Insider Signal Methodology

Scoring SEC Form 3, 4, and 5 filings to distinguish informative insider transactions from routine trading activity.

Overview

Corporate insiders—officers, directors, and large shareholders—have privileged access to material non-public information about their companies. While trading on such information is illegal, insiders often trade based on their general assessment of company prospects. Academic research has shown that aggregate insider trading can predict future stock returns.

Our insider signal methodology transforms raw SEC filing data into actionable intelligence by scoring each transaction based on who traded, what they traded, and how much they traded.

Validation Status

Research results below retain their original samples, uncertainty and refuted claims. Older implementation records are available separately; their role and trade-size weights are not the current scoring formula described below.

Real-data backtest

-1.62 percent

Real Form 4 + split-adjusted daily_returns backtest (corrected bd-rbkns methodology, >=18c7ed4ba: coverage-gap and delisting skips counted, tolerance-only confirmation) still refutes the legacy README magnitude: observed -1.62% 90-day market-adjusted return for late filings over 2021-07-25..2026-07-25, 95% CI [-1.96%, -1.28%], n=69,846 late filings among 1,319,618 priced Form 4 events (the 2026-10-03 run measured -1.74%, CI [-2.08%, -1.39%], over 61,168 late filings; the 2026-11-02 value lies inside that interval). The 2026-11-02 run read a local mirror of production insider_signals (form 4) and daily_returns whose row counts matched production. Its inputs grew more than the one-month window shift explains: the old window now holds 1,369,010 Form 4 events against 1,199,505 on 10-03, because historical rows were added since. Delisting skips rose from 1,727 to 7,610 and no-price skips from 37,230 to 51,472; both stay in the denominator accounting, and the cause of the delisting rise was not decomposed. The e-process value is pinned at its exp(700) numerical ceiling, so it no longer moves with n. The small negative effect itself is very strongly supported (classical clustered p effectively zero at the persisted precision; the classical artifact's result.evalue field is the capped kappa=1/2 p-to-e calibrator of that test). The evalue block here tracks the parallel Grunwald-Vovk betting e-process from late-filing-evalue-latest.json, which tests the winsorized conditional-mean null in filing-date order (presentation adjudication: bd-l6i53). The claimed -8.2% magnitude remains refuted under all of these.

Evidence provenance

late-filing-underperformance-8.2

Computed 2026-11-02; evidence tests/scripts/late-filing-underperformance.test.ts; docs/claims/late-filing-latest.json

Real-data backtest

4-6 percent

Jagolinzer (2009), Management Science 55(2), 224-239, remains the external citation for the historical Stanford 4-6% 10b5-1 abnormal-return finding. The Midas Edge real SEC EDGAR + EODHD validation sample does not establish it: the latest primary 180-day result is 4.22% mean avoided loss, 95% CI [-4.88%, 13.32%], n=25, p=0.33. The producer labels this 'consistent' because the point estimate now falls inside 4-6% (it was 3.97% on 2026-10-02), but the interval still includes zero, so the target range lies inside a wide interval rather than being reproduced.

Evidence provenance

10b5-1-abnormal-returns-4-6

Computed 2026-11-02; evidence tests/integration/insider/ten-b5-one-abnormal-return.integration.test.ts; docs/claims/10b5-1-abnormal-return-latest.json

Real-data backtest

Feature-backed

Corrected real-data methodology (filing-date public-information anchoring, split-adjusted daily_returns, and >=80% per-ticker window coverage) over 1,681,898 Form 4 observations (source real-insider-form4-2026-10-30), computed on a local mirror of production insider_signals (form 4) and daily_returns whose row counts matched production. The week-bucketed Kendall tau is +0.29 with ticker-cluster bootstrap CI [-0.111, +0.644] (p=0.24), and the T+90 top-bottom spread is +0.13% with CI [-2.01%, +2.21%]. The 2026-10-04 run (1,411,603 observations) had tau +0.20 (CI [-0.067, +0.689], p=0.42) and a T+90 spread of +0.30% (CI [-1.58%, +1.28%]); the T+90 interval still includes zero and is now wider. Short horizons still separate (T+1 +0.99%, CI [+0.89%, +1.13%]; T+5 +1.22%, CI [+0.92%, +1.46%]). claimStatus is underpowered and the monotonicity claim remains unverified for marketing use.

Evidence provenance

insider-decile-monotonicity

Computed 2026-11-02; evidence tests/integration/research/insider-decile-realdata.test.ts; docs/claims/insider-decile-latest.json

Historical implementation record

Backed claim

25 percent

Insider signal component weight defined in src/services/insider

Evidence provenance

insider-weight-transaction-type-25

Test evidence: tests/services/insider.test.ts

Historical implementation record

Backed claim

25 percent

Insider signal component weight defined in src/services/insider

Evidence provenance

insider-weight-trade-size-25

Test evidence: tests/services/insider.test.ts

Historical implementation record

Backed claim

20 percent

Insider signal component weight defined in src/services/insider

Evidence provenance

insider-weight-role-20

Test evidence: tests/services/insider.test.ts

Historical implementation record

Backed claim

4 count

Minimum trade count for insider skill calibration

Evidence provenance

insider-historical-threshold-4

Test evidence: tests/services/insider.test.ts

Source: docs/claims.json, generated 2026-09-07.

