Hillscore

Methodology

Published figures for congressional trading disagree with each other by more than most of them admit, and the reasons are all methodological: execution price, measurement window, weighting and benchmark. Why most published congressional trading returns are wrong sets out where the differences come from. This page is the other half of that — exactly what this site does, so its own numbers can be checked rather than trusted.

This page explains exactly how every number on Hillscore is calculated, where the data comes from, and what its limits are. If a figure here disagrees with another tracker's, this is usually why.

Where the data comes from

Trade disclosures come from a commercial feed that parses the House Clerk's Periodic Transaction Reports and the Senate's electronic financial disclosures — the same filings members are legally required to submit under the STOCK Act. Prices are public daily closing data. Neither the filings nor the prices are proprietary: both are the kind of data a reader could pull themselves to check a figure, and the full dataset this site is built from is published for exactly that reason.

Who is tracked, and why not everyone

Every member of the House who clears the bar is tracked. The bar is ten distinct common-stock purchases in the five-year window — not party, seniority, or any judgement about who is interesting. Below ten trades an average says more about luck than about anyone's decisions.

The list is generated from the filings, not compiled by hand, so it needs no revision and cannot quietly become a popularity contest. Members join it as they build a record and leave it when their record ages out of the window.

Congress has 535 voting members and most trade rarely or not at all: about 235 House members have disclosed a single trade in five years, and 64 clear the ten-trade bar.

A bias we had, and removed

Until 27 September 2026, this site tracked members chosen for how often they traded. The list began as the thirty most active traders on a third-party site and was revised by hand five times, always on the same criterion: volume.

That put a bias into every figure published before that date. Trading frequently and trading well are different things, and selecting on the first while measuring the second quietly excludes anyone who trades rarely and does it well. Dwight Evans disclosed seventeen stock purchases in five years — far too few to reach any "most active" ranking, but comfortably enough to be scored now. Pete Sessions, Debbie Wasserman Schultz and others were absent for the same reason.

Coverage is now a rule applied to everyone: every House member with at least ten disclosed stock purchases in the five-year window, read from their own filings. It makes no judgement about how often a member trades, and the list now regenerates itself automatically.

This does not make the sample complete, and it should not be read that way. Ten purchases is still a threshold. Members who file on paper rather than electronically remain unreadable whatever they trade. Senators still come from a commercial feed, because the Senate publishes to a separate system this pipeline does not read. But the selection no longer turns on a quantity related to the thing being measured, which is the part that mattered.

Where the two chambers differ. House members are read from their own Periodic Transaction Reports, published as PDFs by the Clerk of the House. Senators file to the Senate's separate system, which this pipeline does not read, so the senators listed here come from a commercial data feed instead. That feed is also the fallback for House members whose filings we cannot read — roughly 15% of filings are scanned images with no extractable text, and a member who files that way is invisible to the parser however much they trade.

Some tracked members appear with no priced trades at all. Where that's a known gap in the source data rather than an absence of real trading, they stay listed and unranked rather than being quietly dropped — and this page is where that gap is disclosed.

A member needs at least 10 priced trades from the last five years before being ranked — the trades Composite is actually built from. Below that, one good or bad pick moves the average so far that a ranking would be misleading. Older trades still appear on a member's page, but count towards neither the score, the Evidence label, nor this threshold.

One trade per decision

When a member's own account and a spouse's or joint account buy (or sell) the same stock on the same day, the filings list them separately — often purely to split a holding for tax or account reasons. Hillscore counts them as one trade: the amounts are added together and the owners listed, such as "Self + Spouse". The same applies when one account's purchase of a stock is split across several lines on the same day. The site asks whether a member picks well, and that is one pick.

How Composite is calculated

Composite is the plain average return of every disclosed stock buy a politician has made in the last five years. Each buy contributes exactly one number: its realised return if the position has since been sold, or its return marked to today's closing price if it is still held.

There is no time weighting. A trade made nine days ago and one made 900 days ago each count once, equally, as a single win or loss. A $2,000 trade and a $2,000,000 trade also count once each — Composite measures the average pick, not money-weighted performance.

This is deliberate. It answers "was each individual decision a good one," not "how much money did they make." Congressional filings disclose dollar ranges, never exact amounts, so any money-weighted figure would be estimating from a midpoint before it started.

The 1M / 3M / 12M columns

These sit alongside Composite in the rankings and answer a different question: how had a member's positions done at fixed points after they bought? Each is the average return of their buys over the first one, three or twelve months after each buy's own purchase date. A politician's own page shows the same figures at one week and six months as well.

A position still held at a checkpoint is measured at the stock's closing price on that date. A position sold between two checkpoints counts once more, at its return on the day it was sold, in the first checkpoint after the sale — and not in any longer one. The checkpoints run one week, one, three, six and twelve months, so a buy sold at day 20 counts at its one-week close, then at its sale price in the one-month column, and nowhere after. A buy still held but too recent to have reached a checkpoint is left out until it gets there, since its result isn't known yet.

Longer columns therefore rest mostly on the positions a member kept. Hover over a figure in the rankings to see how many buys it rests on, and read the longer columns alongside that number.

