So yeah, I tried to do that, so I figured I could make a short post about it.

I'll start by explaining my methodology. Initially my idea was to make markets as needed. The first market I made for this purpose was

When this market got above some threshold I figured I could make another for 110 seats, 120 seats and so on until I got a market with a percentage low enough that making further markets would be useless. It could also go the other way around, as it happened with the Vox markets. The first one I made was for 42 seats, and because its percentage got too low I ended up making two more markets for 39 and 36 seats. I also gave more importance to the PP and Vox markets, which were the ones that seemed likely to form a government, so in the end I ended up making 6 markets for PP (100 to 150 seats), 6 for Vox (36 to 51 seats), 4 for PSOE (90 to 105 seats) and a single one for Sumar, because it took them a long time to formulate themselves as a party and I thought that the tracking would not be as interesting.

I tracked these markets daily from June 5 until election day itself, July 23, and weighted them to obtain a predicted amount of seats.

Now, with the methodology explained, let's discuss the results. On the morning of July 23 the prediction according to Manifold Markets was 138 (132-143) for the People's Party, 103 (100-106) for the Spanish Socialist Worker's Party and 36 (34-38) for Vox. The prediction was very accurate for the People's Party: 136 seats, within the range of 132-143 and two deputies away from 138. The prediction for Vox did not go so well, but too poorly either. With a result of 33 seats it is out of the range of the prediction, although only by one seat. The prediction failed for the Spanish Socialist Worker's Party, with a result of 122 that completely falls out of range.

The Manifold Markets prediction did not do that well overall, it would seem. But to be fair, most of the polls for the Spanish general election did poorly. Maybe we should see how this prediction stacks with those?

That's a table showing how much each poll deviated from the actual results. The Manifold prediction has a total deviation of 19+2+3=24, which would earn it the 6th position. But since I only made one market for Sumar I should not consider the deviation for Sumar for all the other polls either. After adjusting for that, it shares 7th place with ElectoPanel. The people from ElectoPanel pride themselves in making accurate predictions with online polling. With Manifold I was able to not only get a similar result to ElectoPanel with online questions, but also with way less participants (ElectoPanel had around 2000, Manifold Markets in the morning of the 23 around 80). This is true for every other poll, since they had 1000 participants at least. 

Yes, it's true that it was not very accurate at getting the correct result, but Manifold was still more accurate than most polls (7th out of 20) by having a fraction of their participants. That's pretty impressive and should improve your opinion on prediction markets.

If there are new elections, which is not unlikely to happen, I'll get the tracking going again to check if this was just dumb luck. And that's it, I have nothing more to say. Happy predicting.