Does the score work? — QuantCapitol

Honesty Does the score work?

A ranking is only worth anything if it's honest about whether it works. So we test the Interest Score against real outcomes — the price move relative to the S&P 500 over the 63 trading days after each buy. Because the score is ex-ante (built only from what was knowable at filing time), this is a fair test: the score never got to see these outcomes.

The honest headline: the score does not predict returns. We'll put the conclusion first. On our own data, a higher Interest Score does not reliably lead to a better return — the relationship is essentially flat. That's the truth, and it's more useful than a comforting lie. The score is an attention filter (which trades are large, unusual, clustered, or inside the trader's jurisdiction — worth a look), not a return forecast. The evidence is below, unedited.

Hypothetical backtest, not investment advice. Returns are measured against historical prices over a fixed window; disclosures lag the trade by up to 45 days and report dollar ranges. This evaluates the score — it is not a prediction or a recommendation, and past patterns don't predict future results.

First, the honest baseline — with its uncertainty

-0.13%

avg excess return
vs S&P 500

Across 6,520 congressional buys — measured from the filing date, i.e. when a follower could actually have acted — the average 63-day return beat the S&P 500 by -0.13%, with a 95% confidence interval of [-0.50%, +0.25%]. That interval straddles zero — so the average edge is statistically indistinguishable from nothing. This is the honest, look-ahead-free verdict: following Congress on average does not reliably beat the market.

Why "from the filing date"? A member's trade isn't public until it's disclosed — up to 45 days later. Measuring from the transaction date would credit a return no follower could have captured (look-ahead bias). We measure from the filing date — what you could actually have done. For reference, the member's own trade (from the transaction date) averaged -0.69% — the gap between the two is the value of being the insider versus the follower.

Does a higher score mean a different outcome?

If the Interest Score predicted returns, higher-scoring buys should show consistently higher excess returns — a clean upward staircase across the bands. Below is the average excess return per score band, each with its 95% confidence interval, because an average without its spread is meaningless. The relationship is not monotone (rank correlation 0.2, i.e. essentially no consistent ordering). One middle band may stand out, but the highest-scoring band — the trades we flag as most noteworthy — does not reliably show the best returns. This is the signature of an attention filter, not a return predictor — which is exactly what the score is built to be. We report it plainly rather than dressing a jagged pattern up as "it works".

Interest Score band	Avg excess return vs S&P 500	95% confidence interval	Significant?	Trades
0–20	-0.93%	[-1.7, -0.2]	significant	1531
20–40	+0.03%	[-0.8, +0.9]	not sig.	1188
40–60	+2.02%	[+0.8, +3.2]	significant	755
60–80	-0.62%	[-1.4, +0.2]	not sig.	1501
80–100	-0.02%	[-0.8, +0.7]	not sig.	1545

Method: every congressional BUY with price data is measured 63 trading days out from its filing date (look-ahead-free), as its return minus the S&P 500's over the same window, then grouped by the score it received at filing time. The 95% confidence interval uses the t-distribution on each band's sample; "significant" means the interval excludes zero. Bands marked ⚠ have fewer than 30 trades and are statistically unstable. This is a hypothetical backtest on public data, not actual performance.