Industry: Government and Public Sector Product: ModelRisk Application: Tax Revenue Forecasting
A finance ministry pencilled $350B of tax receipts into next year's appropriation — a single figure carried forward from a deterministic model that grew last year's bases by a central GDP forecast and read off the answer. It is the number that anchors every spending decision in the budget, and it is presented as if it were known. It is not. Tax receipts are the product of macroeconomic conditions and collection behaviour that will only be revealed over the coming twelve months, and a point forecast hides the one thing the treasury most needs: how likely is the budget to come up short.
Re-run the same four revenue lines — income tax, corporate tax, VAT, and excise duties — as a Monte Carlo simulation, with each line's base driven by uncertain GDP and employment surprises and an uncertain collection rate, and the answer changes character. Receipts come out as a distribution with a mean of $339B, a P10 of $316B, and a P90 of $361B. Against the $350B budgeted figure, that is a 74% probability of shortfall — the optimistic budget line sits well up in the right tail of what the economy is likely to deliver. The deterministic forecast could not say this, because a point estimate has no tail to measure the budget against.
Total receipts are a sum of revenue lines, and each line is a base times an effective rate times a collection rate. The textbook grows each base by a central GDP assumption and stops. Every term in that chain is uncertain. GDP growth is carried as a Normal centred on 2.4% with a 1.6% standard deviation; employment growth as a Normal centred on 1.1%; and the collection / compliance rate as a Beta bounded on (0,1) and centred near 93%, because a rate cannot be modelled as an unbounded Normal. Each revenue line then responds through its own elasticity — corporate tax is the most cyclical (a GDP elasticity above 2), excise duties the least.
The decisive structural choice is correlation. The four lines do not move independently — they all ride the same economy. The model imposes this with a shared macro factor: a single standard-normal draw per trial that enters GDP growth, employment growth, and the collection rate, so one recession drags every line down together and one boom lifts them together. The simulation then verifies the link empirically: revenue lines correlate at 0.75 (income–corporate), 0.88 (income–VAT), and 0.63 (corporate–excise) — all strongly positive, exactly as the shared driver intends. The verification matters because correlation is what gives the total its real width: with the macro factor in place the total standard deviation is $17.4B, against just $10.1B if the lines were treated as independent. Modelling the lines as independent would understate the spread by a factor of 1.7 and quietly hide most of the shortfall risk — the classic central-limit collapse that makes a naive sum-of-lines forecast look far more certain than it is.
The distribution puts the budget line in context. The mean is $339B, the P10 $316B, and the P90 $361B — and the $350B budgeted figure sits at roughly the P75 of what receipts are likely to be, leaving a 74% probability of shortfall. A forecast that reports only "$350B expected" tells the treasury nothing about the eleven-billion-dollar gap between the budget line and the median, or about how often the economy fails to clear it.
Because receipts ride the cycle, the same engine was run under three macro overlays — recession, baseline, and expansion — by shifting the shared macro factor and the collection rate together.
The cumulative curves show how fast the shortfall risk moves with conditions. In the baseline, receipts average $339B and the budget falls short 74% of the time. In a recession, the mean drops to $320B and the shortfall probability climbs to 96.5% — the budget is almost certain to miss. Only in an expansion, with a mean of $355B, does the shortfall probability fall to 37%. The spread between those three curves is the planning range the deterministic $350B erased entirely, and it tells the ministry that a recession is not a tail risk to the budget — it is a near-certain miss.
With several uncertain drivers, the next question is which ones move total receipts — so forecasting effort and revenue-protection policy go where they pay.
The tornado ranks them by contribution to the total spread. The corporate-tax line is the largest swing at ±$15.1B, just ahead of the collection / compliance rate at ±$14.5B, with GDP growth (±$11.6B) and employment growth (±$10.5B) behind. Corporate tax leads despite being a smaller line because of its high cyclical elasticity; compliance ranks second and is the one driver the ministry can directly influence. The ranking points the policy lever clearly: a programme that lifts collection — closing the compliance gap — attacks the second-largest source of revenue uncertainty and shifts the whole distribution to the right.
The correlation the model imposes is not a modelling convenience — it is the structural fact that makes downturns dangerous, and it is worth seeing directly.
The matrix exposes how tightly the lines are bound by the shared macro factor: income and VAT co-move at 0.88, income and corporate at 0.75, and even the least cyclical pairing, corporate and excise, at 0.63. Those positive off-diagonals are why receipts cannot diversify their way to safety — when the economy turns, every line turns with it, and the shortfalls arrive together rather than offsetting. A forecast that ignores this structure treats the four lines as a portfolio of independent bets and reports a spread far narrower than the treasury actually faces.
The ministry stopped budgeting to a point and started planning against the distribution:
Monte Carlo turns tax forecasting from "what will receipts be?" into "what is the distribution of receipts, and how likely is the budget we have already committed to come up short?" — and on this budget the answer was a 74% chance that the confident $350B was an over-estimate.