sec-filing-forecasts

Rocket Lab, 8 May 2025

One filing from a pre-registered test of 3,329 SEC filings: when it became public, what a small Transformer forecast after it, and what the stock did next.

The filing

Rocket Lab filed its first-quarter earnings release with the SEC at 16:17 New York time and its 10-Q thirteen minutes later, at 16:30:52. The model may only act at the first market open at least 30 minutes after a filing is accepted: 09:30 the next morning.

market closed16:0020:0000:0004:0008:00Thu 8 MayFri 9 Mayearnings release filed, 16:1710-Q accepted, 16:30SEC bulk file: 20:30, four hours latemodel decides, 09:30

The SEC's own bulk download lists this 10-Q at 20:30:52, four hours late. For about half of the filings in the study it is off in the same way, and for two thirds it is wrong in some way, so every time here comes from the SEC's XBRL archive instead.

The forecast

For the next 20 trading days the network's 10–90 % range ran from −27.3 % to +35.5 %against the industrial sector fund. A forecast that only knows the last 20 days of volatility said −14.9 % to +15.6 %. The network's middle was −0.8 %: no view on the direction, just a much wider range.

−40 %−20 %0 %+20 %+40 %network, 10–90 %volatility, 5–95 %what happened, +24.4 %day 20 · +24.4 %25 days beforeopen, 9 May+5+20 days

What happened

On the first day the stock fell −8.0 % against the sector. After a week it was up +6.0 %, after twelve days +25.3 %, and after 20 trading days +24.4 %. That is inside the network's 10–90 % range, and outside even the volatility forecast's 5–95 % range of −20.7 % to +21.3 %.

In context

I picked this filing because it shows the pattern clearly. It is not a typical case, and plenty of forecasts miss. Over all 3,329 filings from 2023 to 2026, the network's loss was 3.4 % lower than the volatility forecast's, and it got the direction right only about half the time. What it learned is how wide the range should be, not which way the stock goes.