Football Returns

It's been a long offseason.

Too long, indeed.

The football drought is like global warming - you can't feel it until it happens. But the season has begun, and so has the betting.

New format for me this year: less talk, more stats. I'm timed constrained now, so I would rather post frequently with no commentary than infrequently with useless blather. To that end, here's a quick summary of new stats and a taste of what it will look like:

1.) A new process has been added (crafted by, of all people, my lovely wife) that identifies games in which Vegas has a bias. It's a succinct "model of the market." A perfect example of this from 2007, the Pats perfect season: from Week 3 through Week 11, NE was giving high, but earthbound spreads. Starting Week 12, the spreads rose to ludicrosity. This can be explained because there was inherent bettor's bias reflected in the early spreads. Vegas continued to think, "they can't POSSIBLY cover AGAIN." The level of bias was sustained until after Week 11, when Vegas (and bettors) abandoned their bias. Understand that "bias" is non-directional, it simply means that Vegas is setting spreads in a way that doesn't reflect reality, either too high or too low. This has helped isolate games to "arbitrage" using statistical information to correct for excess bias.

2.) A new way to measure team performance volatility by measuring the expected performance gap (how a team should perform) versus actual performance (how a team did perform). Hypothetically, long run results should show that higher ranked teams (teams who perform as expected consistently) should have better results as picked by the model. So far, there is a statistically insignificant performative advantage to lower volatility teams, but I'm hopeful I can iron that out.

3.) A new power ranking system, taking volatility and expected performance into account. It looks a lot more like traditional power rankings, with some nice wrinkles.

Without further ado, here are this week's bets:

1	CHI	@ GB	+3.5
2	PHI	@ CAR	+1.5
3	MIA	@ ATL	+4
4	DET	@ NO	+13
5	SF	@ ARI	+6.5
6	TEN	@ PIT	+6.5
7	STL	@ SEA	+8.5
8	NYJ	@ HOU	+4.5
9	TB	v. DAL	+6
10	JAC	@ IND	+7
11	KC	@ BAL	+13
12	DEN	@ CIN	+4
13	WAS	@ NYG	+6.5
14	MIN	@ CLE	-4
15	SD	@ OAK	-9
16	NE	v. BUF	-10.5

Almost all away, almost all dogs.

Note that bets for Weeks 1-3 are HIGHLY VOLATILE, as there is no seasonal data on which to base performance. I don't recommend any betting until Week 3 at least.

It's That Time of Year...

I feel like the guy in that Office Max commercial, dancing through the aisles as happily buying "going back to school" gear.

Football season fast approaches, and the box and I have spent some very quality time together. Hot hot quality time. I thought I'd get a jump start, though I won't be ready for much commentary until closer to prime time in late August. Here are some changes to expect:

1.) No more "pure" financial measures.

The BOX is now BOX 3.0, a total upgraded version. I am currently starting a green hedge fund (see here for the gist, and here for more details) with a model I've worked on for 2 years, and I have a bunch of new tricks up my sleeves. I may revisit the Financials (alpha, beta, sharpe ratio) later this summer, but for now, they are a thing of the past. They have been replaced by new metrics, most importantly: Game Value, Performance Volatility, and Expected Performance. I'm combining a lot of the hedge fund volatility metrics and applying them here to performance (ie, point performance, performance vs. the spread, statistical field performance, performance against an average, etc.). Let's just say the database got much thicker.

Game Value - the statistically based game value based team performance and record.

Performance Volatility - the volatility of a teams week to week performance statistically on multiple levels.

Expected Performance - a sort of futures derivative for football performance, looks at a combination of Vegas and statistical expected performance.

I'll go over it all in more detail as the season gets closer, suffice to say, I'm pretty stoked.

2.) I cracked the Holy Grail: 60% historical win rate.

It's legitimate. Whether it will hold up is another question, but back to back years of 60% performance using the new metrics. You might be asking, "How did you accomplish such a feat Mr. Genius?" It's a good question, but a key component was something obvious that I overlooked last season. I spent a good deal of time reviewing how many games to bet to maximize my win percentage, but I ignored the simple fact: all weeks are not created equal. GVal (Game Value) assigns realistic, statistically based value to upcoming games, similarly to Mark Myers' TIMP system but using a totally different data set. In fact, I based values on over 15 years of data (it's been a busy summer) to come up with real game values that are organic. I initially resisted what GVal seemed to suggest; don't bet on games that don't have value, even if it means betting on NO GAMES in a particular week! While this isn't much fun, espeically if it's more than one week of no betting in spread games in a row, it's definitely profitable. Limiting myself to the highest probability games as values shift means some weeks I could make 16 bets, others 2 bets, but only bets that have real value.

