Does Insider Buying During Market Crashes Predict Returns?

When markets crash, some corporate insiders start buying their own stock. The logic is simple: they know their business better than anyone, and if they are spending personal money while the stock is down 20%+, they may be seeing value that the market is missing. But does persistent insider buying during drawdowns actually predict higher forward returns? This study tests the hypothesis across every month from 2016 to 2024 where SPY was down 15% or more from its 52-week high.

The question

The generated SQL

Claude generated the following query automatically from the prompt above. This is the raw, unedited output - 70 lines of SQL with 6 CTEs, lateral joins, and insider transaction filtering.

WITH months AS (
  SELECT (DATE '2016-01-31' + (i * INTERVAL '1 month'))::DATE AS month_end
  FROM generate_series(0, 107) AS t(i)
  WHERE (DATE '2016-01-31' + (i * INTERVAL '1 month'))::DATE <= DATE '2024-12-31'
),
spy_drawdown AS (
  SELECT m.month_end,
    entry.close AS spy_close,
    hi.spy_high,
    ROUND((1 - entry.close / NULLIF(hi.spy_high, 0)) * 100, 1) AS spy_dd_pct
  FROM months m
  INNER JOIN LATERAL (
    SELECT close FROM shibui.stock_quotes
    WHERE ticker = 'SPY' AND date <= m.month_end
    ORDER BY date DESC LIMIT 1
  ) entry ON true
  INNER JOIN LATERAL (
    SELECT MAX(high) AS spy_high FROM shibui.stock_quotes
    WHERE ticker = 'SPY'
      AND date BETWEEN m.month_end - INTERVAL '52 weeks' AND m.month_end
  ) hi ON true
  WHERE (1 - entry.close / NULLIF(hi.spy_high, 0)) >= 0.15
),
buy_signals AS (
  SELECT t.issuer_symbol, sd.month_end,
    COUNT(DISTINCT DATE_TRUNC('month', t.tx_date)) AS buying_months,
    COUNT(DISTINCT t.owner_cik) AS unique_buyers
  FROM shibui.insider_transactions t
  CROSS JOIN spy_drawdown sd
  INNER JOIN shibui.general_info g ON t.issuer_symbol = g.symbol
  WHERE NOT t.is_superseded
    AND NOT t.is_derivative
    AND NOT t.is_10b5_1
    AND t.security_title ILIKE '%common%'
    AND t.tx_code = 'P'
    AND t.price_per_share IS NOT NULL
    AND t.tx_date BETWEEN sd.month_end - INTERVAL '12 months' AND sd.month_end
    AND g.type = 'Common Stock'
  GROUP BY t.issuer_symbol, sd.month_end
  HAVING COUNT(DISTINCT DATE_TRUNC('month', t.tx_date)) >= 4
    AND COUNT(DISTINCT t.owner_cik) >= 3
),
with_drawdown AS (
  SELECT bs.*,
    entry.close AS entry_price,
    hi52.wk52_high
  FROM buy_signals bs
  INNER JOIN LATERAL (
    SELECT close FROM shibui.stock_quotes
    WHERE symbol = bs.issuer_symbol AND date <= bs.month_end
    ORDER BY date DESC LIMIT 1
  ) entry ON true
  INNER JOIN LATERAL (
    SELECT MAX(high) AS wk52_high FROM shibui.stock_quotes
    WHERE symbol = bs.issuer_symbol
      AND date BETWEEN bs.month_end - INTERVAL '52 weeks' AND bs.month_end
  ) hi52 ON true
  WHERE (1 - entry.close / NULLIF(hi52.wk52_high, 0)) >= 0.20
),
with_returns AS (
  SELECT wd.month_end, wd.issuer_symbol, wd.entry_price, wd.unique_buyers,
    sd.spy_dd_pct,
    exit_p.close AS exit_price,
    (exit_p.close - wd.entry_price) / NULLIF(wd.entry_price, 0) * 100 AS ret_12m,
    spy_exit.close AS spy_exit_price,
    (spy_exit.close - sd.spy_close) / NULLIF(sd.spy_close, 0) * 100 AS spy_ret_12m
  FROM with_drawdown wd
  INNER JOIN spy_drawdown sd ON wd.month_end = sd.month_end
  INNER JOIN LATERAL (
    SELECT close FROM shibui.stock_quotes
    WHERE symbol = wd.issuer_symbol
      AND date BETWEEN wd.month_end + INTERVAL '11 months'
        AND wd.month_end + INTERVAL '13 months'
    ORDER BY date DESC LIMIT 1
  ) exit_p ON true
  INNER JOIN LATERAL (
    SELECT close FROM shibui.stock_quotes
    WHERE ticker = 'SPY'
      AND date BETWEEN wd.month_end + INTERVAL '11 months'
        AND wd.month_end + INTERVAL '13 months'
    ORDER BY date DESC LIMIT 1
  ) spy_exit ON true
)
SELECT month_end, spy_dd_pct, COUNT(*) AS signals,
  ROUND(COUNT(*) FILTER (WHERE ret_12m > 0) * 100.0 / COUNT(*), 0) AS win_pct,
  ROUND(AVG(ret_12m), 1) AS avg_ret,
  ROUND(MEDIAN(ret_12m), 1) AS med_ret,
  ROUND(PERCENTILE_CONT(0.25) WITHIN GROUP (ORDER BY ret_12m), 1) AS p25,
  ROUND(PERCENTILE_CONT(0.75) WITHIN GROUP (ORDER BY ret_12m), 1) AS p75,
  ROUND(AVG(spy_ret_12m), 1) AS spy_12m,
  ROUND(MEDIAN(ret_12m) - AVG(spy_ret_12m), 1) AS alpha
FROM with_returns
GROUP BY month_end, spy_dd_pct
ORDER BY month_end
LIMIT 50

The results

Ran in 1.1 seconds. Seven months qualified (SPY down 15%+ from its 52-week high):

What the data shows

The March 2020 COVID crash was the standout: 227 stocks met the criteria, 96% were higher a year later, and the median return was +89.9%. That is 28 percentage points of alpha over SPY. Insiders who bought during the fastest market crash in history were overwhelmingly right.

