Check whether a cash forecast's ranges match its actual uncertainty

A low and high cash forecast are useful only when the reader knows what they represent. They might be two management scenarios, a statistical prediction interval, or a lower-bound liquidity policy. Those are different objects. Label the range before using it to determine a funding buffer or payment decision.

A prediction interval describes uncertainty about a future outcome under a model and its assumptions. A scenario describes the outcome of selected assumptions and need not have an assigned probability. Calling a hand-picked downside case an 80% range does not make its coverage measurable unless the probability statement has a defined basis.

Forecasting: Principles and Practice discusses prediction intervals and the evaluation of forecasts against future observations. The operational review below uses those principles to examine whether a cash forecast's stated ranges have been useful over time. It is separate from deciding how much liquidity the business should retain.

A central 80% prediction interval typically aims to leave 10% probability below its lower endpoint and 10% above its upper endpoint, according to the model. An 80th-percentile cost estimate is a one-sided threshold with a different interpretation. Using the same percentage in their labels does not make them interchangeable.

For cash balances, the lower tail is often the immediate concern. State whether the lower endpoint is a particular percentile or merely a downside scenario. Also state whether the balance includes only available cash, whether funding draws are assumed to occur, and whether management interventions are included.

A range for end-of-week cash does not describe the minimum balance within the week. If Monday payroll and Friday receipts create a temporary trough, the relevant funding risk may sit outside the endpoint measure. Define the outcome that the forecast is intended to cover before reviewing its statistical properties.

Suppose a team issued 20 one-week-ahead intervals, each labeled as a central 80% range. Thirteen eventual outcomes fell inside their original lower and upper limits. Observed coverage is 13 divided by 20, or 65%. That is a useful warning to investigate, although 20 observations alone do not establish the long-run coverage precisely.

Separate misses below the lower endpoint from misses above the upper endpoint. If all seven misses are below the range, the issue may involve persistent optimism about cash, an omitted payment category or a shifted collection pattern. A balanced set of misses can indicate a different problem, such as ranges that are too narrow overall.

Preserve each forecast's original issue date and endpoints. Recalculating the range after actuals are known invalidates this coverage check. Review one-week and eight-week forecasts separately because their uncertainty and available information differ.

Inspect coverage in conditions that matter to the decision, such as payroll weeks or weeks with concentrated customer receipts. An overall 80% result could combine excellent coverage in routine weeks with repeated misses during the exact periods treasury needs protection. Keep such subgroup findings proportional to the sample size. A handful of observations can motivate a case review without supporting a precise new probability estimate for that subgroup.

A range stretching from zero to an extremely high balance may contain almost every outcome while giving treasury little help. Review interval width together with coverage. Compare widths across similar horizons and business conditions, and ask whether the range changes the funding or payment decision.

For example, two methods may both cover 16 of 20 outcomes, while one averages a $40,000 width and the other a $160,000 width. The narrower method appears more informative in that sample, but inspect the missed outcomes and the conditions behind them before choosing it. A method that misses the severe shortfall can have a satisfactory average score and still be unsuitable for the intended decision.

Do not shrink ranges merely to make them look more actionable. If available information supports a wide range, preserve that uncertainty and identify which missing customer or payment facts would reduce it. The uncertainty is part of the planning problem.

Historical forecast errors may be a poor guide after a new payment processor, billing policy, large customer or seasonal pattern changes the cash process. Tag these events and review their effect on the distribution of misses. A range built from quiet operating weeks should not silently represent a major transition.

Related cash flows also require attention. Several customers can delay payment because they share a parent company, industry shock or approval process. Treating their outcomes as independent can understate the chance of a concentrated shortfall. Use evidence about the shared exposure and make the dependence assumption visible.

Where probabilities cannot be supported, use clearly described scenarios. A scenario can still be operationally useful when it names the delayed receipts, payment obligations and response options. Its value does not depend on attaching an unsupported percentage to it.

Ask what action follows when the lower range approaches the business's minimum available-cash requirement. The answer may involve collecting specific invoices, moving discretionary spending or arranging funding with enough lead time. The action rule should account for the cost of an unnecessary intervention and the cost of a missed shortfall.

Keep the probability model separate from management's risk preference. A business can choose a conservative funding buffer even when the forecast is well calibrated. Conversely, a narrow buffer does not become prudent because a chart uses a precise-looking interval.

Publish the range definition, horizon, historical coverage, typical width and known coverage gaps with the cash forecast review. Retain the original ranges and actual outcomes for the next evaluation. If the forecast has not yet accumulated enough comparable observations, say which decisions remain supported mainly by scenario judgment.