NFL 1st Half Unders in Primetime

Bob Voulgaris, newly minted director of quantitative research and development for the Dallas Mavericks, is easily the most famous NBA sports bettor out there (although he's now retired from sports gambling and just stays active on NBA Twitter). One of his original profitable strategies was exploiting how sportsbooks set their NBA over/unders for each half:

It all had to do with how most bookmakers set their halftime totals, the predicted number of points scored in each half of the game. Each half, of course, is its own discrete period of play, and the fourth quarters of close games can end in elongated foul-clogged stretches of free throws, timeouts, fast play and, hence, a burst of scoring. But incredibly, bookmakers at the time didn’t account for this fact; they simply arrived at a total for the full game and cut that figure roughly down the middle, assigning some 50 percent of the points to the first half and 50 percent to the second.

Back in 2015, a former coworker came across a similar phenomenon in NFL first half over/unders, resulting in a considerable edge in specific situations. His strategy was so successful that it started 12-0 during the 2015 season, until we promptly bet it heavily on all 3 Thanksgiving Day games that year and it went 0-3.

Nonetheless, I went back and analyzed the past 3 full seasons of NFL first half totals (2015-2017) and found that the strategy still holds up remarkably well. The under has an abnormal success rate in games in which the first half over/under is >= 21 points, and is even more magnified in games that occur in primetime.

	O/U >= 21
All Games	WIN	338
	LOSS	264
	Push	19
	Total	621
	Win %	56.15%

56.15% is a very good success rate, but this market inefficiency skyrockets when the game is in "primetime":

	O/U >= 21
Primetime Only	WIN	91
	LOSS	46
	Push	5
	Total	142
	Win %	66.42%

I'm considering "primetime" to be any game that is the only game on - so Thursdays, Sunday nights, Monday nights, the London games (Sunday mornings), or the miscellaneous Saturday games late in the season. In other words, not in the afternoon on a Sunday.

Why does this edge occur in those spots? Over the years we've settled on the theory that players aren't in their normal rhythm that they experience most other Sundays, whether that's due to a shortened/extended week of practice, or just the game day pattern extending into the night. Therefore, they start slow during the game, resulting in less points in the first half. Regardless of the reason, this strategy hits at almost a 2/3 clip (66.42%).

The threshold of only betting first half O/U of 21+ results in the most profitable yield, but these effects stand up regardless of the number, just at a lower success rate:

	ALL O/U
All Games	WIN	402
	LOSS	344
	Push	22
	Total	768
	Win %	53.89%
	ALL O/U
Primetime Only	WIN	104
	LOSS	64
	Push	6
	Total	174
	Win %	61.90%

Parlaying O/U and ATS in NCAAF Relative to One Another

After my recent post on parlaying O/U and ATS depending upon the size of the spread, I received a request to look at the ratio of spreads and totals in NCAAF. So situations in which the spread is an outlandish percentage of the total, like 35%+. For example, in today's Georgia-Vanderbilt game, Georgia is favored by 25.5 with an over/under of 56. That 25.5 point spread thus represents 45.5% of the overall expected points in the game (25.5/56), which is a fairly lopsided ratio. Is there value in games where one team's score alone is supposed to make up a good portion of that game's total points?

As before, a standard 2-team payout is 13/5 (+260), and at +260, you need to hit above 27.78% of the time to turn a profit. (Note: with heavily skewed NCAAF games, most books actually don't let you parlay a side and an over/under. But this strategy can be applied to individual bets as well). So I looked at varying ratios of spreads to totals over the past 5 years of college football, and combining bets on the favorite and the over is a remarkably profitable strategy as the ratio escalates:

Ratio of Spread to Total
FAV + OVER	Actual	%	Return	ROI	% Occur
20%	414	27.62%	-9	-0.57%	42.18%
25%	339	28.70%	39	3.34%	33.23%
30%	273	30.67%	93	10.43%	25.04%
35%	215	32.33%	109	16.39%	18.71%
40%	158	32.71%	86	17.76%	13.59%
45%	114	34.03%	75	22.51%	9.43%
50%	76	33.78%	49	21.60%	6.33%
55%	58	37.66%	55	35.58%	4.33%
60%	41	37.27%	38	34.18%	3.10%
65%	24	37.50%	22	35.00%	1.80%

The more outlandish the ratio is, the more profitable it is, but of course the higher that ratio gets, the less often it occurs. If you can get any favorite that makes up 30%+ of the overall total, you're looking at a very high ROI.

