Re: Pirateboy\'s NCAAF ML Dogs Experiment - W9
 

silentbob- I don't think he has fine-tuned it. Or at least he hasn't mentioned it here. Are you talking about kdog's post?

Re: Pirateboy\'s NCAAF ML Dogs Experiment - W9
 

Yeah, sorry. I thought you were responding to kdog. Under the fine-tuned system, the only ones betting (based on my approximations) are:

Maryland (although I can only find +140 right now)
Ohio (available at +250 some places)
N. Illinois
North Carolina
North Carolina State

I bet a couple of these for other reasons.

Re: Pirateboy\'s NCAAF ML Dogs Experiment - W9
 

As others have pointed out, the problem with this approach is that the ranges used are too broad; there is a huge difference between a +3.5 dog and a +6 dog. I ran the numbers for road underdogs since 1996 and got the straight-up W/L record for each individual line. I think this illustrates the problems with using broad ranges.

NCAAF Road Underdogs since 1996
<font class="small">Code:</font><hr /><pre>
Line Wins Losses Win % Breakeven ML
+1 57 47 0.548 -121
+1.5 46 39 0.541 -118
+2 28 32 0.467 114
+2.5 79 63 0.556 -125
+3 112 141 0.443 126
+3.5 55 82 0.401 149
+4 40 66 0.377 165
+4.5 28 82 0.255 293
+5 21 41 0.339 195
+5.5 48 65 0.425 135
+6 40 79 0.336 198
+6.5 44 98 0.310 223
+7 59 126 0.319 214
+7.5 40 104 0.278 260
+8 20 44 0.313 220
+8.5 23 80 0.223 348
+9 18 40 0.310 222
+9.5 23 84 0.215 365
+10 35 94 0.271 269
+10.5 14 67 0.173 479
+11 21 36 0.368 171
+11.5 17 37 0.315 218
+12 13 40 0.245 308
+12.5 22 53 0.293 241
+13 17 70 0.195 412
+13.5 23 88 0.207 383
+14 23 112 0.170 487
</pre><hr />

These numbers are interesting, but they also show that you can't use extremely narrow ranges (exact lines) for this type of analysis, either. For example, +5.5 dogs performed *better* than +4.5 dogs by a huge margin. Does that mean we should blindly bet all +5.5 dogs and all -4.5 faves? Probably not. Much more likely that it is just a random effect, even over these fairly large sample sizes.

A better option might be to use a range size in between the two extremes, say +/- half a point, so the ranges would be +1..+2, +1.5..+2.5, +2..+3, +2.5..+3.5, and so on. The line for a particular game would fall into three of these ranges, so if the moneyline showed value for all three, it would probably be a good bet. I'll post these numbers in a bit.

Regardless, I think that the most important thing to get out of this is that these numbers should not be your sole reason for placing a bet. However if you already like a particular road underdog, the fact that the moneyline is better than typical for the given spread would be another factor to consider, that adds a little to your already expected edge.

The exception is with road dogs &lt;= +2.5, which as we've discussed in some other threads, have actually won straight-up 53.7% of the time. I think that, particularly in games expected to be low-scoring, these are good enough to blindly pick unless there are extenuating reasons to pick against them. (Other than cases like Akron this week where they are a dog on the spread but ML is -105.)

Re: Pirateboy\'s NCAAF ML Dogs Experiment - W9
 

mogwai- I assume you have that data in an XL spreadsheet. What fields to you have to make it easy to sort? The raw data I downloaded has DATE, VISITOR, VISITOR SCORE, HOME TEAM, HOME SCORE, LINE. I obv. need to create a column for who covered and for who won the game straight up. Do you have the team names in those columns, or Home/Visitor? I'm just wondering which would make it easier to sort. Any other fields that you have?

(anyone else feel free to chime in as well.)

Re: Pirateboy\'s NCAAF ML Dogs Experiment - W9
 

[ QUOTE ]

A better option might be to use a range size in between the two extremes, say +/- half a point, so the ranges would be +1..+2, +1.5..+2.5, +2..+3, +2.5..+3.5, and so on. The line for a particular game would fall into three of these ranges, so if the moneyline showed value for all three, it would probably be a good bet. I'll post these numbers in a bit.


