Experiment
Signals after filings
After each material filing of a tracked company, the world engine runs on the stock's history, telling months with material filings from months without, and says how likely the stock is to touch certain levels. It is not investment advice.
The “Why” text is AI-generated and checked by automatic rules.
Loading…
Order = the composite score: the mean difference between the probability of the stock touching +5/10/15 % and of touching −5/10/15 %, over 1, 3, 6 and 12 months (from −1 to +1). The two probabilities do not add up to 100 %: a stock can touch both +5 % and −5 % in the same period, so 53 % and 54 % mean «it moves», not «where». «Agreement» = how many horizon × threshold cells lean the same way.
After each material filing of a tracked company, the world engine runs on the stock's history, telling months with material filings from months without, and says how likely the stock is to touch certain levels. It is not investment advice.
Methodology
- The world engine counts, on the stock's history, what followed the months in which the issuer published a material filing (results, contract, M&A, management, capital increase, litigation…), against the months without one, and conditions the price-move probability on that. Classes are not separated: a month with any material filing counts the same. The history starts at the first classed filing (at the SEC, August 2004); with fewer than 36 months of history the signal is not published.
- “Touch” means the price crosses the level at least once up to month k, not that it ends there. The probability of a price at the horizon is not computed.
- Next to every probability sit the random walk (p_rw: no drift, at the current volatility) and how many months of the history held filings (k of n). Without them we do not publish the probability.
- The dominant direction is the one with the higher probability at the 5% move; the “1,000 → X” sum uses the largest move in that direction with probability of at least 50%.
- The “trend + seasonality” band is a plain statistical baseline, unrelated to the engine: it starts from today's price and adds the average monthly drift and the shrunk month-of-year seasonality of the stock's last 36 months (5 years of history); p10–p90 widen with the horizon.
- The “Why” text is written by a language model from the facts of the filing and the signal, passes automatic checks and cannot change any figure.
- It is an experiment: the past score below is public, including when the engine loses to the random walk. On the history of the 40 US companies, telling filing months apart has not so far improved the probabilities over the engine without it.
Past score (walk-forward)
The past score is not published yet.
olivLaw experiment. Model-computed probabilities; not investment advice. AI-assisted content.