Model forecasts
What it shows
A summary of current QWave forecasts for the coins in the basket: P(up), median change and the P10-P90 band for the chosen horizon, with direction hit rate and interval coverage and their settled count below.
How to read it
Start with the P10-P90 band and its width, then check interval coverage with N: if it stays around 80%, the band can be trusted as a measure of uncertainty. Read P(up) carefully, because over short horizons it usually sits close to 50%.
Parameters
- Horizon (1 day, 7 days, 30 days)
- How far ahead the forecast reaches. Each horizon has its own forecast, settlement date and statistics, so 1 day and 30 days can look different without contradiction.
- P(up)
- The share of simulated paths that end the horizon above today's price. 50% is a coin flip; over a few days the value almost always sits near 50%, and that is the honest answer, not a fault.
- Neutral reading: From 45% to 55%: the model sees no edge on either side.
- Median (P50)
- The middle price from the simulation: half of the paths end lower, half higher. A median close to today's price is a normal result.
- 80% forecast interval (P10-P90)
- The price band at the end of the horizon: in 80 cases out of 100 the price should land inside, 10 fall below and 10 above. A wide band means high uncertainty. This is the most important number on the card.
- Direction hit rate (N)
- How often the side indicated by P(up) matched the real price change. N is the number of settled forecasts. Below the minimum sample we write “too little data” instead of a percentage.
- Negative reading: A result around 50% is normal for direction over a few days. The model is judged mainly by interval coverage.
- 80% interval coverage (N)
- How often the real price actually landed inside the P10-P90 band. An honest model scores about 80% here: clearly less means the band is too narrow, clearly more that it is too wide.
- Positive reading: From 75% to 85%: the band tells the truth about uncertainty.
- Historical test or live results
- Until there are enough settled live forecasts, the statistics come from the historical test: forecasts for past days computed only from data available before each day. The two sources are never mixed, and the label always says which one you see.