The 2022 bear market tells a different story. Across seven months of drawdown (June through December 2022), results were mixed. Win rates ranged from 36% to 54%, and median returns were mostly negative. The insider buying signal was not enough to overcome the broad market headwind of rising rates and margin compression. Stocks with heavy insider buying actually underperformed SPY in every 2022 month tested.

The takeaway: insider buying during crashes is a meaningful signal, but it is not a mechanical trading rule. It worked spectacularly during a sharp, V-shaped recovery (2020) and poorly during a prolonged, grinding bear market (2022). The signal may be more useful for identifying which stocks to research further rather than as a standalone buy list.

Which sectors gave the strongest insider buying signal

The aggregate results above mask significant sector-level variation. Financials dominate signal counts in both crashes, accounting for roughly 40% of all insider buying signals. Banks and financial companies have the most officers and directors required to file Form 4, so the aggregate results are heavily weighted toward financial sector performance.

In March 2020, every sector produced positive 12-month returns. The V-shaped recovery lifted everything. Health Care had the highest median return (+129.8%) driven by biotech and pharma companies that rallied on pandemic demand. Consumer Discretionary followed at +245.1% median, Industrials at +147.0%, and Energy at +119.8%. Even the weakest sector, Utilities, was positive at +5.7% median.

March 2020 (SPY -24.0%)

In September 2022, the sector breakdown tells a different story. Only Consumer Discretionary (+17.7% median, 78% win rate) and Industrials (+26.5% median, 80% win rate) generated positive returns from insider buying signals. Health Care, the top performer in 2020, collapsed to -30.1% median with only a 33% win rate. Financials, representing the largest signal group with 72 qualifying companies, had a 33% win rate and -9.8% median return. The aggregate -23.8pp alpha in September 2022 was not uniform: insiders in consumer and industrial companies were right, while insiders in health care and financials were wrong.

September 2022 (SPY -25.6%)

What happens at a lower threshold

The main study uses a 15% SPY drawdown threshold. Lowering it to 10% surfaces 17 qualifying months instead of 7, but mostly by extending existing crashes rather than finding new independent episodes. The 2022 bear market stretches from 7 months to 10. The 2020 COVID crash extends from 1 month to 4 (including the recovery months of April and May). The 2023 SVB banking crisis, which did not qualify at 15%, appears as three new months at the 10% threshold.

Additional episodes at 10% SPY drawdown threshold

The only truly new episode is December 2018, the Christmas crash. Insider buying generated a 72% win rate and +14.5% median return, but SPY itself recovered +28.7% over the following 12 months, so the insider-bought stocks underperformed the market by 14 percentage points. The 2023 SVB months (January through March) produced -25 to -43pp alpha as the insider-heavy regional bank stocks continued to decline while the broader market rallied. The lower threshold confirms the original finding: the signal is not consistent enough to be a mechanical trading rule.

Methodology

• Data source: Shibui Finance, covering NYSE + NASDAQ, 10,000+ securities
• Insider transactions from SEC Forms 3/4/5. Excludes 10b5-1 pre-planned transactions, derivative transactions, and superseded filings.
• Signal criteria: SPY down 15%+ from 52-week high, stock down 20%+ from its own 52-week high, 4+ months of insider purchases by 3+ distinct buyers within the trailing 12 months.
• Forward returns computed over an 11-13 month window to account for trading day availability.
• Survivorship bias: delisted stocks are included in the signal count but may be missing forward price data. This biases returns toward survivors.
• Sector classification uses GICS sectors from the general_info table. Sector assignment is based on the company's classification at the time of the query, not at the time of the trade.

Frequently asked questions

How was this study run?

A single natural-language prompt to Claude with the Shibui Finance MCP connector. Claude generated the SQL automatically and returned the results in under 2 seconds. The full SQL is shown above on this page.

What counts as insider buying in this study?

Open-market purchases (transaction code P) of common stock by officers and directors, filed on SEC Forms 3, 4, and 5. The study excludes pre-planned 10b5-1 transactions, derivative exercises, and superseded filings to isolate genuine conviction buying.

Why exclude 10b5-1 plans?

10b5-1 plans are pre-scheduled transactions that do not reflect a real-time conviction signal. Including them would dilute the signal with non-informative noise. The study focuses on discretionary purchases where insiders actively chose to buy.

Can I reproduce this backtest?

Yes. Connect Shibui to Claude (free), paste the prompt shown above, and Claude will generate and run the query. Results update daily as new insider filings are processed from SEC EDGAR.

Does insider buying work better in some sectors than others?

In March 2020, every sector produced positive 12-month returns from insider buying signals. In September 2022, only Consumer Discretionary and Industrials generated positive median returns. Financials, which account for roughly 40% of all insider buying signals, had a 33% win rate and -9.8% median return in 2022. The sector composition of insider buying matters more than the aggregate signal count.

Is this financial advice?

No. This is a statistical study using historical data. Past performance does not predict future results. The study has known limitations including survivorship bias. Consult a licensed advisor before making investment decisions.