A $1M Insider Buy Is Not Always Large: Measuring Purchases Against Existing Holdings

A million-dollar insider purchase sounds large. Economically, that description is incomplete.

The same disclosed amount can add less than one percent to one reporting owner's position and establish nearly the entire reported position of another. If relative commitment captures conviction better than nominal dollars, purchases that represent a larger share of post-transaction holdings should produce stronger subsequent outcomes.

Using the July 29, 2026 InsiderAlpha Form 4 Research Database, we tested that hypothesis across 7,804 mature direct-ownership purchase events. Larger position increases showed a small raw advantage at 180 days, but the relationship was not monotonic and did not survive the stronger issuer-matched diagnostic.

Relative purchase size adds economic context. This release does not support using it as a universal multiplier for an insider-buying signal.

The key result

The 20% to under 50% cohort produced the highest median 180-day outcome, at +8.22%. The 50% to 100% cohort did not improve on it, and purchases under 1% had the highest positive-outcome rate.

Purchase as share of post-trade holdings	Events	Issuers	Median disclosed value	Median 30D	Median 90D	Median 180D	Positive at 180D
Under 1%	2,031	210	$13,063	+1.50%	+4.82%	+7.16%	66.3%
1% to under 5%	1,798	315	$46,645	+2.14%	+3.85%	+6.92%	63.7%
5% to under 20%	1,936	376	$103,804	+1.92%	+5.51%	+7.47%	63.3%
20% to under 50%	1,042	337	$150,501	+2.20%	+5.84%	+8.22%	64.9%
50% to 100%	997	356	$186,390	+1.91%	+4.25%	+7.83%	65.8%

The ordering is not a staircase. Moving from a median purchase fraction of 0.33% in the first cohort to 99.69% in the last cohort did not create a corresponding progression in returns or positive-outcome rates.

Across all 7,804 events, the Spearman rank correlation between purchase fraction and 180-day outcome was 0.002. In practical terms, relative size did not rank subsequent outcomes in this sample.

What the percentage measures

For each eligible owner event, we calculated:

purchase fraction = shares purchased / shares beneficially owned after the event

If an owner purchased 10,000 shares and reported 200,000 shares held afterward, the fraction is 5%. If the same 10,000 shares equaled the entire post-transaction position, the fraction is 100%.

This is not purchase value divided by wealth, compensation or market capitalization. Form 4 does not provide a complete personal balance sheet. The measure describes the purchase relative to the holdings reported for that security and ownership form.

Earlier research has used shares traded relative to holdings to distinguish economically different insider transactions. For example, the study Some Insider Sales Are Positive Signals used a holdings-relative measure when examining insider sales. Our test asks a narrower question about reported purchases in the InsiderAlpha coverage universe.

Sample construction

We applied the following rules:

1. Keep transaction code P, which the SEC defines as an open-market or private purchase.
2. Start on January 1, 2007 and require calculated 30-, 90- and 180-day market-price outcomes.
3. Keep current, non-derivative transaction rows with positive purchased shares and positive post-transaction holdings.
4. Restrict the primary analysis to direct ownership (D). Indirect holdings can belong to different trusts, partnerships or other accounts whose denominators should not be combined casually.
5. Aggregate price tranches to one ticker-reporting owner-transaction date-security title-ownership nature event.
6. Sum purchased shares and disclosed value, use the maximum reported post-transaction holding for the event, and use the median price outcome when source rows share the event.
7. Exclude 79 aggregated events whose calculated fraction was above 100%, because those combinations did not provide a defensible denominator.

The resulting primary sample contains 7,804 events across 528 issuers and 2,583 reporting owners, dated February 6, 2007 through January 16, 2026.

Coverage narrows materially. The release contains 40,799 mature code P rows from 2007 onward, but only 15,920 mature non-derivative rows have positive purchased shares and usable post-transaction holdings. The study therefore describes the canonical holdings-covered subset, not every historical purchase in the database.

The SEC Form 4 itself separates transaction amount, holdings after the transaction, and direct or indirect ownership. Those fields should remain separate until the unit of analysis is defined. See the official Form 4 and instructions and the InsiderAlpha field dictionary.

High versus low relative size

To simplify the comparison, we combined purchases under 5% into a low-relative-size cohort and purchases from 20% through 100% into a high-relative-size cohort.