I haven't expanded the notion to moneyline, teasers, and parlays yet, but I still have many modifications to go before the season starts.

3.) No more wondering how much to bet.

While I still haven't devised a way to have the BOX suggest a minimum bet, I have figured out how to enter a minimum (ie, $5 at my online "broker"), and the BOX does the rest. I plan on adding the "Doubler" again in a few test cases, but the allocations now act more like stock portfolios than football betting. Box 2.0 devised allocations as well, but needed multiple variables each week, like how much to bet and how much the minimum is every week. Now, the BOX says "Bet this much assuming a minimum of X".

4.) The spread/juice stock ticker.

One of the last things Vegas wants, in sharp contrast to equity markets, is real time data. Betting arbitrage would be a disaster. Imagine a system where Vegas sets the line of NE v. KC at NE -10, -125. Money comes pouring in for NE - bettors think the spread is too low, for whatever reason. Suddenly, in order to compensate for this, Vegas needs to a.) raise the spread, b.) raise the vig, or c.) both so they don't end up on the tremendous losing end of this game. So within a day, the new line is NE -15.5, -115. Imagine if, along the way, you had a ticker giving you the line. As you saw the line pump up, you realized - hey, I should get in on the ground floor here - and you place a bet at NE -12, -120. When the line jumps above 14, you think maybe KC has a chance here, and because it's happening in real time, you can lock in a bet at KC +14.5, +110, hedging yourself on the upside. But real time markets mean real time money and real time hedging, all of which are bad for Vegas.

So in lieu of Vegas, the BOX now takes multiple ticks in the "market". I haven't tested this yet, as I lack the back data (it's impossible to get in usable form), it's a step forward in actual betting arbitrage. So watch for this later this season, it could prove awesome.

So that's the scoop. I don't imagine I'll be posting much again until preseason in mid/late August, but it's fast approaching! Get psyched!

Week 16 - True Win Percentages, Sighs Abound

So as I mentioned in my last post, I found a major flaw in my programming. A major flaw. What was it you ask? Let me tell you and rant about my moron-acy.

On a weekly basis, I enter the scores into the Box to update my performance. I went to enter the scores from Week 15 and update my performance when I noticed something odd happened -- my picks changed. Magic? I think not. So I deleted the scores and checked my picks, and they were back to the picks I posted here. "Hmmm..." I thought. That's odd. So I went back through my math, picking around, thinking maybe I just entered something wrong. Lo and behold, my stupidocity went far beyond a wrong entry.

Part of my pick algorithm figures in historical pick win percentages. For instance, if my number 6 pick historically is winning 60% and my number 5 pick is winning 50%, the Box is trained to say, "that's weird, we should switch them". As more data comes in, the Box should get more accurate as the best picks float to the top. It's a simple way to train the box using wins and losses, but it's been working effectively. This week, I noticed that the Box is not using historical win percentages (as in, based on overall win percentages on Week 14, the Box is trained for Week 15), it was using live win percentages (as in, as scores come in, the Box changes current picks to look like it wins more). That explains my win percentage of a ridiculously astounding 87% for my number one pick - the Box keeps CHANGING the number one picks going backwards to make them look like winners.

Oops.

So, I have fixed the error. And I have cried and lied in the fetal position. I need a binky. Is all lost? Not even close.

Here are my true win percentages (note: these are before modification using the financials - actual win percentage should be slightly higher), and they are far more realistic:

	Parlay	Tease	Ind
1 Team Bet	0.00%	0.00%	60.61%
2 Team Bet	39.39%	69.70%	59.85%
3 Team Bet	18.18%	48.48%	53.54%
4 Team Bet	9.09%	39.39%	50.38%
5 Team Bet	3.03%	27.27%	50.61%
6 Team Bet	0.00%	18.18%	50.76%
7 Team Bet	0.00%	15.15%	52.16%
8 Team Bet	0.00%	15.15%	52.08%
9 Team Bet	0.00%	15.15%	51.35%
10 Team Bet	0.00%	6.06%	51.82%
11 Team Bet	0.00%	3.03%	52.89%
12 Team Bet	0.00%	3.03%	52.90%
13 Team Bet	0.00%	3.03%	52.91%
14 Team Bet	0.00%	3.45%	52.59%
15 Team Bet	0.00%	5.26%	52.11%

