Summary
The obvious rule is to buy what members of Congress buy and avoid, or short, what they sell. Across 17,859 investable common-stock disclosure events, that rule fails.
One year after a disclosure became public, the typical purchase event trailed the S&P 500 by 5.36 percentage points and the typical sale event trailed by 6.39 points. Purchases did only 1.13 points better than sales on matched disclosure dates, with a Newey-West t-statistic of 1.15. That difference is not distinguishable from zero.
The reporting delay is not what destroys the result. Starting after the transaction date, and therefore crediting a follower with information that was not public, purchases still trailed by 5.71 points and sales by 6.67. The buy-minus-sell spread was -0.02 points with a t-statistic of -0.16.
There is one qualified result. Purchase windows were better than later windows on the same stocks. Under a fully disjoint two-year shift, the paired improvement was 5.46 points with a t-statistic of 3.27. Sales showed no statistically reliable improvement. That is evidence of limited purchase timing, but not a market-beating edge: the real purchase window still lagged the S&P 500.
“Better than this stock later” and “better than the market now” are different claims. The first has evidence behind it. The second does not.
Abstract
We measure raw closing-price returns after 17,859 congressional common-equity disclosure events, aggregated from 27,790 purchase and sale rows filed between January 2020 and August 2025. The nested one-year panel covers 1,601 stocks and 1,168 disclosure dates.
An event is one stock, one filing date and one direction. Investable entry is the first market close strictly after filing. Returns are measured at 21, 63, 126 and 252 market sessions and compared with the S&P 500 price index over identical dates. We require traded entry and exit rows and exclude every window crossing an authoritatively captured stock split. Dividends are excluded from both sides.
At one year, purchases have a median S&P-relative return of -5.36% and sales -6.39%; only 42.87% of purchases and 41.30% of sales beat the index. Disclosure-date-clustered t-statistics are -2.47 and -2.73. The direct buy-minus-sell spread is 1.13 points with a t-statistic of 1.15.
A transaction-date robustness test leaves the directional spread at zero. Shifted-window placebos are more nuanced: purchases improve on their own placebo by 2.51 points under a 252-session shift and 5.46 points under a 504-session shift, with t-statistics of 2.01 and 3.27. We interpret that as timing content inside a group of stocks that still underperforms the market, not evidence for a copy-Congress strategy.
Background and research question
Congressional trading attracts two claims that are often blended together.
The governance claim is that elected officials can encounter information, influence and conflicts unavailable to ordinary investors. Whether a trade makes money is not the test for whether that conflict exists.
The investment claim is narrower: public disclosures let an investor copy purchases and treat sales as warnings. That claim requires a public entry date, a benchmark and proof that purchases behave differently from sales.
Older studies made congressional trading look unusually profitable. Ziobrowski and co-authors reported that a portfolio mimicking House purchases earned roughly 55 basis points a month over 1985–2001. Post-STOCK Act evidence is less dramatic. Belmont, Sacerdote, Sehgal and Van Hoek (2022) found no superior performance in aggregate or among senators accused of informed trading. Chen and Sacerdote (2026) extended the evidence through 2023 and concluded that congressional portfolios underperformed or, at best, matched market benchmarks.
A newer working paper by Pyun (2026) separates transaction and report dates and finds stronger predictability from the transaction date. That distinction is central to any copying strategy.
We ask whether purchases or sales predict returns after disclosure, whether purchases beat sales on the same dates, whether an advantage exists at the private transaction date, and whether the real window beats a shifted window on the same stock.
The disclosure clock
The STOCK Act creates a delayed public record, not real-time reporting.
The House Committee on Ethics says a qualifying transaction must be disclosed by the earlier of 30 days after awareness or 45 days after the transaction. The Senate Select Committee on Ethics states the same outer limit. Covered transactions over $1,000 generally require a Periodic Transaction Report.
That creates two dates:
- The transaction date, when the member, spouse or dependent account traded.
- The filing date, when the record became public.
Our main specification enters at the first market close strictly after the filing date. A transaction-date backtest is reported only as an impossible-foresight robustness test.
Data and sample construction
What counts as a stock
The disclosure corpus also contains funds, notes, preferred shares, warrants and rights. We never classify them by ticker or name patterns.
We use the security type from the issuer’s filed 12(b) title, SEC industry codes and Form N-PORT filings. Preferred shares, debt securities, warrants, rights, registered funds and authoritatively identified exchange-traded products are excluded. Unknown security types remain eligible because exclusion requires positive evidence.