[/ QUOTE ]

NCAAF Road Underdogs since 1996
<font class="small">Code:</font><hr /><pre>
Line Wins Losses Win % Breakeven ML
+1..+2 131 118 0.526 -111
+1.5..+2.5 153 134 0.533 -114
+2..+3 219 236 0.481 108
+2.5..+3.5 246 286 0.462 116
+3..+4 207 289 0.417 140
+3.5..+4.5 123 230 0.348 187
+4..+5 89 189 0.320 212
+4.5..+5.5 97 188 0.340 194
+5..+6 109 185 0.371 170
+5.5..+6.5 132 242 0.353 183
+6..+7 143 303 0.321 212
+6.5..+7.5 143 328 0.304 229
+7..+8 119 274 0.303 230
+7.5..+8.5 83 228 0.267 275
+8..+9 61 164 0.271 269
+8.5..+9.5 64 204 0.239 319
+9..+10 76 218 0.259 287
+9.5..+10.5 72 245 0.227 340
+10..+11 70 197 0.262 281
+10.5..+11.5 52 140 0.271 269
+11..+12 51 113 0.311 222
+11.5..+12.5 52 130 0.286 250
+12..+13 52 163 0.242 313
+12.5..+13.5 62 211 0.227 340
+13..+14 63 270 0.189 429
+13.5..+14.5 46 200 0.187 435
+14..+15 23 112 0.170 487

</pre><hr />


As an example, my lines on UNC today were +5.5 ATS and +200 ML. +5.5 falls into the ranges +4.5..+5.5, +5..+6, and +5.5..+6.5. The break-even moneyline values for these three ranges are +194, +170, and +183, so +200 is likely a good value on the moneyline. However this line is pretty close to the anomalous +4..+5 range, which has a BE ML of +212. In any case, since I already felt that UNC had greater than a 1/3 chance of winning the game, this data added to my belief that the +200 ML bet had a significant edge.

Re: Pirateboy\'s NCAAF ML Dogs Experiment - W9
 

[ QUOTE ]
mogwai- I assume you have that data in an XL spreadsheet. What fields to you have to make it easy to sort? The raw data I downloaded has DATE, VISITOR, VISITOR SCORE, HOME TEAM, HOME SCORE, LINE. I obv. need to create a column for who covered and for who won the game straight up. Do you have the team names in those columns, or Home/Visitor? I'm just wondering which would make it easier to sort. Any other fields that you have?

[/ QUOTE ]

My data does have several more columns, but they aren't too relevant here - what you have should work just fine. First filter out everything except road underdogs (or whatever subset you want to look at). I made a column "SU Result" that has a value of 1 if the road team won and -1 if the road team lost. I made a cell that did subtotal(9, ) of that column. Then I can just autofilter on the line or range of lines I'm interested in, and autofilter on the SU Result column for 1 to get the wins count and -1 to get the losses count. I'm sure there are ways to automate it even more, but this works for me.

Re: Pirateboy\'s NCAAF ML Dogs Experiment - W9
 

I just placed a bet on UNLV +350. They are a 10 point dog on the spread but you have to go all the way to +13..+14 before you see a break-even ML worse than +350.

Re: Pirateboy\'s NCAAF ML Dogs Experiment - W9
 

[ QUOTE ]
[ QUOTE ]
No value on the Fresno ML on Friday night, so here are the Saturday plays:

Connecticut +170

North Carolina +200

Pittsburgh +330

Texas A&amp;M +140

UNLV +10

USC +130

Arizona +150

EMU +175

Georgia +255

Maryland +145

NC State +155

Ohio +240

South Carolina +125

Northern Illinois +145

Penn State +150

Cal +135

ULM +215

Arkansas State +160

North Texas +425

[/ QUOTE ]

Maybe I don't understand, but it seems you are going against that article now. You have a lot of bets there on teams that are 3-6.5 point dogs, but you aren't getting the +166 necessary to qualify for your experiment. NC State, Penn State, Cal, Maryland...

[/ QUOTE ]

I have since received an even better data set, thus, I can get far more specific.

Re: Pirateboy\'s NCAAF ML Dogs Experiment - W9
 

I noticed I put "UNLV +10" which is obv wrong. It was +350.

Re: Pirateboy\'s NCAAF ML Dogs Experiment - W9
 

[ QUOTE ]
[ QUOTE ]
[ QUOTE ]
No value on the Fresno ML on Friday night, so here are the Saturday plays:

Connecticut +170

North Carolina +200

Pittsburgh +330

Texas A&amp;M +140

UNLV +10

USC +130

Arizona +150

EMU +175

Georgia +255

Maryland +145

NC State +155

Ohio +240

South Carolina +125

Northern Illinois +145

Penn State +150

Cal +135

ULM +215

Arkansas State +160

North Texas +425

[/ QUOTE ]

Maybe I don't understand, but it seems you are going against that article now. You have a lot of bets there on teams that are 3-6.5 point dogs, but you aren't getting the +166 necessary to qualify for your experiment. NC State, Penn State, Cal, Maryland...

[/ QUOTE ]

I have since received an even better data set, thus, I can get far more specific.

[/ QUOTE ]

[img]/images/graemlins/grin.gif[/img] [img]/images/graemlins/grin.gif[/img] [img]/images/graemlins/grin.gif[/img] [img]/images/graemlins/grin.gif[/img] [img]/images/graemlins/grin.gif[/img] [img]/images/graemlins/wink.gif[/img]