I finally sat down and read the actual academic literature on congressional trading a while back, mostly because I was tired of arguing about it from vibes. The online debate has two settings. Either members of Congress are running the greatest insider fund in history, or the whole thing is survivorship bias and cherry-picked screenshots. The papers say something quieter than both, and if you are thinking about building any kind of copy strategy off disclosure feeds, knowing what the evidence actually shows will save you from some expensive assumptions.
Where the legend comes from
The headline numbers everyone quotes trace back to Alan Ziobrowski and his coauthors. Their 2004 study looked at common stock trades by US senators during the 1990s, and the result was genuinely striking. Stocks senators bought went on to beat the market by a wide margin, historically somewhere around ten to twelve percent a year of abnormal return on the purchase side, while stocks they sold tended to underperform after the sale. That combination, buys that win and sells that dodge losses, is the classic footprint of informed trading. Retail investors show nothing like it, and even corporate insiders, who legally trade with deep knowledge of their own companies, historically showed weaker timing than that Senate sample.
A follow-up paper on the House found the same pattern at roughly half the strength, which the authors read as representatives having less power and less access than senators. That dose-response relationship made the story more convincing, since more power tracking with more edge is exactly what you would predict if information were the mechanism.
Two caveats before you extrapolate. The samples were small and dominated by a minority of members who traded actively, so a handful of skilled or connected traders could drive the averages. And the data came from an era when disclosure meant annual paper filings that almost nobody read. Whatever edge existed was operating in the dark. Nobody was copying these trades because in practical terms nobody could see them, and that detail matters for everything that came after.
Then the evidence got messier
Later work pushed back hard. Eggers and Hainmueller published a study memorably titled Capitol Losses that examined congressional portfolios in the mid 2000s and found no outperformance at all. If anything the average member lagged a boring index fund, with a heavy tilt toward local companies and familiar names. Their conclusion was that Congress in aggregate is a mediocre investor, and that the earlier results may have been specific to a small group of active traders in an unusual market period. Reconciling the two camps is hard because they use different windows, different weighting schemes, and different definitions of abnormal return. My honest read is that the informed trading in the older data was real but concentrated, a minority of members doing most of the damage while the median member picked stocks about as well as your uncle does.
Then the STOCK Act arrived in 2012 and changed the information environment. It made explicit that trading on material nonpublic information obtained through congressional work is illegal, and it required trades above a small dollar threshold to be disclosed within 45 days instead of buried in an annual filing. Most research on the period after 2012 finds aggregate abnormal returns shrinking toward zero, and before-and-after comparisons generally agree that the average congressional trade stopped looking special once it became visible within weeks instead of a year.
There are two readings of that shrinkage, and both are probably partly true. Disclosure and media scrutiny deterred the most aggressive behavior, and whatever remained migrated to places that are harder to measure, spouse accounts, options structures, private placements, timing that resists clean attribution. Either way, the observable edge weakened at almost exactly the moment it became easy to observe, and that relationship between visibility and returns is the most important calibration point for anyone building a strategy on these feeds.
The subsets that still carry signal
Averages hide the interesting stuff, and the post-STOCK Act literature is mostly a story about subsets. A few keep showing up across different papers and methods.
- Purchases over sales. Members sell for a hundred reasons, tuition, a house, diversification, optics before a tough vote. They buy for roughly one reason. Across nearly every study, whatever excess return exists lives on the buy side, and sales carry little usable information.
- Committee-relevant trades. Purchases of companies that fall under a member's committee oversight have historically performed better than their unrelated trades in several samples. This is the closest thing to a smoking gun in the modern data, because committee work is exactly where nonpublic information would flow.
- Seniority and leadership. Power correlates with performance in multiple samples, which rhymes with the original gap between the Senate and House results.
- Clustered buys. When several members buy the same name inside a short window, the historical signal is stronger than for any single trade, for the same reason that clustered corporate insider buying beats a lone purchase.
One more finding worth carrying around. Several papers document that prices react on the disclosure date itself, separate from anything that happened on the trade date. Some of that is copy flow from retail traders and funds watching the filings, which means part of the return you see after a disclosure is other followers pushing the price, and that kind of flow-driven pop tends to fade rather than continue. Distinguishing information from attention is the hardest part of trading this data.
Calibrating before you copy anything
If I compress all of this into practice, it comes out as four rules.
First, you are trading disclosures, not trades. The gap between the transaction and the filing can legally run to 45 days and often does. Any backtest that enters on the transaction date is measuring returns you could never have captured, and this is the single most common failure mode I see in copy strategies. It routinely turns a fantasy double-digit edge into something between small and zero. Key every backtest to the filing timestamp or do not bother running it.
Second, filter hard. Buys only. Weight committee relevance if you can map tickers to oversight areas. Prefer clusters of members over a single filer, and give more weight to senior members than to freshmen. A raw feed of every congressional trade is mostly noise with occasional theater, and the studies that find nothing are usually the ones measuring that raw feed.
Third, size for a modest edge. A realistic expectation for a well-filtered strategy in the post-STOCK Act era is low single digits of annualized excess return before costs, with wide error bars around that. If your position sizing assumes the old Senate numbers you will be disappointed, and if you are paying spread and slippage on thinly traded names the edge can vanish entirely.
Fourth, know what you are actually harvesting. Some of the return is real information leaking slowly, some is other copiers arriving after you, and the mix shifts trade by trade. Watching how a name behaves in the first few sessions after the filing tells you which one you got. This is roughly why we built the political trading feed in Blockcircle around filing timestamps with committee mappings attached, because without those two fields the data reads as entertainment rather than signal.
My read after all of it is that a modest, filterable edge survives, and it rewards the people who treat the footnotes as the product. Hold your expectations to low single digits, trade the buys, respect the lag, and test only against dates you could have acted on. That alone puts you ahead of most of the people copying screenshots.