58 lines
2.5 KiB
Plaintext
58 lines
2.5 KiB
Plaintext
TODO
|
|
|
|
o Technical
|
|
o Handle different keys for team names, such as JAX/JAC and SD/LAC
|
|
o Double-check that csv file has the right data (week, team names, scores)
|
|
|
|
|
|
o Data
|
|
o Historical odds at game time (ESPN's are frozen early in the week)
|
|
o Create strip_stats database
|
|
|
|
|
|
o Strategy
|
|
o Record historical error values with respect to popular confidence in each game. Important to measure accuracy of crowd and identify what biases cause the crowd to be wrong
|
|
|
|
Math:
|
|
|
|
confidence (c: -1.0 to 1.0, negative value menas home team is favorite):
|
|
p = home team popularity (0.0 to 1.0, > .5 means home team is favorite)
|
|
c = 2(0.5 - p)
|
|
|
|
accuracy (a: accuracy of spread, negative value means home team was overrated)
|
|
s = spread
|
|
m = margin of victory (away team score - home team score)
|
|
a = s - m
|
|
|
|
rightness (r: positive values = right, negative values = wrong):
|
|
r = -ca
|
|
|
|
| popularity |(confidence)| away score | home score | (mov) | spread |(accuracy)|(rightness)|
|
|
| 0.150 | +0.700 | 20 | 10 | +10 | -5 | -15 | +10.50 |
|
|
| 0.800 | -0.600 | 10 | 20 | -10 | -5 | +5 | +3.00 |
|
|
| 0.520 | -0.040 | 16 | 20 | -4 | -5 | -1 | -0.04 |
|
|
| 0.520 | -0.040 | 15 | 20 | -5 | -5 | 0 | 0.00 |
|
|
|
|
o Identify biases
|
|
o Persistent biases (like home underdogs and other stats where people notoriously miscalculate their importance)
|
|
o Temporary biases (maybe like hating on a team for some reason)
|
|
o Last week bias (check if/how much people tend to overestimate teams that won in previous week(s) and/or underestimate teams that lost in previous week(s))
|
|
o Streaks that cause people to miscalculate odds
|
|
|
|
o Take close look at home underdog games since oddsmakers deliberately place the "wrong" odds to improve their payout (they take advantage of known crowd bias in favor of away team)
|
|
|
|
o Measure instantaneous rate at which people are choosing home vs. away. This is important to recognize when the market is moving in one direction, possibly due to new information
|
|
|
|
o Change weight of games with anomolous events like large turnover ratio, significant injuries, etc.
|
|
|
|
o Curiosities
|
|
o Overall accuracy of crowd
|
|
o Compared to
|
|
|
|
|
|
Streaks
|
|
Track:
|
|
- win/loss streak (negative = losses)
|
|
- instantaneous season record (wins/losses/ties)
|
|
-
|
|
|