Cohort	Events	Median fraction	Median disclosed value	Median 180D	Positive at 180D
Under 5%	3,829	0.89%	$25,036	+7.05%	65.1%
20% to 100%	2,039	49.07%	$160,432	+8.16%	65.3%

The raw median difference is +1.11 percentage points. The positive-outcome rates are nearly identical. This is a modest descriptive separation, not evidence that multiplying a position should mechanically improve a signal.

The within-issuer test

Relative purchase sizes are not distributed randomly across companies. Some issuers have ownership requirements, concentrated founders or recurring small purchases. A broad comparison can therefore confuse issuer composition with relative commitment.

We isolated 141 tickers with at least three low-relative-size and three high-relative-size mature events. For each ticker, we calculated:

median 180D outcome after high-relative-size events - median after low-relative-size events

Matched-issuer diagnostic	Result
Matched issuers	141
Median high-minus-low spread	+1.06 pp
Issuers where high-relative-size median was higher	51.8%
Issuer bootstrap 95% interval	-1.56 to +2.92 pp

The interval crosses zero, and the high-relative-size cohort led in only slightly more than half of matched issuers. This release does not establish a stable within-company premium.

A stricter owner-and-ticker comparison produced a +3.99-point median spread, but only 37 owner-ticker pairs had at least two events in each cohort. Its bootstrap interval was wide, from -1.94 to +10.08 points. That diagnostic is directionally interesting and statistically inconclusive.

Apparent new positions

The 50% to 100% bucket contains two economically different cases:

• 502 events between 50% and under 100%, with a median 180-day outcome of +8.48%;
• 496 events at exactly 100%, with a median outcome of +7.65%.

An observed 100% fraction means the purchased shares equal the post-transaction holdings reported for that direct security position. It is reasonable to label this an apparent new reported position, but not proof that the person had no derivative, indirect or otherwise undisclosed economic exposure.

Sensitivity checks

The raw high-minus-low difference remained positive in two broad time periods: +0.58 points during 2007-2015 and +1.26 points during 2016-2026. Neither period produced a large step change.

Including both direct and indirect ownership expanded the valid sample to 10,214 events and flattened the upper buckets further. Median 180-day outcomes were +7.51% for the 20% to under 50% cohort and +7.47% for the 50% to 100% cohort. This sensitivity supports the same restrained conclusion, while also showing why direct and indirect denominators should not be treated as interchangeable.

What the result changes

Nominal value and relative size answer different questions:

• nominal value measures disclosed transaction scale in dollars;
• relative size measures how much the reported position changed;
• neither field alone captures personal wealth, issuer valuation, liquidity or the information available when the filing became public.

A defensible model can retain relative size as a feature, especially in interactions with insider role, repeat buying, cluster context and issuer characteristics. What the evidence rejects is a one-dimensional rule such as:

larger percentage increase = proportionally stronger signal

The earlier signal-framework study found no monotonic premium for nominal purchase value. This study reaches a similar conclusion for relative position size: the field is more informative as context than as an isolated ranking mechanism.

Limitations

• Outcomes begin from the transaction-date market-price baseline, not the first tradable price after EDGAR acceptance.
• Returns are absolute security-price changes, not market- or sector-adjusted alpha.
• The sample covers only events with usable post-transaction holdings and therefore excludes much of the validated legacy history.
• Post-transaction holdings describe the reported security and ownership form, not total wealth or every economic exposure.
• Code P includes private purchases as well as exchange purchases.
• Repeated events from the same owner and issuer are not fully independent.
• Bucket boundaries were selected before reviewing outcomes, but alternative boundaries can still change descriptive medians.

This is an observational event study, not an executable backtest or investment recommendation.

Conclusion

Relative purchase size improves the description of an insider transaction. A $1 million purchase can be economically small for one owner and transformative for another.

But better context is not the same as a standalone edge. High-relative-size events showed a small raw 180-day advantage, while rank correlation was effectively zero and the issuer-matched interval crossed zero. The defensible use is to combine position change with transaction semantics, owner identity, filing chronology and issuer context—not to assign a universal conviction premium.

The versioned data fields used in this study are documented in the research database, and a free 10,000-row sample is available for independent inspection.