We Backtested Copying Congress Stock Trades

Copying politicians’ stock trades is one of the most repeated money ideas on the internet. The pitch practically writes itself. They have information you do not, they have to tell you what they bought, so just buy the same thing.

We wanted to know if it works. Not as a vibe, as a number. So we pulled every Periodic Transaction Report the House filed between January 2021 and May 2026 straight from the Clerk’s public archive, parsed 22,810 transactions out of the PDFs, and tested 8,329 stock purchases against the S&P 500.

The short version: it did not work. And the reason it did not work is not the reason almost everyone says.

Do congress stock trades beat the market?

No. Buying at the first price an ordinary person could get after a filing went public returned 4.07% over the next 90 days. The S&P 500 returned 4.79% over those exact same windows. That is a gap of roughly negative 0.71% per trade, and once we account for the fact that many trades get filed on the same day and share that day’s market move, the result is not distinguishable from zero.

Copying Congress was not a disaster. It was a wash, with more work and more risk.

How we set this up

Every quoted politician return you see is measured from the day they traded. You cannot buy on that day, because you do not know yet. You can only buy after the filing becomes public.

So we used the filing date, and we were strict about it. Each PDF actually contains two dates, and this trips up a lot of analysis. There is a “notification date,” which is when the member was told about the trade, and there is the date the report itself was filed with the Clerk. Only the second one is public. We used the filing date, then entered at the opening price of the next trading day, because filings post at all hours and plenty land after the market closes.

We compared each trade to SPY over identical entry and exit dates. That matters more than it sounds. It means market timing cancels out completely, so whatever is left is purely a question of whether the stock they picked did better than the index.

We also charged 10 basis points per round trip for spread and slippage, and we tested at both a 30-day and a 90-day hold.

	90-day hold	30-day hold
copying the congressman	+4.07%	+1.54%
S&P 500, same windows	+4.79%	+2.05%
difference	−0.71%	−0.51%
beat the index on	45.0% of trades	45.5% of trades

At 30 days the shortfall is small but statistically real. At 90 days it sits just inside the noise. Neither one is an edge.

The win rate is the part people miss

Here is the number that reframed this for us. Over the same 90-day windows, the congressional picks finished positive 58.3% of the time. SPY finished positive 79.9% of the time.

Think about what you are signing up for. You take on all the risk of individual companies, you accept a hit rate more than 20 points worse, and in exchange you get a slightly lower average return. The median copied trade underperformed the index by 1.81%.

The averages look close because a small number of picks did extremely well. Strip out the ten best trades from more than eight thousand and the average excess return drops from negative 0.71% to negative 0.92%. Strip out the best hundred and it falls to negative 1.65%. This is the same shape as a lot of stock picking. A few winners do the heavy lifting and you have no way of knowing in advance which ones they’ll be.

If concentration risk is the thing you’re trying to avoid, we wrote about why spreading out beats getting clever in our three fund portfolio breakdown.

The number that almost fooled us

Our first pass produced something exciting. A portfolio built on these trades compounded at 19.29% a year while the S&P 500 returned 15.27% over the same stretch. Four points a year of free outperformance.

It was not real, and the way it was not real is worth understanding, because this exact mistake is everywhere in copy-trading content.

Congressional filings do not arrive evenly across time. They cluster. And it turned out the stretches of calendar those trades occupied were better than average for the market as a whole. The average SPY return across the specific windows we were invested in worked out to roughly 20.6% a year, well above the 15.27% the index actually delivered over the full period. We were not measuring good stock picking. We were measuring the fact that the strategy happened to be invested during good months.

The test that settles it is simple. Run the same portfolio, same trades, same entry and exit dates, but buy SPY every time instead of the congressman’s pick.

identical trades and dates	annual return
copy the congressman’s stock	19.29%
buy SPY instead	19.68%

The schedule was doing the work. The stock selection cost money.

We think a lot of “copying Congress beats the market” claims are this exact artifact, because comparing a strategy’s return to a buy-and-hold benchmark over different deployment periods is the easiest mistake in backtesting to make and one of the hardest to notice.

What about only following the fast filers?

This is the most popular fix, and it is the one part of the theory that partly held up.

Trades sorted by how quickly they were disclosed do line up in the right order:

filed within	trades	vs S&P 500	beat the index
7 days	724	+0.71%	49.0%
8 to 14 days	877	−0.05%	48.7%
15 to 30 days	3,297	−0.22%	45.9%
31 to 45 days	2,099	−0.86%	44.3%
more than 45 days	1,332	−2.92%	39.6%

Slow filings are clearly worse. That gradient is real and it is the strongest pattern in the data.

But the fast bucket does not survive a significance test. That +0.71% carries a confidence interval running from negative 0.82% to positive 2.24%. We cannot rule out zero, and we cannot rule out a small loss. Its hit rate is 49%, which is below a coin flip. A positive average combined with a losing hit rate tells you a few outliers are doing everything.

There’s also a practical problem. Only 11.9% of trades in our sample were filed within 10 days. The median lag was 28 days. Building a strategy around fast filers means waiting around for about one trade in eight, and the payoff for that patience is a number we cannot distinguish from nothing.

So where does the edge actually go?

Here is where it gets interesting, and where we ended up disagreeing with the standard story.

The usual explanation is that the disclosure delay eats the edge. The trade was good, but by the time you can see it, the move already happened.

We tested that directly by pretending we had a time machine. What if you could enter at the member’s own trade date, the price they got, which no copier can ever actually access?

That version underperformed the index too, by about 0.36%, with a confidence interval spanning zero.

Read that again, because it undoes the whole premise. Even buying at the exact moment the politician bought, you would not have beaten just owning the index. And at the 30-day hold, entering at the public filing date actually did slightly better than entering at the trade date, which is what a non-effect looks like when you poke it.

The lag is not the reason copying fails. The picks are.

What we would actually do

We are not going to tell anyone that tracking congressional trades is pointless. It’s genuinely interesting information, it says something real about disclosure and accountability, and if you want to follow it we’ve written a guide to how to follow congress stock trades using the free tools.

What we would not do is reorganize a portfolio around it. Eight thousand real trades over five and a half years produced no measurable advantage over buying the index and leaving it alone, and the version that looked like an advantage turned out to be a calendar artifact.

If you want to see what the boring alternative does over the same kind of timeframe, run the numbers yourself in our compound interest calculator. The gap between a strategy that works and one that merely sounds like it works usually shows up there.

One honest caveat on our own numbers. About 6.9% of the buys we found had no usable price history, mostly companies that got delisted, acquired, or renamed. Those skew toward bad outcomes, so if anything our results are slightly kinder to copying than reality was. We also only covered the House. Senate filings live on a separate system and are not in this sample.