If the glass was half empty, I just lost 27% of my number one pick win rate. If the glass is half full, 60% is INCREDIBLY PROFITABLE if you bet real money. So it's really bitter sweet. The teaser bets are a little misleading - I hit a 15 team teaser once last season, and it's skewed the results slightly. If you remove my 2006 numbers, my numbers get much uglier for 2007:

	Parlay	Tease	Ind
1 Team Bet	0.00%	0.00%	50.00%
2 Team Bet	31.25%	62.50%	51.56%
3 Team Bet	12.50%	31.25%	45.83%
4 Team Bet	6.25%	31.25%	46.09%
5 Team Bet	6.25%	12.50%	46.88%
6 Team Bet	0.00%	6.25%	47.40%
7 Team Bet	0.00%	0.00%	47.77%
8 Team Bet	0.00%	0.00%	48.05%
9 Team Bet	0.00%	0.00%	46.18%
10 Team Bet	0.00%	0.00%	45.94%
11 Team Bet	0.00%	0.00%	45.74%
12 Team Bet	0.00%	0.00%	45.31%
13 Team Bet	0.00%	0.00%	45.91%
14 Team Bet	0.00%	0.00%	47.19%
15 Team Bet	0.00%	0.00%	44.81%

The favorite trend is killing the Box (and Vegas, for that matter - I read that handicappers can't figure out how high to set spreads, since this year the faves are just killing everything). But despite losing, the teasers win. And win with a buffer underneath them. Which tells me that the Box is roughly 6 points away from a killer win percentage this year. Sigh.

The real underlying issue now is confidence. My self confidence is mildly shattered. Well, that's a lie. I'm a cocky bastard. It might be sprained a bit, but nothing some HGH can't fix.

So the question now is, what to bet? Based on these win percentages, this is what I should do:

1.) 2 Individual Spread bets
2.) 2 Team Parlay (cheap if I can get it)
3.) 4 Team Teaser

That's right, the teaser. The teaser I so ardently mocked turns out to be my most profitable bet. Expect to see it this week in my bets, as I swallow my pride and go for the bucks.

Stay tuned for later today, and don't forget if the Steelers/Rams are in your line up, BET TODAY, it's crappy football Thursday on NFL Network!

Week 14 - Interests Include Parlays, Teasers, and Long Walks on the Beach

The Capper likes to talk crazy. So crazy, he's doing a 10 team progressive parlay, 3 three team teasers, and a partridge in a pear tree. Since the Box's bets are in (posted again here for posterity), we'll do a little analysis of this betting style.

Bets for Week 14

Spread: ARI +7 (-105) v. SEA - $10.33 to win $9.84

Spread: KC +6.5 (-110) v. DEN - $7.79 to win $7.08

Spread: STL +6.5 (-115) v. CIN - $7.96 to win $6.92

Spread: SD 0 (-110) v. TEN - $18.28 to win $16.62

Spread: PIT +10 (-110) v. NE - $8.31 to win $7.55

Parlay: KC +6.5 (-110) v. DEN, ARI +7 (-105) v. SEA, STL +6.5 (-115) v. CIN, PIT +10 (-110) v. NE, MIA +7 (-110) v. BUF - $5.00 to win $55.00

With that out of the way, on to the analysis.

First, we'll analyze the payout schemes. Now, when the bookmakers give you a payout schedule, it's typically an estimation depending on the juice. You can expect more or less what they tell you, but I found that depending on the type of bet (ie, moneyline parlays), the payouts can vary. For the purpose of argument, we'll say the payouts are set in stone, because it'll simplify the analysis.

Parlay Payouts

	Friday Only Special Parlay Odds[-105]	Regular Daily Parlay Odds
2 Teams	2.8/1	13/5
3 Teams	6.4/1	6/1
4 Teams	13.5/1	10/1
5 Teams	27/1	25/1
6 Teams	54/1	40/1
7 Teams	107/1	75/1
8 Teams	210/1	150/1
9 Teams	411/1	300/1
10 Teams	804/1	700/1
11 Teams	1,570/1	1,100/1
12 Teams	3,066/1	1,800/1

Teaser Payouts for football:

