Forecasting Construction Prices When the Estimate Has to Last

Some engineer’s estimates get tested in six weeks, when the bids come in. Others have to survive five years — carried through phase gates, capital programs, and funding cycles before a single contractor prices the job. And increasingly, the forecast itself is graded: come in more than 10% off the escalated number, and you’re explaining yourself to the client.

That changes what estimating is. You’re no longer pricing a project. You’re pricing a project and predicting a market — and the second half is where estimates die.

The forecast is part of the estimate now

On long-horizon public work, the engineer’s estimate does double duty. It’s the basis for the owner’s budget request today, and it’s the yardstick the owner will hold up on letting day, years from now. The line items you can defend with takeoffs and unit-cost history. The escalation assumption is the exposed flank — a single number, compounded across years, that nobody on the project team picked with much more than convention behind it.

Three tools show up on nearly every engineer’s desk for that number. Each is useful. None of them, alone, is a forecast.

1.Flat escalation factors

The workhorse. NJDOT’s Cost Estimating Guideline specifies a 3% simple (not compounded) inflation factor, applied for the number of years between the estimate date and the project letting date. Consultants working NJTA programs apply the Authority’s cost escalation guidance — 3.5% per year — the same way. Agency-published factors have one great virtue: they’re defensible by citation. Nobody gets fired for applying the circular.

The problem is what a flat factor assumes: that asphalt, structural steel, drainage pipe, electrical, and labor all inflate at the same rate, in every county, in every year. They don’t — they never have. Run the math on what that costs you:

2.OMB Circular A-94

Engineers on federally funded work reach for OMB Circular A-94, the White House guidance that sets discount rates for evaluating federal programs, lifecycle costs, and lease-purchase decisions. For calendar year 2026, its published real discount rates run 1.1% (3-year) to 2.0% (30-year), with nominal rates from 3.4% to 4.1%, refreshed each year in an OMB memo.

A-94 is the right tool for a specific job: comparing alternatives over a lifecycle. Discounting tells you whether the concrete deck beats the steel deck once you account for maintenance and replacement on a common present-value basis. What it does not do is predict what a contractor will charge for either deck in 2031. A discount rate is a statement about the time value of money — not about the future price of structural steel. Treating the two as interchangeable is one of the quieter ways long-range estimates go wrong: the arithmetic looks rigorous, and the citation is impeccable, but no market information has actually entered the number.

3.Standard regression

The most sophisticated of the three: fit a trend line to historical unit prices and extend it. Done per line item, regression beats any flat factor, because it lets asphalt behave like asphalt and steel behave like steel. But anyone who has actually run the regression on bid data knows where it hurts:

• The inputs are polluted. Bid tallies are full of unbalanced bids — front-loaded mobilization, pennied-out items, money moved between pay items to pull cash forward. A regression can’t tell a market price from a cash-flow strategy.
• Outlier handling is a judgment call. Which points you exclude changes the slope, and a slope you hand-tuned is a slope a client can challenge.
• Small samples, big items. The line items that carry the estimate — the half-million-dollar structural items — are exactly the ones with the fewest comparable observations in your firm’s repository.
• Straight lines don’t do cycles. A trend fitted through a demand spike, a materials shock, or a tariff regime extrapolates that anomaly five years forward.

A flat escalation factor isn’t a forecast. It’s a treaty — everyone agrees to be wrong by the same amount.

What a defensible five-year forecast looks like

You can’t remove the uncertainty from a five-year horizon. You can change what you’re escalating from, what you’re escalating with, and how well you can show your work when the number is questioned. In practice:

• Escalate line items, not lump totals. Materials diverge. A project-level factor averages away exactly the information that determines whether your big-ticket items sink the estimate.
• Anchor to the current market, not your last project. If the base number is already two years stale before you escalate it, the forecast inherits that error on day one. Start from what the market is pricing now, in the project’s actual geography.
• Refresh at every phase gate. A five-year forecast shouldn’t survive five years untouched. Re-benchmark the estimate against current market pricing at each design milestone, and let the escalation assumption shrink as the letting date approaches.
• Carry a range, not a point. A single escalated number invites a single verdict: right or wrong. A distribution — here’s the sweet spot, here’s the tolerance band — turns the bid-day conversation from “you missed” into “we landed inside the range we documented.”
• Write the methodology down. When the forecast is scrutinized, the difference between an awkward meeting and a routine one is whether you can produce the basis: the data, the trend, the assumption, the date it was made.

Where PinPoint fits

Escalation circulars, A-94, and regression all still have seats at the table — for compliance, for lifecycle comparisons, for trend context. What they were never built to do is tell you what a contractor will bid in a specific county in a specific season. That’s a market question. Answer it with market data.