Victoria 2026 Forecast
Politics · The auspol Project · Victoria 2026
Victoria 2026 Forecast
A candidate-level simulation of all 88 Legislative Assembly districts for 28 November 2026. Each of 20,000 runs draws a statewide swing from the poll trend, gives every seat its own noise, distributes preferences and counts the chamber. The model is retrained on every past election it can see and scored against those same elections before anything ships.
How to read it. A seat’s rating is how often its favourite wins across the 20,000 runs: a toss-up is under 60%, lean 60 to 79%, likely 80 to 94% and safe 95% or more. A primary vote is the forecast first-preference share, before preferences are distributed. The chamber counts include every district, and the hung-parliament figure is simply the runs where nobody reaches 45.
How it is built. The poll trend gives each party a statewide vote; each seat starts from its last result and the statewide swing, adjusted for who is actually standing (returning members, departed members, by-election winners); a gradient-boosted model trained on 22 past state and federal elections corrects that baseline. At state elections outside Western Australia, each seat is then nudged by how it swung at the last federal election. Any seat that has been polled is pulled toward its seat polls. Then the preference count is simulated with flows learned from past distributions. Every past election is scored the way it would have been run at the time, using only what was known before it. On 660 seats across seven recent elections, against the best public benchmark, it is closer on primary votes (root-mean-square error 4.9 points against 5.6, weighted by vote share, on the 654 of those 660 seats where both give every party), calls more seats right (88.9% against 86.8%), and its seat probabilities score better (log loss 0.274 against 0.285, lower is better). Source and full method: github.com/peteowen1/auspol.