From rows to events
The event is one stock, one filing date and one direction. Multiple same-direction lines or filers become one event. A stock carrying both purchases and sales on one date contributes one event to each direction.
| Measure | Value |
|---|---|
| Disclosure events | 17,859 |
| Underlying trade rows | 27,790 |
| Purchase events | 8,444 |
| Sale events | 9,415 |
| Distinct stocks | 1,601 |
| Distinct filing dates | 1,168 |
| Filing-date range | 2 Jan 2020 to 18 Aug 2025 |
| Price-data cut | 21 Aug 2026 |
Events are excluded when the one-year horizon has not matured, the exact primary listing lacks a traded entry or exit row, or the window crosses a captured split.
Prices, benchmark and statistics
We use raw closes on the exact current primary listing, with positive volume at entry and exit. Any one-year window crossing a captured split is excluded, and the same clean panel is used for shorter horizons.
Dividends are excluded. The benchmark is the S&P 500 price index, so both sides are price returns. Horizons are 21, 63, 126 and 252 S&P 500 market sessions.
“Relative return” means the stock’s raw closing-price return minus the S&P 500 price-index return over identical dates. It is not a factor-model alpha.
The pooled median is primary. For inference, we collapse each direction to one median per filing date and compute a Newey-West t-statistic, using a lag equal to the fourth root of the date count. The direct directional test keeps dates carrying both directions and subtracts the sale median from the purchase median.
We also shift each one-year window forward by 252 or 504 sessions. Real and placebo windows must both be complete, traded and split-free.
Main result: purchases and sales look alike
| Direction | Horizon | Events | Stock return | S&P 500 return | Relative return | Beat S&P | t |
|---|---|---|---|---|---|---|---|
| Purchase | 1 month | 8,443 | 1.50% | 2.42% | -0.52% | 47.01% | 0.46 |
| Purchase | 3 months | 8,443 | 3.40% | 6.07% | -1.79% | 45.02% | -2.27 |
| Purchase | 6 months | 8,442 | 6.28% | 10.78% | -2.63% | 44.14% | -1.71 |
| Purchase | 1 year | 8,444 | 10.95% | 19.73% | -5.36% | 42.87% | -2.47 |
| Sale | 1 month | 9,414 | 1.34% | 2.42% | -0.54% | 46.98% | -0.77 |
| Sale | 3 months | 9,412 | 3.48% | 6.16% | -2.22% | 43.83% | -1.46 |
| Sale | 6 months | 9,412 | 5.95% | 11.71% | -4.06% | 41.79% | -2.41 |
| Sale | 1 year | 9,415 | 11.30% | 21.50% | -6.39% | 41.30% | -2.73 |
Both directions rise in absolute terms. Both trail the index.
“Stocks bought by Congress gained 10.95%” is true and incomplete. The market gained much more over the same windows.
A sale is not a mirror-image purchase. People sell to rebalance, pay taxes, diversify or fund spending. If sales carried a clean bearish signal, sold stocks should perform materially worse than bought stocks. They do not.
Congress versus corporate insiders
Congressional purchases performed better than corporate-insider purchases in our two studies, but they did not beat the market. The typical congressional purchase lagged the S&P 500 by 5.36 points after one year, compared with 12.95 points for 47,458 corporate-insider open-market purchases. The congressional deficit was 7.59 points smaller.
The raw-return figures point the same way: 10.95% for the typical congressional purchase versus 2.36% for the typical insider purchase. This is a descriptive cross-study comparison, not a controlled head-to-head test. The studies cover similar dates but differ in issuer mix, disclosure systems, event construction and corporate-action exclusions. It does not establish superior congressional skill; it shows that congressional purchases underperformed the index by less in these two samples.
Do members do better in sectors their committees oversee?
A stricter version of the information-advantage argument is that a member might do better when buying companies in an industry their committees oversee. We tested that possibility separately.
The original analysis merges everyone buying the same stock on the same filing date. This extension instead uses one member, one stock and one filing date as the event, because committee exposure belongs to the member. It keeps the same eligible-security rules, public entry date, one-year raw-close return, S&P 500 price-index benchmark and split exclusion. After requiring a usable official assignment and a complete return window, the panel contains 8,263 purchase events.