These columns do not feed Composite. Think of Composite as "did this pick work out," and these as "how had it done at fixed points after they bought" — answered separately, on purpose.

On a politician's own page the same figures are split into committee-flagged buys and everything else, with the number of buys behind each one shown in brackets. That count matters: only 52 of the 60 ranked members have any flagged buys inside these windows at all, and the subset runs from a couple of hundred trades down to one. A return drawn from a single buy should not be read like one drawn from two hundred.

The buy-timing table

The buy timing page asks a narrower question than Composite does: not "was this a good pick" but "what did the stock do straight after they bought it, and was that any better than its own sector."

Only positions that have been sold appear there. An open position has no sale to measure against, and the headlines are drawn from the window between purchase and sale, which does not exist yet while it is still held.

Each row shows the stock's move at one week, one month and three months from the purchase date. Underneath each is a Δ figure: the same window's move for that stock's sector benchmark ETF, subtracted. A stock that rose 8% while its sector rose 9% shows +8.0% with Δ−1.0%, and the second number is the one that carries information — the first mostly reflects what was bought rather than when. Which ETF stands in for which sector, and how each stock was assigned to one, is published in full, and the ETF closes themselves are downloadable so the subtraction can be checked.

Consumer staples and consumer discretionary are measured against separate funds — XLP and XLY. They are both "consumer" sectors and they move quite differently, so holding Procter & Gamble against the discretionary fund was crediting or penalising it for a gap that had nothing to do with the stock.

Δ is blank where the benchmark has no price for that window. That happens for buys made before an ETF existed — XLC launched in 2018, so older trades in the communications sector have no comparison to draw.

Headlines on that page

The news beside each closed position is every headline the source returned for that ticker between the purchase date and the sale, gathered automatically. No one selects them, and no editorial judgement is applied to which appear.

They are context for what the stock did, and nothing more. They are not evidence that a member read them, acted on them, or knew anything ahead of the market — a headline published during a holding period is simply a headline published during a holding period.

The performance chart

Each politician's page carries two charts. The first plots how their disclosed positions performed week by week against the same money, on the same dates, held in each stock's sector ETF. The second shows three groups of their trades — committee-flagged, no committee overlap, and Cluster excluding mega-caps — each net of its own sector ETFs, so above zero means that group was beating its sectors. Groups with fewer than five positions are left off.

It is a time-weighted return. Each week's return is measured across the positions held right through that week and then chained onto the running total, so money arriving never moves the line — a new purchase joins from the following week. Sales are matched oldest-first, and a sold position is fixed at what it went out for. The line starts at zero on the left, so it reads as the return over the period drawn rather than carrying in gains from before it.

The obvious alternative — total value divided by total cost — is rejected for two reasons. It falls when someone buys more, because the new cost lands in the denominator immediately while the gain has not happened yet, so the line can drop sharply on a day when nothing lost value. Across seven hundred positions that is barely visible; across six it dominates the chart. And it would make the two lines incomparable, since whichever category has more money behind it would win by construction rather than by performance.

Estimated profit

Each politician's page carries an estimate of money made — gains banked on positions they have sold, plus the paper gain on those still held. The full calculation is set out on the page itself, directly beside the figures, so it can be read at the point it actually matters rather than looked up here.

Two things are worth repeating in both places. Every figure is prefixed "c." because it is an estimate: filings disclose amount bands, never exact sums, so each trade is counted at the midpoint of its band, and the totals are rounded hard instead of being given to the dollar.

And percentages lead everywhere else on this site — including the rankings, which are built on Composite — because the ratio between entry and exit price stays right even when the size attached to it is wrong. A money total has no such protection.

Why prices are the close on the trade date

STOCK Act filings disclose a transaction date, not an execution price. The actual fill price a member paid is never public. Every price here is the stock's ordinary closing price on the disclosed transaction date — the only consistent choice that doesn't require inventing an undisclosed number.

This is not presented as the politician's actual purchase price, and no figure on this site should be read as one. Trade size, similarly, is the midpoint of the disclosed range.

What "Notable" and "Cluster" mean

Every trade gets one of three committee tags:

Cluster is a separate marker. A buy is marked Cluster when at least one other tracked member of Congress bought the same stock within 30 days of it, regardless of committee membership. It records members moving on the same name at the same time, whatever the reason, so a trade can be both Notable and Cluster. Popular stocks cluster often, because many members buy them.

A third marker, Mega-cap, shows on any trade in a company capitalised above $200bn. That is the whole definition, recomputed weekly, and 82 companies currently qualify. The list is published with each company's figure and where it came from, so the call can be checked rather than trusted. Until recently this was a hand-written list of 105 tickers with no rule behind it, which had Zoetis in at $70bn and Broadcom out at ten times that.

American Depositary Shares are marked mega-cap only where the figure comes from a source that already accounts for the depositary ratio. Where it does not, they are left out entirely: a foreign issuer files its share count in ordinary shares while the price we hold is the ADR's, and one ADR can stand for any number of ordinary shares, so multiplying the two gives a number that means nothing. Computed that way Taiwan Semiconductor comes out at $11.6tn rather than $2.3tn, which is the kind of error that would quietly reshape a comparison group.

