複製美國國會議員的股票交易有用嗎？我測試了 17,859 份披露紀錄

The popular rule is simple: buy what members of Congress buy and avoid what they sell. I tested it on 17,859 common-stock disclosures filed under the STOCK Act from January 2020 to August 2025 (27,790 trade lines, 1,601 stocks), entering at the first close after each filing became public and comparing against the S&P 500 over the same dates. The rule fails, and the reason isn't the reporting delay.

Medians of each disclosure's return minus the S&P 500 over identical dates. Both directions rose in absolute terms (about 11% after a year); the index rose more.

Five findings

1. Buys don't beat sells. On the 760 filing dates with both purchases and sales, the one-year buy-minus-sell spread was 1.13 points with a t-statistic of 1.15. At one, three and six months it was zero. If sales were a bearish signal, sold stocks should do clearly worse than bought ones. They don't, probably because people sell to rebalance, pay taxes or fund spending.

2. Perfect foresight doesn't rescue it. I reran everything from the day after the trade itself, which gives a follower information nobody had yet. Purchases still trailed by 5.71 points and sales by 6.67, and the spread between them was -0.02 points.

3. There is some purchase timing. Purchase windows did better than the same stocks' later windows: shifting each one-year window forward 504 sessions, the paired improvement was 5.46 points (t = 3.27). Sales showed nothing reliable. But the real purchase windows still lagged the index, so this is "less bad than the same stock later", not "better than the market now".

4. Committee seats didn't help. Purchases in industries a member's committee oversees (436 events, 58 members, mapped from official rosters and SEC industry codes) gained 16.2% against 11.0% for other purchases. Against the market over the same windows the gap shrinks to 0.40 points, and matched-date and within-member comparisons point the other way.

5. Faster filers did less badly. The median purchase was disclosed 28.5 days after the trade. Purchases filed within 15 days trailed the index by 1.96 points after a year; those filed after more than 45 days trailed by 7.55. Sales show no such pattern, and even the fastest filings don't beat the market.

The traps I hit

• Classify securities from filings, not tickers. The disclosures include funds, notes, preferred shares and warrants. I used the security type from each issuer's 12(b) registration, Form N-PORT and SEC industry codes; guessing from a ticker or a name is how an ETF ends up counted as a stock pick.
• Use the filing date, not the trade date. A follower can only act after the report is public. The trade-date version is useful only as a ceiling.
• Collapse lines to events. One stock, one filing date, one direction. Otherwise a long report with many lines outweighs a short one.
• Compare like with like. Dividends are left out of both sides: raw closes against the S&P 500 price index.
• Exclude split windows. Any one-year window crossing a recorded stock split was dropped, so no return mixes two price bases.
• Median, then cluster by date. Disclosures bunch on filing dates, so the t-statistics come from one median per date, not 17,859 independent observations.

For comparison, the typical corporate insider purchase in my insider study trailed the S&P 500 by 12.95 points after a year, so Congress's purchases lagged by less. Different samples, though, so that's a description rather than a head-to-head test.

Check one member

With the Equibles MCP server connected to Claude or ChatGPT, a prompt like "Using Equibles, list the stock purchases Nancy Pelosi's filings disclosed in the last year, with the filing lag and how each stock has done since the filing date versus the S&P 500" runs the same comparison for one person.

The full study has the disclosure-lag table, the committee test with its roster sources, the placebo details and the reproducible specification: Do Congressional Stock Trades Beat the Market?