How well MEQIRI warns

How the number is made

MEQIRI recalculated the score of every stock in the US and Europe over 20 years, using only the statements published on each day. Where a statement has no publication date, an annual report counts only 120 days after the fiscal year end, a quarter after 90. For every month and stock we know how it stood and whether it crashed afterwards. A disclosed calculation learns from this which features went along with crashes more often in the past: the eight score categories, the price change and volatility of the past 12 months, the distance to the 52-week high, company size and whether it is a bank or insurer or belongs to a cyclical sector. For the test above it learned its weights only from data up to 2014 and was tested on the years after. The test is not fully blind: the score formula and the choice of features date from 2026, when the outcome of those years was already known. The number on the stock pages comes from the same calculation, learned from 605,040 cases.

What the number cannot do

It is not a forecast for a single stock and not a guarantee. It says how often stocks with similar features crashed in the past. No calculation predicts sudden market-wide crashes like March 2020, and a large share of the crashes fell in 2020 and 2022. It is measured on month-end closes without dividends. Lows between two month ends do not count, with daily prices there would be more crashes. Companies that no longer exist are missing from the data, such as Silicon Valley Bank or Credit Suisse. This probably understates crashes at high risk most. For some US companies our data provider supplies restated figures for earlier years. Without the affected cases the result barely changes. Among the 100 largest stocks the calculation separates crashes clearly better than the price swings of the last twelve months alone; for smaller stocks the swings alone do about as well.