Scott Armstrong bets that he found a method that will help to further increase the accuracy of the combined PollyVote forecast.
In particular, Scott proposes that adding a naive model will improve accuracy when uncertainty is high, such when the election is still far away and when the parties’ eventual nominees are yet to be known. This suggestion is based on the Golden Rule of Forecasting, which recommends considering a naive model if the situation involves uncertainty.
In operational terms, Scott suggests the following:
- Until the candidates from both parties are nominated on July 28, 2016: add the naive model, which predicts a vote share of 50% for each party, as a seventh component to the PollyVote.
- From July 29th: move on with the regular PollyVote.
The chart above shows how the forecasts of the original PollyVote and Armstrong’s version that includes the naive model have developed from January until July 28th (after that, the forecasts are of course similar).
We will keep you posted on whether Scott is able to beat Polly for long-term forecasts!