We assigned each transaction to its Congress, then joined the member to the official House rosters for the 116th, 117th, 118th and 119th Congresses or the Senate's official assignment publications. A purchase counted as committee-aligned only when the issuer's SEC industry code fell inside a narrow, written mapping from committee jurisdiction to agriculture and food, defense and aerospace, financial services, energy and natural resources, health, transportation or telecommunications. Industry codes come from the SEC's official SIC list, not from ticker or company-name guesses. “Other purchase” therefore means no strict match under this map, not proof that the committee had no possible connection to the company.
| Purchase group | Events | Members | Stock return | S&P 500 return | Relative return | Beat S&P |
|---|---|---|---|---|---|---|
| Committee-aligned | 436 | 58 | 16.20% | 20.50% | -4.82% | 43.81% |
| Other purchases | 7,827 | 153 | 10.96% | 20.11% | -5.22% | 43.29% |
The raw comparison looks favorable: aligned purchases gained 5.24 points more. After accounting for the market over the same windows, however, their median advantage shrinks to 0.40 points, and their S&P-beating rate is only 0.52 points higher. Neither group beats the index.
Composition checks do not rescue the result. On the 182 filing dates carrying both groups, the median aligned-minus-other gap is -2.01 points, with a Newey-West t-statistic of -0.94. Among the 57 members who made both aligned and other purchases, the median within-member gap is -1.65 points; the cross-sectional t-statistic is 0.16. Those estimates disagree in sign and are statistically weak.
This is not evidence that committee assignments create an investment edge. The higher raw return is largely absorbed by the benchmark and does not survive matched-date or within-member comparison. The test also has limits: the official rosters are Congress-level snapshots rather than a daily appointment ledger, committee jurisdictions are broader than SIC buckets, and disclosures can cover a spouse or dependent. The sector subsamples are too small to support claims that one committee is uniquely good at trading.
Buys do not beat sells
| Horizon | Matched dates | Median buy-minus-sell spread | t |
|---|---|---|---|
| 1 month | 760 | 0.00 pts | 1.03 |
| 3 months | 760 | -0.06 pts | 0.02 |
| 6 months | 760 | 0.00 pts | 1.08 |
| 1 year | 760 | 1.13 pts | 1.15 |
There is no directional separation at any horizon. The one-year spread is positive but statistically weak.
Perfect foresight does not rescue it
Starting after the transaction date gives a follower information that was not public.
| Direction | Events | One-year relative return | Date-level mean | t |
|---|---|---|---|---|
| Purchase | 9,990 | -5.71% | -4.73% | -6.05 |
| Sale | 10,264 | -6.67% | -4.11% | -4.36 |
| Purchase minus sale | 1,197 dates | -0.02 pts | -0.17 pts | -0.16 |
The delay did not erase a profitable copying signal. Even at the private date, both directions underperform and their spread is zero.
This cannot prove that no isolated informed trade exists. It shows that “Congress bought it” is not, by itself, a market-beating rule.
Disclosure delay
The median purchase is filed 28.5 calendar days after the transaction and the median sale after 29 days. Recorded lags above 45 days appear in 17.75% of purchases and 15.40% of sales.
Those observations are not findings of legal violations. The store can include amendments, corrected or backfilled records and filing-date revisions. We also do not observe when a filer was notified.
| Direction | Recorded lag | Events | One-year relative return | Beat S&P |
|---|---|---|---|---|
| Purchase | 0–15 days | 1,749 | -1.96% | 46.88% |
| Purchase | 16–30 days | 3,167 | -5.59% | 42.56% |
| Purchase | 31–45 days | 2,029 | -6.23% | 41.99% |
| Purchase | More than 45 days | 1,499 | -7.55% | 40.03% |
| Sale | 0–15 days | 1,714 | -5.55% | 42.12% |
| Sale | 16–30 days | 3,497 | -5.29% | 42.55% |
| Sale | 31–45 days | 2,754 | -9.48% | 37.58% |
| Sale | More than 45 days | 1,450 | -4.59% | 44.34% |
Faster purchase disclosures are followed by less underperformance. Sales do not show the same monotonic relationship, and eliminating the lag still does not create market-beating returns.
The placebo: some purchase timing survives
| Direction | Shift | Events | Real relative return | Placebo relative return | Paired improvement | t |
|---|---|---|---|---|---|---|
| Purchase | 252 sessions | 6,139 | -4.54% | -6.10% | 2.51 pts | 2.01 |
| Sale | 252 sessions | 6,940 | -5.78% | -6.28% | 1.39 pts | 0.77 |
| Purchase | 504 sessions | 5,351 | -3.97% | -7.63% | 5.46 pts | 3.27 |
| Sale | 504 sessions | 5,890 | -4.94% | -8.05% | 4.43 pts | 0.78 |
Purchase windows are better than the same stocks’ later windows. But every real purchase number remains negative.
