identify incorrect data. Suppose a manager has historically reported between 25 and 40 unique positions per quarter and suddenly appears with 150 positions. That does not automatically mean the filing is wrong, but it is a strong anomaly that deserves investigation.The same applies to position values.
If a $100 million position suddenly becomes $10 billion without a corresponding change in shares, security price, corporate action, or other filing information, the change needs to be investigated rather than blindly published.
This is why Radient's pipeline compares current filings with the manager's historical disclosures.
The historical record becomes part of the validation process.
The impact of a bad 13F record is rarely limited to one row. Most institutional users do not consume a single filing. They build datasets and models on top of thousands of filings across managers and time. A single incorrect holding can therefore propagate into multiple downstream calculations.
A mis-scaled position changes the manager's portfolio value and consequently its calculated weight in ranked institutional holder tables.
Flows are calculated as: Position(t) − Position(t−1)
If a restatement is mishandled, the resulting difference can look like a trade that never happened. A manager may appear to exit and subsequently re-enter a security even though the apparent trade is simply an artifact of how the filings were reconstructed.
13F option positions are reported in terms of the underlying security's share equivalents. Treating these as ordinary common-stock positions without accounting for the instrument type can produce misleading ownership calculations.
Multiple managers can report the same position under shared investment discretion. Without correctly resolving the relevant filer and other-manager information, the same economic position can be counted multiple times. The further downstream the data travels, the harder these errors become to detect. By the time an incorrect position has been incorporated into a crowding score, consensus score, ownership percentage, or portfolio factor, the original anomaly can be averaged into a number that looks completely reasonable.
That is the dangerous part of data quality.
Reliable holdings data requires considerably more than parsing the SEC information table. At Radient, our 13F processing and validation includes several layers.
CUSIP is an important reported identifier, but it is not necessarily a permanent identifier for a security across its entire history. CUSIPs can change following corporate actions, reorganizations and other events. We validate reported identifiers and map securities to stable internal identifiers so that the same economic security can be consistently tracked across historical filings.
Splits, mergers, ticker changes and share-class reorganizations can create artificial breaks in a holdings time series. Historical share counts need to be interpreted in the context of these events so that a corporate action is not incorrectly interpreted as a portfolio transaction.
CIK provides an important anchor for identifying a filer, but manager names can change and large investment organizations may have multiple filing entities. We connects individual filers to the appropriate managers or manager family.
We distinguish between additional-holdings amendments and true restatements and reconstruct the portfolio accordingly. This is particularly important when building historical holdings and quarter-over-quarter position changes.
Some holdings can initially be withheld from a filing and subsequently disclosed through an amendment. When those holdings become available, they need to be incorporated into the historical dataset. This can also change previously calculated historical flows, which makes proper versioning and historical reconstruction important. Radient accounts for all these cases.
Finally, individual holdings are checked against historical disclosures, security prices, share counts and portfolio-level information. Large changes that cannot be explained by the underlying data are flagged rather than automatically published.
The result is a dataset that is cleansed, identifier-mastered, corporate-action adjusted and amendment-resolved before it reaches the end user.
N-PORT presents similar challenge but much larger at scale. While 13F provides institutional investment manager-level holdings on a quarterly basis, N-PORT operates at the fund and series level and provides much more granular portfolio information. The resulting data volume is substantially larger, and the underlying schema introduces additional complexity.
N-PORT portfolios can contain a wide range of instruments, including equities, fixed income, derivatives, repos and private placements. Not every instrument has a clean CUSIP or ISIN that can simply be mapped to an equity security master.
For a usable dataset, the raw filing information needs to be parsed, normalized and linked across the relevant entities. This requires dedicated validation around the underlying fund, security and reporting information rather than treating the filing as a flat table of positions.
At Radient, we also link N-PORT series back to Form ADV, creating a connected layer across:
Fund → Adviser → Portfolio → Security → Historical Holdings
This relationship is particularly useful for data teams that need to move between fund-level characteristics and the underlying portfolio rather than maintaining separate disconnected datasets.
The distinction we make at Radient is between filing ingestion and data reconstruction. Downloading a filing answers one question:
Did we receive the source document?
A reliable holdings dataset requires answering several more:
These checks happen upstream of what the customer receives.The goal is not simply to ingest every filing. The goal is to correctly reconstruct the holdings represented by those filings.This matters because holdings data is rarely the final output. It becomes the foundation for institutional ownership research, portfolio analytics, manager screening, crowding analysis, consensus signals, factor models and systematic investment research.
A small error at the filing level can therefore become a much larger error in the model built on top of it.