Insider Selling and Drawdowns: What 194,000 Trades Show

Planned sells carry no bearish signal — discretionary sells underperform by half

The most important filter when evaluating insider sells is whether the transaction was pre-arranged under a Rule 10b5-1 trading plan. A planned sell is a schedule the insider set before acquiring any information advantage; once in place it executes automatically, independent of their current views. A discretionary sell is a decision made in the present.

Splitting the full dataset of 194,750 open-market sells (code S, derivatives and superseded rows excluded) on this single dimension:

Absolute, split-adjusted returns anchored at the first close on or after the filing date. Sample: open-market sells (code S), Oct 2023 – Feb 2026.

Planned sells produced +4.99% at 3 months and +9.33% at 6 months — normal to strong drift, no bearish signal. Discretionary sells underperformed by roughly half: +2.36% at 3 months, with a median of +0.01% — essentially flat. The 6-month gap widens to 3.75 points (+5.58% vs +9.33%).

This directly validates the Form4API default of excluding 10b5-1 plan trades from all sell signals. If you evaluate insider selling without filtering by plan status, you are mixing a no-signal group (101,815 planned transactions) with the informative subset, diluting any predictive content into noise. The planned-vs-discretionary split is not an analytical nicety — it is the first filter that makes the signal visible.

Where the bearish signal actually concentrates — and it is not at the top

Within the 92,935 discretionary sells, we split trades by the stock's proximity to its prior 52-week high at the time of the filing:

Discretionary sells only (code S, no 10b5-1 plan). Price context = filing close ÷ max(prior-365-day close) − 1.

The result contradicts the most popular narrative about insider selling. The "smart money taking profits at the top" story is not supported: insiders who sell when the stock is within 5% of its 52-week high were followed by +3.66% at 3 months and +7.47% at 6 months. Those momentum names kept rising.

The bearish signal concentrates in the opposite bucket: insiders selling a stock that is already significantly below its high. Discretionary sells into stocks 15% or more below their 52-week high produced a median 3-month return of −4.19%. The average (+1.29%) is less extreme because of right-tail recoveries in a subset of names, but the median is the cleaner read of what happens to the typical stock in this bucket — continued weakness.

The pattern has a compelling interpretation: when a stock is already in decline and an insider who has no pre-arranged plan chooses to sell anyway — ignoring the lower price, ignoring the typical reluctance to sell at a loss — that is a meaningful signal about their assessment of the business. They are not trimming profits; they are exiting a weak situation. That is structurally different from a sale at a 52-week high, and the data reflects it.

The sell signal is weaker than the buy signal — by a wide margin

Even taking the strongest bearish cell above — discretionary sells into weakness, median −4.19% at 3 months — the corresponding buy-side signal is materially stronger: insider buys after a deep selloff (stock ≥20% below its high) averaged +10.89% at 3 months across 22,245 transactions. The median for the most bearish sell bucket barely reaches the magnitude of the average for the most bullish buy bucket.

This asymmetry is consistent with the academic literature and has a structural explanation. When an insider buys on the open market with their own capital, there is essentially one plausible motivation: conviction that the stock will rise. Selling has no such purity — even a discretionary sell can reflect personal liquidity needs, concentration reduction, tax planning, or a change in financial goals unrelated to any view of the company.

The practical implication: run the sell signal as a secondary filter, not a primary one. Cluster sells — multiple unrelated insiders selling discretionarily in the same window — raise the probability that the trigger is informational rather than personal. A single discretionary sell into weakness is a weak signal; several concentrated in a short window is a much stronger one. The signal-vs-noise guide covers the clustering logic in detail.

Methodology

• Sample. All open-market sells (transaction code S) in the Form4API database with a matured post-trade return, filed between October 2023 and February 2026.
• Exclusions. Derivative transactions (IsDerivative = true), superseded/amended rows (IsSuperseded = true), and any filing without a ticker or a matured return are excluded. The Is10b5Plan flag partitions planned vs discretionary — it does not exclude either group from the population; it is the analytical split variable.
• Return definition. Absolute, split-adjusted price return anchored at the first market close on or after the filing date (FilingClose), measured at approximately 1, 3, and 6 calendar months. Returns are stored as fractions in the database (0.10 = +10%).
• Price context. Prior drawdown = filing close ÷ max(daily close over the prior 365 days) − 1. Buckets are defined at the <5%, 5–15%, and ≥15% drawdown levels.
• Median vs mean. Both are reported. The means are right-skewed in the positive direction (a minority of large recoveries pulls averages up); the medians give a cleaner read of what happens to the typical trade.
• Not market-adjusted. Returns are absolute, not excess vs the S&P 500 or any sector benchmark. The findings are about relative differences between groups, which largely cancel common market drift.

Reproduce it yourself

The discretionary sell subset this study is based on is directly queryable. Fetch open-market sells with plan trades excluded:

# All discretionary open-market sells — 10b5-1 plan trades excluded
curl "https://api.form4api.com/v1/transactions?code=S&exclude_10b5=true&has_returns=true" \
  -H "X-Api-Key: $FORM4API_KEY"

# Each record includes:
# {
#   "transactionCode": "S",
#   "is10b5Plan": false,
#   "return1m": 0.0066,   // fraction (0.0066 = +0.66%)
#   "return3m": 0.0236,
#   "return6m": 0.0558,
#   "pricePerShare": "142.80",
#   "insiderName": "...",
#   ...
# }

To replicate the price-context split, join on prior-365-day max price from daily price bars and bucket by filingClose / priorHigh − 1. The full filter grammar — codes, date ranges, return thresholds — is in the /docs reference.

Frequently asked questions