Underdogs and Moneylines

I took your data and had a play in Excel.

I did some estimates looking only at 0-7, 3-10 and 7-14 spreads.

I also did my own estimate of a true curve for win percentage at different dog spreads.

The polynomial produced are :

Overall : Percentage Win Chance = 0.00003x^3 - 0.00059x^2 - 0.02708x + 0.50000 (x = spread offered)
0-7 : %age win = 0.0018x^3 - 0.0172x^2 + 0.0069x + 0.5
3-10 : %age win = -0.0002x^3 + 0.0034x^2 - 0.0462x + 0.5
7-14 : %age win =0.0002x^3 - 0.0024x^2 - 0.0258x + 0.5

The R^2 fit was :

Overall = 0.91
0-7 = 0.73
3-10 = 0.61
7-14 = 0.20

Meaning only the first two are particularly good fit lines for the data.

I set them all to intercept the y-axis at 0.5 (hence the last term). This might be an error....

What that looks like is :



Dashed Line = Overall estimate
Green Line = Estimate for 0-7 point spreads
Red Line = Estimate for 3-10 point spreads
Blue Line = Estimate for 7-14 point spreads

Cheers.

Quote: Wizard

You're right the data was not aligned properly. Bug in Excel. It is no easy task getting these charts the way you want. I could break this down by home dog and away dog, but it wouldn't make much different, trust me. In my opinion an x-point underdog has just as good a chance of winning at home as away.

I was more meaning the win/loss against the spread rather than the money line. Do the spreads mostly split 50/50 for dogs and favourites, or are there spreads where the dog has historically won (with points) much more than 50%

Quote:

About the game tonight, all the late action seems to be coming on Detroit. Stations even has them at -7. You can now get Bears money line at +270 at the Hilton and Lucky's, which I think is a great bet.

Bears -7 is a bet I'd make if I wasn't wearing my Lions jersey for tonights game, and they weren't having such a great season so far. A lot will depend on the Chicago Rush offence versus the Lions Run defence. If the Bears don't get behind early, and Cutler doesn't have to throw to win the game, then this will be at least close, if not a Bears win.

If the Lions get up early, Cutler won't get the pass protection he needs, and Suh and KVB will have a field day.

And if Stafford goes down... :(

I don't use parabolic regression, because the probabilities can dip below zero for extreme values. This is more of an issue in college football, where you get very large points spreads.

Going against the spread, any math-based sports bettor knows that underdogs are the way to go. There are no special numbers, other than I think you need to get to about +3 to see the benefit. Wong's book suggests betting home underdogs only, but I think any underdog is fine.

That why I looked to see if there was any patterns for short segments of the graph... there seems to be not enough data at the extremes anyways, and with the estimate giving a 10% chance to a +15 underdog, when none of the 2 dozen 3 score under dogs have won would suggest to me there is no value at all in betting on the money when the lines are that extreme. So why not ignore those outliers and look at the regions we might bet at?

I was guessing that a exponential regression might make a nicer line, but not fit the data as well, which is what I saw when I was looking at the short regions. There's a suggestion that around 11-14 point dogs are generally getting too many points and would be good Money Line bets.

Will be something I will take into account anyways.


I just checked this page to see if you'd broken down dogs by points...:

http://wizardofodds.com/sports/sports_apx2.html

Table 4, where you look at home dogs over a season... I think the headings Win and Loss are reversed.

As I'm curious, I should go look up the data myself :)

Quote: Wizard

CM, I'm not sure I understand your comment above. The red line shows what I think is a fair, no advantage, money line, not considering the effect of key numbers in the NFL, mainly 3 and 7.

This is pretty much what I was trying to say: that you calculated the money lines to have no advantage. If I would have read the thread more carefully, I would have realized that in the first place.

More good stuff for you guys.

Here are the results by point spread in college football. The FML column is the Fair Money Line. For example, a money line on a 7-point underdog of +231 would be exactly fair. Data is from 1983 to 2005.