SEC Filing Data Sources

All insider transactions must be reported to the SEC via Forms 3, 4, or 5:

Form 3 - Initial Statement

Filed within 10 days of becoming an insider (officer, director, or 10%+ owner). Establishes the baseline of securities held.

Form 4 - Statement of Changes

Filed within 2 business days of any transaction. This is the primary source of timely insider trading signals. The tight deadline makes Form 4 filings highly actionable.

Form 5 - Annual Statement

Filed within 45 days after fiscal year end for transactions exempt from Form 4 reporting or inadvertently not reported. Less timely but captures small transactions and gifts.

Signal Scoring Formula

The final score ranges from -100 to +100: positive for purchases, negative for sales, and zero for neutral transaction types. Magnitude combines role, transaction type, dollar value and holdings context.

\displaystyle B_{\text{TS}} = 100(0.30R + 0.25T + 0.35D + 0.10H + 0.05)

Where:

$R$ = Role weight (0-1): Significance of the insider’s position
$T$ = Transaction type weight (0-1): Informativeness of the transaction type
$D$ = Dollar value weight (0-1): Magnitude of the transaction
$H$ = holdings weight (0-1): square root of the fraction of prior holdings traded, capped at 1; zero when holdings context is unavailable
$P$ = 10b5-1 penalty (0 or 0.4): Reduction for pre-scheduled trades

The TypeScript path starts with the base magnitude above, applies the plan penalty, late-filing penalty, short adoption-window boost, person credibility, role multiplier, unusual personal trading, transaction-age decay and insider-cluster multiplier, then rounds, assigns direction and clamps the score.

The WASM adapter path has a different base: 10 + 30R + 25T + 35D, with the plan penalty and an initial rounding and cap. It then adds the rounded 10H holdings contribution and applies credibility, late-filing, short-window, role, personal-trading, age and cluster adjustments with intermediate rounding. The stored rationale records contributions; interpreting them requires the scorer path, since the two paths are not numerically identical.

For example, the TypeScript calculation for a director buying $1m, trading 25% of prior holdings on the scoring date, with no plan or other flags and no person-credibility record, yields +63. The corresponding sale yields -63; tax withholding remains neutral at 0.

Role Significance Weights

Not all insiders are created equal. A CEO buying shares is more significant than a 10% owner selling. We assign weights based on the insider's likely access to material information:

Role	Weight	Rationale
CEO / Chief Executive Officer	100%	Highest decision-making authority, direct control over company strategy
CFO / Chief Financial Officer	95%	Deep visibility into financial performance, accounting, and forecasts
COO / Chief Operating Officer	90%	Operational insight, day-to-day business performance
President	85%	Senior leadership with broad company visibility
CTO / Chief Technology Officer	80%	Product and technology roadmap knowledge
General Counsel	70%	Legal matters, M&A activity, regulatory issues
Officer (other)	65%	Executive-level position with material information access
Director	60%	Board member with fiduciary duty and strategic oversight
VP / Vice President	55%	Senior management with departmental visibility
10% Owner	50%	Large shareholder with significant economic interest
Treasurer	50%	Cash management and capital structure visibility
Secretary	40%	Administrative executive role

Transaction Type Classification

The type of transaction determines both its direction (bullish/bearish/neutral) and informativeness. Open market purchases and sales carry the most signal; gifts and administrative transactions carry the least.

P · Purchase

Direction: bullish
Weight: 100%
Description: Open market purchase - strongest bullish signal

S · Sale

Direction: bearish
Weight: 100%
Description: Open market sale - strongest bearish signal

F · Tax Withholding

Direction: neutral
Weight: 5%
Description: Shares sold to cover tax obligations on vesting

A · Award/Grant

Direction: neutral
Weight: 5%
Description: Stock grant or award - company compensation decision

M · Exercise/Conversion

Direction: neutral
Weight: 5%
Description: Option exercise or conversion - often routine

G · Gift

Direction: neutral
Weight: 0%
Description: Gift to family, charity, or trust - tax planning

D · Disposition to Issuer

Direction: neutral
Weight: 5%
Description: Return of shares to company

J · Other

Direction: neutral
Weight: 5%
Description: Other acquisition or disposition

Code	Type	Direction	Weight	Description
P	Purchase	bullish	100%	Open market purchase - strongest bullish signal
S	Sale	bearish	100%	Open market sale - strongest bearish signal
F	Tax Withholding	neutral	5%	Shares sold to cover tax obligations on vesting
A	Award/Grant	neutral	5%	Stock grant or award - company compensation decision
M	Exercise/Conversion	neutral	5%	Option exercise or conversion - often routine
G	Gift	neutral	0%	Gift to family, charity, or trust - tax planning
D	Disposition to Issuer	neutral	5%	Return of shares to company
J	Other	neutral	5%	Other acquisition or disposition

Dollar Value Weighting

Larger transactions carry more signal. We use a logarithmic scale to normalize across the wide range of transaction sizes:

Normalize the transaction value, then clamp it to the 0-1 range:

\displaystyle d = \frac{\log_{10}(\text{Value}) - 4}{4}

\displaystyle D = \min(1, \max(0, d))

This maps transaction values to a 0-1 scale:

$10K

$100K

25%

$1M

50%

$10M

75%

$100M+

100%

10b5-1 Trading Plan Detection

Rule 10b5-1 allows insiders to establish pre-scheduled trading plans when they are not in possession of material non-public information. These plans then execute automatically, providing an affirmative defense against insider trading claims.

