Why 1-Day, 5-Day, and 20-Day Forecasts Behave So Differently
2026-11-12 ยท hedgewing.ai Research
A common first reaction to a multi-horizon forecast is confusion: the one-day call leans down, the five-day call leans up, and the twenty-day call is nearly neutral. Which one is right? Usually all of them, because they are answering different questions. The statistics of returns change with the horizon, and so does what a model can and cannot know.
Noise dominates short horizons
Daily stock returns are almost entirely noise. The expected move over one day is a few hundredths of a percent, while the typical move is one or two percent, so the signal-to-noise ratio is tiny. A one-day forecast with 55% confidence is doing real work, but it will be wrong often, and a run of five wrong calls in a row is unremarkable. The upside of the short horizon is that it resolves quickly, which means calibration can be checked on many outcomes and the model gets fast feedback.
Signal accumulates at medium horizons
Over five to ten trading days, persistent effects such as momentum, mean reversion after overextension, and post-earnings drift have room to express themselves, and the day-to-day noise partly averages out. This is where features built from volatility, trend, and volume structure tend to carry the most information, and where the four models' different views of the sequence are most complementary. Confidence at these horizons is often higher than at one day, not because the models are more certain about any specific day but because the cumulative move is more predictable than any single step.
Events dominate long horizons
At twenty days, the forecast window is likely to contain something the price history cannot anticipate: an earnings report, a macro release, a sector rotation. The expected-return signal is larger in absolute terms but so is the variance, and the distribution of outcomes grows fat tails. A well-behaved model responds by widening its range and pulling confidence toward neutral when a known event sits inside the window. A model whose twenty-day confidence is as sharp as its one-day confidence is not accounting for what it does not know.
Matching horizon to holding period
The practical rule is to read the horizon that matches how long you intend to hold, and to use the others as context. A swing trader holding for a week should anchor on the five-day forecast and treat the one-day call as a timing hint for entry. A position investor should anchor on twenty days and treat short-horizon noise as exactly that. Disagreement across horizons is itself informative: a positive short-term call with a negative long-term call often describes a bounce inside a downtrend, which is a very different trade from a stock whose horizons all agree.
What hedgewing.ai shows
hedgewing.ai reports separate calibrated confidence for the 1-, 5-, and 20-day horizons, and the per-model breakdown is available on each. The concepts behind this article are covered at /glossary/time-series-forecasting, /glossary/momentum, and /glossary/post-earnings-announcement-drift. These are research outputs, not recommendations; backtests on the platform are in-sample and illustrative, and this article is not investment advice.