Congress appears to buy these stocks at a relatively better time than another date in their later history while still choosing stocks or periods that trail the S&P 500. That is timing content without market-beating selection.
How this fits the literature
The broad conclusion matches Belmont and co-authors and the newer NBER work by Chen and Sacerdote: legislators do not display aggregate stock-picking skill after a market comparison.
The purchase placebo is closer to Pyun’s disclosure-timing result. We find information in purchase dates relative to the same stock’s later path, but not positive S&P-relative performance.
Older pre-STOCK Act findings study different eras and often use calendar-time portfolios rather than equal-weighted disclosure events. Our result is specifically about recent public disclosures and what a present-day follower can act on.
Limitations
No risk adjustment. The S&P 500 comparison does not control for beta, size, value, momentum, industry or volatility.
No dividends. We compare raw closes with the S&P 500 price index.
Current primary listings. A historical share class or ticker can be unavailable even when the issuer is resolved.
Split-free windows. Excluding split-crossing windows avoids mixed price bases but changes the sample.
Survivorship. Delisted failures are more likely to be absent than surviving winners, which can bias stock returns upward.
No transaction price. Disclosures give dates and value ranges, not an executable public price.
Direction is not intent. Sales can reflect taxes, diversification or liquidity. Purchases can come from spouses or joint accounts.
Event aggregation. Combining lines prevents report length from setting the weight but discards size and member variation.
One period. January 2020 to August 2025 includes the pandemic, the 2022 drawdown and a concentrated large-cap rally.
Aggregate evidence can hide isolated cases. A null average does not prove every trade was uninformed.
Reproducible specification
Sample: purchase and sale rows linked to authoritative listed common equity, filed from 2 January 2020. Exclude preferred shares, debt, warrants, rights, registered funds and exchange-traded products using filed security type, Form N-PORT and SEC industry classification.
Event: one CommonStock, filing date and direction.
Investable entry: first S&P 500 market-calendar close strictly after filing.
Perfect-foresight entry: first market close strictly after the transaction date.
Horizons: 21, 63, 126 and 252 sessions.
Prices: exact primary listing, raw close, positive volume. Exclude a one-year event when a captured split is effective after entry and on or before exit.
Benchmark: S&P 500 price-index return over identical dates.
Statistic: pooled median event-relative return; Newey-West t-statistic on filing-date medians.
Directional test: purchase-date median minus sale-date median on dates carrying both.
Placebo: shift entry by 252 or 504 sessions and require both one-year windows to be complete, traded and split-free.
Final panel: 17,859 events from 27,790 rows, 1,601 stocks and 1,168 filing dates. Price and benchmark cut: 21 August 2026.
Conclusion
Three findings survive.
First, the public copying rule fails. Purchases trail the S&P 500 by 5.36 points over one year and sales by 6.39. Fewer than 43% of either direction beats the index.
Second, purchases are not meaningfully different from sales. The one-year spread is 1.13 points with a t-statistic of 1.15. Perfect foresight still produces no directional separation.
Third, purchase timing is not entirely empty. Purchases perform better than the same stocks do later. But “less bad than this stock later” is not “better than the market now.”
Congressional disclosures are useful evidence about conflicts, holdings and behavior. They are not a substitute for an investment thesis.
This note is for informational and educational purposes only. It is not investment advice.
References
Belmont, W., Sacerdote, B., Sehgal, R., and Van Hoek, I. (2022). “Do Senators and House Members Beat the Stock Market? Evidence from the STOCK Act.” Journal of Public Economics, 207, 104602. https://doi.org/10.1016/j.jpubeco.2022.104602
Chen, H., and Sacerdote, B. (2026). “Capital in the Capitol: Congressional Trades Resemble Uninformed Retail Trading.” NBER Working Paper 35041. https://doi.org/10.3386/w35041
House Committee on Ethics. Financial Disclosure Instruction Guide. https://ethics.house.gov/wp-content/uploads/2024/11/FDInstructionGuide_current_2023.pdf
Newey, W. K., and West, K. D. (1987). “A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix.” Econometrica, 55(3), 703–708.
Pyun, C. (2026). “Congressional Trading, Informational Advantages, and Disclosure Timing.” SSRN. https://doi.org/10.2139/ssrn.5295880
Senate Select Committee on Ethics. Financial Disclosure. https://www.ethics.senate.gov/public/index.cfm/financialdisclosure
Ziobrowski, A. J., Boyd, J. W., Cheng, P., and Ziobrowski, B. J. (2011). “Abnormal Returns from the Common Stock Investments of Members of the U.S. House of Representatives.” Business and Politics, 13(1), 1–22.