I also looked at dogs + unders, in case you want to cover your risk a bit and hedge on the other extreme. There is some value in doing so, but only once you get to the sweet spot of a 45-60% ratio:

Ratio of Spread to Total
DOG + UNDER	Actual	%	Return	ROI	% Occur
20%	403	26.88%	-48	-3.22%	42.18%
25%	320	27.10%	-29	-2.46%	33.23%
30%	230	25.84%	-62	-6.97%	25.04%
35%	178	26.77%	-24	-3.64%	18.71%
40%	134	27.74%	-1	-0.12%	13.59%
45%	100	29.85%	25	7.46%	9.43%
50%	71	31.56%	31	13.60%	6.33%
55%	49	31.82%	22	14.55%	4.33%
60%	35	31.82%	16	14.55%	3.10%
65%	17	26.56%	-3	-4.38%	1.80%

NBA Spreads of 10+ and Totals of 200+, Are Unders More Likely

A former coworker suggested this hypothesis: when looking at large NBA spreads (say where a team is favored by 10+), coupled with a large total (like 200+), are unders more likely? The reasoning being that the heavy favorite is likely to run away with the game (regardless of whether they cover or not) and then coast for a large part of the game, leading to lower scoring.

I looked at the past 5 seasons of NBA spreads and totals, and first I just checked ATS of 10+ and O/U of 200+. This occurred 7.38% of the time, and returned an "under" percentage of 51.61%, which isn't enough to hurdle the juice at -110 (52.38% is needed).

Next I ran a non-linear optimization algorithm to maximize the percentage of unders, while changing the spread and totals variables. And there is a combo threshold that results in fairly optimal unders cashing: spreads of 9+ and totals of 210+ saw the under hit 54.73% of the time. This situation occurs much less frequently, only showing up 3.18% of the time.

NBA Unders	% Occur	Win %	Return	ROI
ATS 10+, O/U 200+	7.38%	51.61%	-7	-1.43%
ATS 9+, O/U 210+	3.18%	54.73%	9	4.53%

Parlaying O/U and ATS

Every sporting event with a spread and an over/under has four possible combinations of outcomes, disregarding a push:

• FAV + OVER
• FAV + UNDER
• DOG + UNDER
• DOG + OVER

Most books let you parlay these combinations together on the same event, for the standard 2-team payout of 13/5 (+260). So is there any value to doing so? In theory, each combination should occur ~25% of the time. At +260, you need to hit above 27.78% to turn a profit (parlays are very rarely an efficient bet compared to individual ones).

I looked at the last 5 seasons for NBA, NFL, and NCAAF (NCAAB data can be spotty, especially for O/U) to see if any trends stood out. Over that many games nothing hurdles above 27.78%:

NBA	Actual	%	Exp	Dif
FAV + OVER	1,515	23.94%	1,582	-66.75
FAV + UNDER	1,638	25.89%	1,582	56.25
DOG + UNDER	1,561	24.67%	1,582	-20.75
DOG + OVER	1,613	25.49%	1,582	31.25
Total	6,327	100.00%
NFL	Actual	%	Exp	Dif
FAV + OVER	282	24.02%	294	-11.5
FAV + UNDER	289	24.62%	294	-4.5
DOG + UNDER	304	25.89%	294	10.5
DOG + OVER	299	25.47%	294	5.5
Total	1,174	100.00%
NCAAF	Actual	%	Exp	Dif
FAV + OVER	917	25.80%	889	28.5
FAV + UNDER	857	24.11%	889	-31.5
DOG + UNDER	920	25.89%	889	31.5
DOG + OVER	860	24.20%	889	-28.5
Total	3,554	100.00%

What about at the extremes, when one team is favored by a lot? Some small trends did stand out on larger spreads, albeit at smaller sample sizes of course.

NBA really doesn't do much. 7% of games matched the criteria of a spread >= 12, and there's not much of a return to be had:

NBA, ATS >= 12	Actual	%	Return	ROI
FAV + OVER	80	18.02%	-156	-35.14%
FAV + UNDER	124	27.93%	2	0.54%
DOG + UNDER	129	29.05%	20	4.59%
DOG + OVER	111	25.00%	-44	-10.00%
Total	444	100.00%

NFL does it a bit better. 12% of games had spreads >= 10, and there's some nice value in combining the underdog and the under:

NFL, ATS >= 10	Actual	%	Return	ROI
FAV + OVER	41	28.08%	2	1.10%
FAV + UNDER	30	20.55%	-38	-26.03%
DOG + UNDER	45	30.82%	16	10.96%
DOG + OVER	30	20.55%	-38	-26.03%
Total	146	100.00%

In the end, NCAAF has the largest opportunity, with 18% of games having spreads >= 20.5 and some really nice value pairing the favorite and the over:

NCAAF, ATS >= 20.5	Actual	%	Return	ROI
FAV + OVER	204	31.83%	93	14.57%
FAV + UNDER	129	20.12%	-177	-27.55%
DOG + UNDER	164	25.59%	-51	-7.89%
DOG + OVER	144	22.46%	-123	-19.13%
Total	641	100.00%