Why 10b5-1 Plans Matter

Transactions executed under 10b5-1 plans are less informative because:

The plan was established in advance, possibly months before execution
The insider cannot modify the plan while possessing MNPI
Timing and size are predetermined, not based on current information

We apply a 40% penalty to signals from 10b5-1 plans. This is reflected in the penalty term $P$ in our scoring formula.

\displaystyle \text{Adjusted Score} = \text{Raw Score} \times (1 - 0.40) = 0.60 \times \text{Raw Score}

Signal Direction Classification

Each transaction is classified as bullish, bearish, or neutral:

Bullish

Open market purchases (P). Insider is putting personal capital at risk, signaling confidence in future prospects.

Bearish

Open market sales (S). Insider is reducing exposure, though sales may be for diversification or personal reasons.

Neutral

Tax withholding (F), grants, exercises, gifts, and other administrative transactions. These reflect compensation plans or estate planning, not market views.

Aggregated Insider Sentiment

Individual transactions can be noisy. We aggregate across all insiders for a given security to compute net sentiment:

\displaystyle \text{Net Value} = \sum_{\text{buys}} |\text{Value}_i| - \sum_{\text{sells}} |\text{Value}_i|

Bullish Sentiment: Net Value > 0 (more buying than selling)
Bearish Sentiment: Net Value < 0 (more selling than buying)
Neutral Sentiment: Net Value ≈ 0 (balanced activity)

We also track the count of net buyers vs. net sellers over a 90-day window to provide additional context beyond dollar values.

Red Flag Detection

Certain patterns in insider activity may indicate elevated risk:

Cluster Selling

Multiple insiders selling within a short window (e.g., 3+ insiders in 2 weeks) may indicate shared negative outlook.

C-Suite Exit

CEO, CFO, or multiple executives selling significant portions of their holdings can signal fundamental concerns.

Unusual Size

Transactions significantly larger than the insider's historical pattern may warrant attention.

Earnings Window

Transactions close to earnings announcements (within blackout windows) may indicate timing awareness.

Confidence Score

Confidence describes available data and filing recency. It is not the probability of a profitable trade.

\displaystyle C = 100(0.20 + 0.40R_{\text{recency}} + 0.40C_{\text{completeness}})

The shared calculation uses 20% base, 40% filing recency, and 40% field completeness.

Completeness checks role, transaction type, value or shares, and holdings context.
Recency weights are 1 through 7 days, 0.8 through 30, 0.6 through 90, 0.4 through 180, and 0.2 thereafter; missing dates use 0.3.
The TypeScript path adds 5 percentage points when a person-credibility record exists, capped at 100.
The WASM adapter path averages the shared result with its own role/value/type completeness score, without that extra addition.

A complete filing dated on the scoring date receives 100 in the shared calculation; with all four completeness fields absent it receives 60. These are data-quality examples, not return forecasts.

Access by Subscription Tier

Every plan, Free included, sees the full history of insider signals. Scoring, including the 10b5-1 plan penalty, is also the same on every plan.

Academic References

Seyhun, H. N. (1986). Insiders' profits, costs of trading, and market efficiency. Journal of Financial Economics
Lakonishok, J., & Lee, I. (2001). Are insider trades informative? Review of Financial Studies
Jagolinzer, A. D. (2009). SEC Rule 10b5-1 and insiders' strategic trade. Management Science, 55(2), 224-239. — External historical basis for the 4-6% 10b5-1 abnormal-return claim; our claims registry also links the current SEC/EODHD replication sample.
Jagolinzer, A. D., Larcker, D. F., & Taylor, D. J. (2011). Corporate governance and the information content of insider trades. Journal of Accounting Research
Cohen, L., Malloy, C., & Pomorski, L. (2012). Decoding inside information. Journal of Finance

Limitations & Known Biases

Form 4 timeliness varies; late filings are explicitly tracked but cannot be assumed to carry the same signal across regimes.
10b5-1 plan flags depend on footnote quality and may miss ambiguous or omitted plan language.
Dollar-value thresholds can overstate signal for founders or executives with concentrated personal wealth.
Signals are company-level research inputs, not investment recommendations or evidence of unlawful insider trading.

What runs behind this

Insider transaction processing and scoring
Reading SEC Forms 3, 4 and 5
Timing unusualness and event proximity scoring
10b5-1 plan analysis

Disclaimer

Insider trading signals are derived from public SEC filings and reflect historical transactions. Past insider activity does not guarantee future stock performance. Insiders may trade for personal reasons unrelated to company fundamentals (diversification, liquidity needs, tax planning). This analysis is for informational purposes only and does not constitute investment advice.