Spread	wins	games	Actual	Estimated	FML
0	472	944	50.0%	50.0%	100
1	157	331	47.4%	47.0%	113
1.5	132	284	46.5%	45.5%	120
2	141	280	50.4%	44.1%	127
2.5	164	376	43.6%	42.6%	135
3	246	540	45.6%	41.1%	143
3.5	134	358	37.4%	39.7%	152
4	135	337	40.1%	38.3%	161
4.5	93	259	35.9%	36.9%	171
5	74	213	34.7%	35.5%	182
5.5	101	294	34.4%	34.1%	193
6	109	343	31.8%	32.8%	205
6.5	121	383	31.6%	31.5%	217
7	146	493	29.6%	30.2%	231
7.5	84	332	25.3%	29.0%	245
8	73	243	30.0%	27.8%	260
8.5	67	227	29.5%	26.6%	276
9	62	225	27.6%	25.5%	293
9.5	64	250	25.6%	24.3%	311
10	79	320	24.7%	23.3%	330
10.5	43	210	20.5%	22.2%	350
11	56	217	25.8%	21.2%	372
11.5	38	142	26.8%	20.2%	395
12	37	179	20.7%	19.3%	419
12.5	38	176	21.6%	18.4%	445
13	42	237	17.7%	17.5%	472
13.5	42	210	20.0%	16.6%	501
14	56	328	17.1%	15.8%	532
14.5	28	194	14.4%	15.0%	565
15	19	157	12.1%	14.3%	599
15.5	20	123	16.3%	13.6%	636
16	22	171	12.9%	12.9%	675
16.5	14	144	9.7%	12.2%	717
17	29	239	12.1%	11.6%	761
17.5	17	134	12.7%	11.0%	808
18	8	121	6.6%	10.4%	858
18.5	6	89	6.7%	9.9%	910
19	18	142	12.7%	9.4%	966
19.5	10	83	12.0%	8.9%	1,026
20	9	136	6.6%	8.4%	1,089
20.5	12	109	11.0%	8.0%	1,156
21	10	138	7.2%	7.5%	1,227
21.5	4	94	4.3%	7.1%	1,303
22	5	128	3.9%	6.7%	1,383
22.5	7	71	9.9%	6.4%	1,468
23	7	115	6.1%	6.0%	1,558
23.5	5	76	6.6%	5.7%	1,654
24	6	122	4.9%	5.4%	1,756
24.5	3	68	4.4%	5.1%	1,864
25	5	62	8.1%	4.8%	1,978
25.5	2	61	3.3%	4.5%	2,100
26	1	78	1.3%	4.3%	2,229
26.5	4	55	7.3%	4.1%	2,366
27	1	79	1.3%	3.8%	2,512
27.5	2	56	3.6%	3.6%	2,666
28	6	93	6.5%	3.4%	2,830
28.5	2	41	4.9%	3.2%	3,004
29	1	52	1.9%	3.0%	3,189
29.5	1	38	2.6%	2.9%	3,385
30	1	52	1.9%	2.7%	3,594
30.5	3	33	9.1%	2.6%	3,815
31	2	58	3.4%	2.4%	4,049

Here is a nice graph for you.



If you compare these results to what you can find on the open market a good bet is tough to find. Your best odds are shopping the books around Vegas, where you can find some marginally good bets on some games. It is tougher betting offshore. Overall, the advantages are greater betting NFL money lines.

Here is a rare example of what I deem a good bet, see bet 111 on Indiana.



In other words, 210 to 1 on Indiana. My graph doesn't go that high, but I show a 39.5-point underdog fair at 112 to 1. Wish me luck!

Looks like there's a much better fit for the line on this one than in the NFL... is this a case of just a tonne more data?

I hope that was a light bet on the Hoosiers. I don't know much about College Football, but I know the Badgers are a hot team right now... :s

I wonder, could you go the reverse direction? Take the the Money Line from above and generate your own points line and see if this varies enough from the Vegas lines?

(E.g Money Line is +200 for a team, which suggests they should be +6. They are +7 (say for a big shift). Would it be worth taking the points?

Just a thought...

Pretty small bet on Indiana. Probably the biggest long shot I have ever bet in my life.

Yes, there is much more data on college football.

With football, the point spread always comes first, and other bets tend to be based on that. You could always use my chart to go the other way. I would use the money line on the underdog to get the spread, not the favorite.

Quote: Wizard

Pretty small bet on Indiana. Probably the biggest long shot I have ever bet in my life.

Yes, there is much more data on college football.

With football, the point spread always comes first, and other bets tend to be based on that. You could always use my chart to go the other way. I would use the money line on the underdog to get the spread, not the favorite.

Has a 40 point underdog ever won a college football game?

Quote: FinsRule

Has a 40 point underdog ever won a college football game?

Based on the data I have from 1983 to 2005 (12,128 games) the biggest upset was a 35-point underdog winning.