When Cluster buys are compared with the rest, mega-caps are left out of both sides: a large share of Cluster buys are mega-caps, and leaving them in would measure large-company momentum rather than members moving together.

Where that line sits changes the answer, so here is the sensitivity. With the threshold anywhere between $100bn and $250bn, Cluster buys trail non-clustered ones by between 2.3 and 3.1 percentage points and the 95% range stays below zero. Raise it to $300bn and the gap narrows to about 1.5 points with a range that crosses zero — no longer distinguishable from chance. The threshold is set at $200bn and the finding is reported as the data at that setting supports it, but a reader who would draw the line higher should know the result is not robust to that choice.

These are screening signals, not evidence of wrongdoing. Committee jurisdiction is broad and plenty of matches are coincidence. Two limitations apply specifically: only current committee membership is checked, so an older trade is measured against today's seats rather than the one held at the time; and the committee-to-sector mapping is a hand-curated judgement, not an official classification.

How a stock gets its sector. We take Yahoo Finance's sector and industry where they exist, and fall back to the industry code in the company's own SEC filings where they don't. The SEC code is only a fallback because the scheme dates from 1937, was last revised in 1987, and is self-reported: it files Airbnb under "Services-To Dwellings & Other Buildings" and puts Accenture, Visa and Uber together in one bucket called "Business Services, NEC". A handful of symbols — closed-end funds, business development companies and preferred lines — have no sector from either source, because they are holding vehicles rather than operating companies. Those are excluded from sector-relative comparisons and shown everywhere else.

Because a stock's sector determines both its benchmark and whether a trade can be flagged at all, the full classification is published: every tracked symbol, the sector applied, where that came from, and which committees claim each sector.

Appropriations is mapped at subcommittee level — Defense, Energy and Water, Homeland Security and the rest. Those subcommittees write the actual spending bills for named sectors via 302(b) allocations, which makes them narrower signals than most full committees, not broader ones. Treating the parent committee as "too broad" measured the wrong level of granularity.

Budget, Rules and Ethics remain excluded. Budget drafts the concurrent budget resolution allocating across all twenty budget functions; it does not control the 302(b) subdivisions and never reaches specific programmes, let alone companies. Because its jurisdiction spans the whole economy, a flag would fire on nearly every trade a Budget member makes and would carry no information. That is a limitation of this signal, not a claim that Budget members lack access to anything. Rules and Ethics concern procedure and member conduct, so there is no sector to map at all.

What isn't priced

Only common-stock trades are priced into any return figure. Bonds, municipal debt, fund holdings and private placements are excluded from every performance number, because no free, reliable daily close series exists for them.

Options are listed in full on the options page, and on the page of each member who trades them, with the underlying stock's move after the trade date. They carry no return of their own: the premium paid is never disclosed, so there is no cost to measure from, and the feed we receive drops the contract terms, so we cannot even say whether a member bought a call or a put. Every politician's page now also shows what they disclosed by instrument, so the share of their activity that carries a score is stated rather than implied.

Filers sometimes code preferred stock as ordinary stock, so the security's description is checked as well as its type code. Preferred shares, including depositary shares representing an interest in a preferred share, along with warrants and rights, are excluded. Alphabet's Series A preferred lines (GOOGN, GOOGM) were being scored as if they were the common stock until this was tightened.

Three kinds of holding are deliberately kept, because excluding them would remove genuine equity exposure on a technicality:

A meaningful minority of records carry a transaction type of "Other" rather than a clean buy or sell. These come from annual holdings-disclosure filings mixed into the same feed as ordinary transaction reports. Where the real direction can be recovered from the filing text, it is; where it can't, the record stays visible but is excluded from performance maths rather than guessed at.

Data-quality exclusions

A small number of filings carry transaction dates that cannot be true — a mistyped year, or a trade dated after the filing that reports it. Filings are typed by hand by congressional staff, so this is a feature of the source data rather than of any tracker.

Hillscore drops these from performance calculations rather than correcting them: when a date is impossible there is no reliable way to know which field is wrong without inventing data. Rejected records are logged and counted, not silently deleted.

Duplicate records are also removed. The source feed occasionally returns the same disclosure line twice — identical symbol, date, amounts, owner and filing status — which is a parsing artifact rather than a second trade.

How current this is

The site rebuilds daily. Coverage rotates because of a data-provider rate limit, so any single politician's history refreshes every few days rather than instantly.

Corrections

Spot a number that looks wrong? Tell us. Confirmed errors are corrected, and this page is updated whenever the method changes.

How sure are these findings?

Every comparison on the site gives its sample size, but a gap between two averages can still be chance. For the three findings the site leads on, here is how clearly each stands out, using a 95% confidence range on the three-month return against each stock's own sector. These are recalculated with the data every night, on buys from the last five years — the same rolling window as the rankings.

These ranges treat every trade as independent. They are not quite: the same member often buys related stocks around the same time, so the true ranges are somewhat wider, and a result near the edge of its range deserves that caution.

Data last updated .