Football:	6 Points	6.5 Points	7 Points
2 Teams	-110	-120	-130
3 Teams	+180	+160	+140
4 Teams	3/1 (+300)	5/2 (+250)	2/1 (+200)
5 Teams	4.5/1	4/1	3.5/1
6 Teams	6/1	5.5/1	5/1
7 Teams	10/1	9/1	8/1
8 Teams	15/1	12/1	10/1
9 Teams	20/1	15/1	12/1
10 Teams	25/1	20/1	15/1
11 Teams	35/1	25/1	20/1
12 Teams	50/1	35/1	25/1
13 Teams	75/1	50/1	35/1
14 Teams	100/1	75/1	50/1
15 Teams	150/1	100/1	75/1

Progressive Parlay

	All Teams Won	1 Team Loss	2 Teams Lose	3 Teams Lose
4 Teams	5/1	1/1	--	--
5 Teams	11/1	2/1	--	--
6 Teams	20/1	9/2	--	--
7 Teams	40/1	4/1	1/2	--
8 Teams	75/1	10/1	2/1	--
9 Teams	150/1	20/1	5/2	--
10 Teams	250/1	25/1	3/1	1/1
11 Teams	400/1	60/1	5/1	1/1
12 Teams	900/1	80/1	10/1	3/1

Now, I did LOTS of math, and I had to make some assumptions, so here are the assumptions and results:

Assumption 1: Vegas is an efficient market. For Vegas to be perfectly efficient, it has to set spreads that win exactly 50% of the time. This means they always make money. So far this year, Vegas is batting roughly 49%. It's safe to assume that, over the long haul, Vegas wins.

Assumption 2: Teaser points ADD VALUE to your bet. I had to make assumptions that teasers tip odds of guessing correctly slightly in the bettors favor. This assumption is likely wrong, but without proper data, I'm not sure I could prove it either way. Instead, I give the bettor the benefit of the doubt.

I weighted progressive parlays so that the "cost per bet" includes the advantage you gain by allowing for losses an still getting payouts. The dollar values in the below chart are essentially cost of the payouts over time. Negative numbers are bad indicating you're losing money over time. Positive numbers are good in the long run. If I were to bet 3 team parlays all day long placing $1 bets (most minimums are $5, but $1 makes the math easy) with a win percentage of 52.38%, I would essentially make $0.01 per bet if I bet over a long enough time frame. Hardly worth it, I would say. I chose 52.38% because that's roughly the win percentage you need on individual bets to break even over time. Anything over that is profitable, anything under... well.. maybe betting's not for you. 52.38% is an estimate and makes the assumption that the juice is -110 for every game, so it's sort of an "ideal world" number. Good for the argument though.

My results are interesting, and may be hard to read, but here's a chart:

Prog = Progressive parlay
6 Pt Tease = 6 point teaser (the % are the % advantage getting points)
Chp Parl = cheaper parlays (some books have "Wacky Fridays" where the juice is lowered - here the juice is -105 instead of -110)
Reg Parl = regular parlay
Ind Bet = individual bets

		1.00%	1.50%	2.00%			Ind Bets
	Prog	6 Pt Tease	6.5 Pt Tease	7 Pt Tease	Chp Parl	Reg Parl	52.38%
1 Team Bet							$0.00
2 Team Bet		$0.03	$0.02	$0.00	$0.04	-$0.01	$0.00
3 Team Bet		$0.16	$0.14	$0.11	$0.06	$0.01	$0.00
4 Team Bet	-$0.64	$0.19	$0.15	$0.12	$0.09	-$0.17	$0.00
5 Team Bet	-$0.67	$0.16	$0.15	$0.13	$0.10	$0.03	$0.00
6 Team Bet	-$0.69	$0.12	$0.11	$0.11	$0.14	-$0.15	$0.00
7 Team Bet	-$0.83	$0.11	$0.10	$0.09	$0.17	-$0.18	$0.00
8 Team Bet	-$0.83	$0.09	$0.07	$0.06	$0.20	-$0.14	$0.00
9 Team Bet	-$0.84	$0.06	$0.05	$0.04	$0.22	-$0.11	$0.00
10 Team Bet	-$0.90	$0.04	$0.04	$0.03	$0.25	$0.09	$0.00
11 Team Bet	-$0.91	$0.03	$0.02	$0.02	$0.28	-$0.10	$0.00
12 Team Bet	-$0.91	$0.02	$0.02	$0.01	$0.31	-$0.23	$0.00
13 Team Bet		$0.02	$0.01	$0.01			$0.00
14 Team Bet		$0.01	$0.01	$0.01			$0.00
15 Team Bet		$0.01	$0.01	$0.01			$0.00

