We backtested congress trading across the whole field — copying every member, entering at the first price available after each filing became public, same rules for everyone.
153 members cleared every threshold. 1 finished ahead of the S&P 500.
The list
| # | Member | vs S&P 500 | Disclosed buys | Filled | Since |
|---|---|---|---|---|---|
| 1 | Thomas Tuberville | +9.0 | 541 | 485 | 2021 |
Thomas Tuberville, by 9.0 points over five years. That is the entire list.
This list used to have thirteen names, and every one of the other twelve was our bug
An earlier version of this article ran a table of thirteen winners, led by margins of +662 and +490 points. Before that, twenty. Both are gone, and the reason is worth more than the table was.
Three separate defects in our own engine, each found only after fixing the one before it:
- No minimum history. The board ranked one member on a single day — eight buys public the day before it was computed, showing +0.03% against the index. Eleven of 194 ranked members had under a year.
- No minimum fills. Disclosed buys and copied positions are not the same number. A member could disclose five and have one priced, and be ranked on it.
- A dammed fill queue. This is the one that invented the winners. A planned entry whose ticker had a dead price series — a company renamed years earlier, leaving a stub — blocked every entry queued behind it. Accounts froze for years, then executed the whole backlog at original prices with current cash. Years of appreciation, booked risk-free, with no market exposure in between.
That last one is why the old table's top entries looked superhuman. They were not returns. They were an accounting artefact of trades that never happened in the order the simulation thought.
Fixed, the field looks like this:
| earlier version | corrected | |
|---|---|---|
| Members ranked | 156 | 153 |
| Beat the index | 13 | 1 |
| Median vs index | −159.4 | −230.1 |
| Best margin | +662.4 | +9.0 |
Every fix moved the number the same way. That is the part we would ask you to weigh: at no point did correcting our own engine make copying Congress look better.
Restricting to members whose simulation filled at least 80% of their disclosed buys doesn't rescue it either — 1 of 105, median −186.0.
What the one winner looks like
Tuberville's margin is 9.0 points across five years, on 485 filled positions out of 541 disclosed buys. It is a real result and a thin one — comfortably inside the range a different position-sizing assumption could erase, and the filings don't disclose sizes, so that assumption is ours.
A quarter of his copied book (25.1%) was options, which a filing cannot convey — no strike, no expiry — so those are bought as the underlying. His figure describes copying the filing, not what the position did.
What it does not show
It does not show anyone acting improperly. A disclosure record that outperforms is not evidence of anything except an outperforming disclosure record. Members trade for the same reasons everyone else does, frequently through managed accounts they don't direct, and often the filings belong to a spouse.
It does not predict anything. The one member ahead is identifiable only in hindsight, by nine points. Nothing here suggests it will be the same member next year, or that there will be one — and if picking the right member in advance were possible, that would be the product, not the leaderboard.
It does not capture options. Purchases of options are modelled as purchases of the underlying, because a filing discloses no strike or expiry. Corpus-wide that's 1.6% of buys, but for Nancy Pelosi it's 30.7% and for Tommy Tuberville 25.1%.
The number that actually matters
Not the one. The other 152.
The median member trailed the S&P 500 by 230.1 percentage points. The conclusion has survived every filter and every engine fix we have thrown at it — and each one made it starker, not softer.
Following a member of Congress chosen at random has been substantially worse than buying an index fund and ignoring it.
The fine print
Entry at the first open after each filing became public; 5% of account equity per position; exits mirror the member's own disclosed sales. Runs on adjusted closes, so dividends are reinvested; taxes, market impact and commission are not modelled. Members need at least three disclosed buys, at least three years of history, and at least three positions the simulation could actually fill. Prices available for 89.5% of disclosed buys. Figures generated from the stored board as of 2026-08-13, and they move as filings land. Educational analysis, not investment advice, and not a claim about any individual's conduct.