Revenue projections should be accompanied by prediction intervals rather than ordinary parameter confidence intervals, because the goal is to quantify uncertainty about future observations, not only about estimated model parameters. Rob J Hyndman, Monash University explains in accessible forecasting guidance that forecast intervals capture both parameter uncertainty and the inherent variability of future outcomes. George Box, University of Wisconsin–Madison emphasized that model error and structural change must be recognized explicitly when communicating forecasts.
Recommended interval levels
Common practice is to show multiple intervals to support different decision tolerances. A central 50% interval communicates a core expectation around the median, an 80% or 90% interval shows moderate downside and upside risk useful for operational planning, and a 95% interval captures tail risk important for capital allocation and contingency planning. Central banks and macro-institutions often present fan charts or multiple percentiles to convey these layers of uncertainty, and the Bank of England uses such visualizations to make asymmetric risks visible to policymakers.
Causes and consequences of interval choice
Interval width and credibility depend on sample size, model specification, volatility of the underlying market, and exogenous shocks such as regulatory change or supply-chain disruption. Choosing only a single, narrow interval creates overconfidence and can lead to under-reserved budgets, missed covenants, or abrupt operational shortfalls. Conversely, presenting only very wide intervals can be unhelpful for routine decision making by obscuring actionable ranges and reducing trust. For companies operating across territories, cultural and institutional differences matter: emerging markets often require wider intervals because political and currency risks make revenue outcomes more dispersed, while mature markets may support tighter short-term forecasts.
Practical application requires labeling and methodology transparency. State explicitly that intervals are prediction intervals derived from the selected time-series model or simulation method, and disclose whether intervals incorporate parameter uncertainty, structural-break scenarios, or correlated forecast errors across product lines. Use out-of-sample backtesting and forecast skill measures to justify interval calibration; documented validation increases credibility for investors and regulators.
For many business contexts, presenting a combination of 50%, 80% and 95% prediction intervals, explained with the chosen model and validation results, balances clarity and prudence. This multi-tiered approach aligns statistical practice with managerial needs by showing both typical outcomes and plausible extreme scenarios while recognizing model limitations and real-world risks.