Basically, you need HUGE BALLS to win over time. This suggests because of the giant payout, the 12 team cheap day parlay is the best bet over time. Over a LOOOOOOOONG time. Here's the chart assuming a 55% win rate, very respectable and hard to hit:

		1.00%	1.50%	2.00%			Ind Bets
	Prog	6 Pt Tease	6.5 Pt Tease	7 Pt Tease	Chp Parl	Reg Parl	55.00%
1 Team Bet							$0.05
2 Team Bet		$0.08	$0.06	$0.05	$0.15	$0.09	$0.10
3 Team Bet		$0.22	$0.19	$0.16	$0.23	$0.16	$0.20
4 Team Bet	-$0.58	$0.25	$0.20	$0.16	$0.33	$0.01	$0.40
5 Team Bet	-$0.61	$0.22	$0.20	$0.18	$0.41	$0.31	$0.80
6 Team Bet	-$0.61	$0.17	$0.16	$0.15	$0.52	$0.13	$1.60
7 Team Bet	-$0.80	$0.16	$0.15	$0.14	$0.64	$0.16	$3.20
8 Team Bet	-$0.79	$0.13	$0.11	$0.10	$0.77	$0.26	$6.40
9 Team Bet	-$0.80	$0.10	$0.08	$0.07	$0.90	$0.39	$12.80
10 Team Bet	-$0.87	$0.07	$0.06	$0.05	$1.04	$0.78	$25.60
11 Team Bet	-$0.87	$0.05	$0.04	$0.03	$1.19	$0.53	$51.20
12 Team Bet	-$0.89	$0.04	$0.03	$0.02	$1.35	$0.38	$102.40
13 Team Bet		$0.04	$0.03	$0.02			$204.80
14 Team Bet		$0.03	$0.02	$0.02			$409.60
15 Team Bet		$0.02	$0.02	$0.01			$819.20

Clearly, bet every game individually here.

If it's a 45% win rate, here's your look:

		1.00%	1.50%	2.00%			Ind Bets
	Prog	6 Pt Tease	6.5 Pt Tease	7 Pt Tease	Chp Parl	Reg Parl	45.00%
1 Team Bet							-$0.14
2 Team Bet		-$0.11	-$0.12	-$0.13	-$0.23	-$0.27	-$0.28
3 Team Bet		$0.01	-$0.01	-$0.02	-$0.33	-$0.36	-$0.56
4 Team Bet	-$0.79	$0.04	$0.02	$0.00	-$0.41	-$0.55	-$1.13
5 Team Bet	-$0.83	$0.04	$0.03	$0.02	-$0.48	-$0.52	-$2.25
6 Team Bet	-$0.86	$0.03	$0.02	$0.02	-$0.54	-$0.66	-$4.51
7 Team Bet	-$0.93	$0.03	$0.02	$0.02	-$0.60	-$0.72	-$9.02
8 Team Bet	-$0.94	$0.02	$0.02	$0.01	-$0.65	-$0.75	-$18.04
9 Team Bet	-$0.95	$0.01	$0.01	$0.01	-$0.69	-$0.77	-$36.07
10 Team Bet	-$0.97	$0.01	$0.01	$0.00	-$0.73	-$0.76	-$72.15
11 Team Bet	-$0.98	$0.00	$0.00	$0.00	-$0.76	-$0.83	-$144.29
12 Team Bet	-$0.98	$0.00	$0.00	$0.00	-$0.79	-$0.88	-$288.58
13 Team Bet		$0.00	$0.00	$0.00			-$577.16
14 Team Bet		$0.00	$0.00	$0.00			-$1,154.33
15 Team Bet		$0.00	$0.00	$0.00			-$2,308.65

Now, teasers seem useful for LOSERS.

Conclusion: PROGRESSIVE PARLAYS SUCK SCROTE. I will not be betting them in the future unless I can get my win rate up. Why you ask? It's seems counterintuitive you say? Well, it's easy really. The payouts are TERRIBLE. Comparatively, a regular parlay pays out 10/1 on a 4 team parlay. Nice payout. But the progressive? 5/1 on the same parlay, because you're "paying" for the ability to lose a game. The payout on hitting the 3 out of 4? 1/1. A 3 team straight parlay pays 6/1, so you are giving up essentially 500% to be "progressive". You are far better off paying for total combination of parlays (1 four team parlay, and 3 three team parlays) because it costs less in the end.

So teaser and parlay madness in the future? I think not. I'll stick to my winning guns.