Multi-Year Financial Forecasting for Small and Mid-Sized Governments

Multi-year forecasting is standard practice in large jurisdictions and sporadic below a certain size, usually for a stated reason — insufficient staff — and an unstated one: the last forecast was wrong and became an argument against the department that produced it. Both problems have the same solution, which is to design the forecast so that being wrong is informative rather than embarrassing.

Forecast the drivers, not the line items

A forecast built by escalating four hundred general ledger accounts at assumed growth rates is impossible to explain and impossible to defend. A forecast built on eight to twelve drivers can be discussed by a council member.

Typical drivers for a general fund:

• Assessed value growth and the effective levy rate
• Taxable sales growth, where applicable
• Headcount by function, and the negotiated or assumed wage schedule
• Health benefit cost trend
• The actuarially determined pension contribution
• Debt service on issued and planned debt
• Contractual escalators on major service agreements
• Non-personnel inflation on everything else

Everything else moves with one of these or is small enough not to matter. The discipline of naming twelve drivers forces a useful conversation about which assumptions the forecast actually depends on.

Three scenarios, not a point estimate

Publishing a single number invites treating it as a prediction. Publishing three — a baseline, a downside, and an upside — invites treating them as a range, which is what they are.

The scenarios should differ in a small number of driver values, stated explicitly. A downside case in which every driver simultaneously takes its worst plausible value is not a scenario; it is an arithmetic exercise with a probability approaching zero, and experienced readers discount it accordingly. Two or three linked drivers moving together — a recession affecting sales tax, permit revenue and investment earnings — is realistic and therefore persuasive.

Keeping score

Every forecast should be accompanied by a retrospective on the previous one: what was projected, what happened, and which driver accounted for the variance. Departments resist this and are wrong to. A forecast that missed by four per cent because assessed value grew faster than assumed is a good forecast with an identifiable cause. A forecast that missed by four per cent for reasons nobody can decompose is a black box, and it will be treated as one.

The retrospective also has a governance function: it establishes that missing is expected. Forecasts are not commitments, and the first time that distinction is tested is usually the worst possible moment to be establishing it.

Practical scope for a small entity

A defensible five-year general fund forecast for an entity with a fifty-million-dollar budget is a workbook of perhaps six sheets, updated twice a year, taking a competent analyst three or four days per update after the first build. That is achievable. The failure mode is attempting to forecast every fund at the same level of rigour in the first year. Start with the general fund and the largest enterprise fund; add others when the process is stable.

Common technical errors

• Escalating a base year that contained one-time items. Normalise the base before growing it, or the anomaly compounds for five years.
• Forecasting revenue and expenditure with independent inflation assumptions. If wages are assumed to grow at four per cent, the sales tax base assumption should be consistent with an economy in which wages grow at four per cent.
• Omitting the capital plan's operating impact. A new facility carries staffing, utilities, insurance and maintenance in the operating budget from the year it opens.
• Holding transfers constant. Interfund transfers are frequently the plug that makes a forecast balance. If they are growing in the forecast, the forecast is describing a subsidy that will eventually be questioned.

Presenting it

One chart of projected fund balance under three scenarios against the reserve policy floor, one table of driver assumptions, one sensitivity table, and two pages of narrative. Anything longer will be summarised by someone else, and the summary is what will be discussed.