I hope to have finished the major load on my forecasting model algorithm. I also conclude this project with a new set of clear graphs, although, I still feel working on the colors and scales.
I’ve received a few emails asking for more details on the simulations and to make the code available, which is part of my plans since ever. But, first I’ve to conclude all the robust checks, and I’ve been quite busy these days working on other stuff.
Below is the Expected outcome simulated. The graphs come next. Last, is the convergence diagnostic graph.
R> summary(mcmc2014) forecast class object: MCMC: 1000 samples with a burnin of 500 and a thin of 5. Predicted at 2014-09-10, using 32 polls from 6 houses. Evidence starting at 2014-02-19. Election at 2014-10-05. Machine runtime: ~ 28 minutes. Expected results + swing voters: Expected results PT 0.35051 PSDB 0.18537 PSB 0.21591 Others 0.05782 None 0.08691 Swing 0.10349 Probability Intervals (95%): 2.5% 50% 97.5% PT 0.33273 0.34964 0.37397 PSDB 0.16884 0.18478 0.20501 PSB 0.17963 0.21576 0.25258 Others 0.04282 0.05766 0.07406 None 0.06892 0.08701 0.10305 Swing 0.08873 0.10387 0.11728
DILMA ROUSSEFF (PT)
MARINA SILVA (PSB)
AÉCIO NEVES (PSDB)
As it becomes clear, there are some strong spikes towards the end of the chain. I’ll look at this next.