Why two credible forecasts disagree

Two firms run the same market through the same class of software and land eleven dollars apart on the 2032 annual average. Neither is incompetent. Neither is being dishonest. This happens constantly, and the usual response — pick the number you like, or split the difference — is worse than either forecast taken on its own.

The disagreement is almost never in the engine. Production cost models are close to commodity at this point. Give two competent teams the same inputs and the same topology and they will produce results that differ at the margin, not at the level. The gap comes from the assumption set, and the assumption set is the part nobody reads.

There are perhaps six inputs that carry most of the variance in a long-dated price forecast: the gas curve, load growth, the interconnection queue and what fraction of it you believe, retirements, transmission expansion, and the reserve margin logic that governs when scarcity pricing shows up. Every one of those is a judgment. Several of them are judgments about policy, which is to say judgments about politics.

So when two forecasts disagree, the useful question is not which one is right. It is: where do these two assumption sets diverge, and does the divergence matter for the decision in front of me. Sometimes it does not — a lender stress-testing debt sizing may find both decks clear the coverage test, in which case eleven dollars is noise. Sometimes it decides everything.

That question is answerable. It requires the forecaster to publish an assumption sheet, and it requires you to read it. Both are rarer than they should be.¹

Footnotes#

1. In my experience the assumption sheet usually exists. It is simply never requested. ↩