Will a single manifold market beat the FiveThirtyEight World Cup Predictions for which teams make it to the round of 16?
8
352
170
resolved Dec 2
Resolved
NO

This market resolves YES if the Brier score for the predictions of the manifold market below is lower (lower is better) than that for the FiveThirtyEight 2022 World Cup Predictions: https://projects.fivethirtyeight.com/2022-world-cup-predictions/

I will compare the predictions before the start of the World Cup (November 20).

By the nature of the manifold market (multiple choice with 16 answers resolved with equal percentage) I will multiply all probabilities in the market by 16. I will assume that any probability larger than 6.25% percent in the manifold market is equal to 1.

Since I will be calculating the scores, I won't bet on the market after I gather the data (November 20). The market resolves at the end of the group stage when all teams for the Round of 16 are determined.

Brier score: https://en.wikipedia.org/wiki/Brier_score

Manifold market to compare:

Get Ṁ200 play money

#NameTotal profit
1Ṁ21
2Ṁ8
3Ṁ6
4Ṁ5
5Ṁ4
Sort by:
predicted NO

They were way closer than I expected:

Brier fivethirtyeight: 0.201

Brier manifold market: 0.206

in fact the manifold market would have won if South Korea didn't get that last minute classification

predicted NO

if you want to check:

outcomes = {"Qatar":0,

"Senegal":1,

"Netherlands":1,

"England":1,

"Iran":0,

"United States":1,

"Wales":0,

"Argentina":1,

"Saudi Arabia":0,

"Mexico":0,

"Poland":1,

"France":1,

"Australia":1,

"Denmark":0,

"Tunisia":0,

"Spain":1,

"Costa Rica":0,

"Germany":0,

"Japan":1,

"Belgium":0,

"Morocco":1,

"Croatia":1,

"Brazil":1,

"Serbia":0,

"Switzerland":1,

"Cameroon":0,

"Portugal":1,

"Ghana":0,

"Uruguay":0,

"South Korea":1}

print("Brier fivethirtyeight:", sum([(fivethirtyeight[country] - outcome)**2 for country, outcome in outcomes.items()])/32)

print("Brier manifold market:", sum([(mm_[country] - outcome)**2 for country, outcome in outcomes.items()])/32)

predicted NO

Just downloaded the data. These are the probabilities that I will be using:

mm = {'England': 1,

'Argentina': 1,

'Brazil': 1,

'Spain': 0.9451968360534888,

'France': 0.8974907339737943,

'Portugal': 0.849977244235089,

'Germany': 0.8112415168267147,

'Netherlands': 0.7971968295073112,

'Belgium': 0.7595755431348061,

'Senegal': 0.7426855519809036,

'Croatia': 0.741065876527692,

'Denmark': 0.6719708710471769,

'Uruguay': 0.617516701467408,

'Switzerland': 0.49974494455889923,

'Mexico': 0.49965919586782787,

'Poland': 0.4608358796614559,

'United States': 0.45698850032681404,

'Wales': 0.36538907075213484,

'Serbia': 0.3057351303302865,

'Morocco': 0.27690844073942406,

'Australia': 0.18374603476500043,

'Cameroon': 0.15410561203952933,

'South Korea': 0.14789233306591248,

'Iran': 0.09774239358780965,

'Ghana': 0.09141480094632379,

'Japan': 0.08992401674617002,

'Tunisia': 0.08298078235641886,

'Saudi Arabia': 0.06596123866028777,

'Qatar': 0.028929836130672053,

'Costa Rica': 0.02812887522247895}

fivethirtyeight = {'Brazil': 0.91045,

'Spain': 0.80686,

'France': 0.82686,

'Argentina': 0.83995,

'Portugal': 0.81087,

'Germany': 0.76424,

'England': 0.80426,

'Netherlands': 0.79006,

'Denmark': 0.64835,

'Uruguay': 0.65274,

'Belgium': 0.624,

'Croatia': 0.54419,

'Switzerland': 0.47861,

'United States': 0.52953,

'Mexico': 0.53636,

'Senegal': 0.51129,

'Morocco': 0.46473,

'Serbia': 0.41295,

'Japan': 0.34492,

'Poland': 0.3792,

'South Korea': 0.35636,

'Tunisia': 0.30862,

'Iran': 0.34207,

'Wales': 0.32414,

'Cameroon': 0.19799,

'Saudi Arabia': 0.24449,

'Australia': 0.21617,

'Qatar': 0.21656,

'Ghana': 0.18003,

'Costa Rica': 0.08398}

predicted NO

@egroj Nicer format for comparison (but rounded):