How Far Can You See

Opening Scene

A navigator can predict the ship’s position in the next five minutes with real confidence. Predicting its exact position in five days is a genuinely different, much harder claim, even using the exact same instruments and the exact same underlying method. The further out a prediction reaches, the more room there is for small errors, unpredicted currents, and genuine chance to compound. Every forecast faces this same fundamental limit.

In Plain English

The forecast horizon is how far into the future a prediction reaches, and forecast accuracy reliably degrades as that horizon extends. A model that forecasts tomorrow’s value with genuine precision may be nearly useless for a value six months out — not because it’s a bad model, but because uncertainty compounds with distance into the future. Treating “how accurate is this model” as a single, horizon-independent number is one of the most common mistakes in applied forecasting.

The Old Way

Before this had a formal statistical treatment, people already understood the underlying limit intuitively:

A short-range weather forecast is trusted far more than a long-range one, a distinction embedded in everyday intuition long before formal meteorology existed.
A navigator trusted a position estimate for the next hour far more than one for next week, without needing a formal mathematical treatment of why.
A financial forecaster’s near-term projections were always treated with more confidence than long-range ones, an intuitive distinction long predating formal time-series methods.

The intuition that “further out means less certain” is close to universal; formal forecasting simply quantifies exactly how much less certain, and why.

What’s Changing (and Why AI Is the Reason)

Modern forecasting practice now routinely reports accuracy as a function of horizon explicitly, rather than presenting a single, horizon-independent accuracy figure that obscures this fundamental limit.
Multi-horizon forecasting methods, increasingly common in both classical and deep learning approaches, produce a full curve of predictions at multiple future horizons simultaneously, rather than one single-point forecast.
This has sharpened organizational practice around setting horizon-appropriate expectations — a supply chain team relying on a six-month forecast needs a genuinely different confidence framework than one relying on a next-day forecast, a distinction covered directly in the next article on prediction intervals.

The Metaphor, Fully Extended

The Voyage	Forecast Horizon Concept
Confidently predicting position five minutes ahead	High accuracy at a short forecast horizon
A much less certain prediction five days ahead	Degraded accuracy at a longer forecast horizon
Compounding small errors and unpredicted currents over distance	Compounding uncertainty over a longer forecast horizon
Reporting confidence differently for near and distant predictions	Reporting accuracy explicitly as a function of horizon

For Beginners: What to Actually Do

Always evaluate and report forecast accuracy at multiple horizons separately, never as a single, horizon-independent number.
Build real intuition for how quickly accuracy degrades on a specific problem by plotting error against horizon directly.
Practice explaining to a non-technical stakeholder why a forecast is trustworthy at one horizon and much less trustworthy at another.

For Practitioners and Leaders: The Deeper Layer

Match the forecast horizon explicitly to the actual decision it supports — don’t request a longer horizon than the decision genuinely requires, since accuracy at that horizon will likely be weaker.
Require horizon-specific accuracy reporting as standard practice, not a single aggregate metric that hides this fundamental degradation.
Set organizational expectations around long-horizon forecasts carefully — this is often where forecasting projects lose stakeholder trust unnecessarily.

Quick Recap

Forecast accuracy reliably degrades as the forecast horizon extends further into the future.
This isn’t a sign of a bad model; it reflects genuinely compounding uncertainty over time.
Modern practice reports accuracy explicitly as a function of horizon, not as one single number.
Matching forecast horizon to actual decision needs is a key part of setting realistic expectations.

Where This Fits in the Series

Article 13 covered how far a forecast can genuinely see. Article 14 covers how to honestly communicate the uncertainty that remains at whatever horizon you’re looking.