just to inform you all, i will join the trend market and will offer the trends i found on my own, none of your trends will be included. the trends i found will be featured in the spreadsheet too, the trend market has more value to the people not so familliar with SDQL in my opinion.
-
-
Hey Nash, check out NBA 72. It's designated as an ATS trend, but it actually does a bit better as an Over trend (after a 2-10 start ATS this season). Do you want to add it again designated as a over trend so the Analyzer sees it both ways (I think that's how it works)?Comment -
i will take a look at it right now.Comment -
tA(P4) -5 > oA(P4) and game number >= 10
is 127-113 on totals. that is not valid enough for me.Comment -
i think you mean Nba73Comment -
and 73 and 125 are dividing the trend in under and ats betsComment -
-
It was a great day for the system yesterday. If only the wizards would have covered, the profit would have been off the charts.
+21 units since I started tracking with the program
and that's in one weekComment -
i need your gmail account to connect it to the spreadsheet. you need min of 50 posts, after that you can message me.Comment -
Comment -
how do I transcribe this in sdql?, "halftime margin of negative 6 to negative 10". thanx in advance!Comment -
Action on the Bulls/Pelicans tonight. 3,13,21,81,189 all have Bulls -2. My Bulls are a mess right now but hopefully they snap out of it long enough to slow down The Brow.
97,211,215 are also on Over 196.5 for this game.Comment -
Bulls devastated Pelicans with a 35-point lead
Comment -
Not exactly. Game was tied 35 - 35 when Davis, best player in the league this season got hurt, left the game and didn't come back. Not saying Pelicans would have won with him, but you can't ignore that as well...Comment -
So the question is, let's assume that Anthony was announced not to be playing for the game before it started, would the line have been -35 then ?
Comment -
No, but would it change the queries that fit? I won't check them each now, but maybe we would have seen queries supporting Pelicans or one two queries would have fallen off from those that fit the Bulls?Comment -
Every day, for maybe a a month and a half or so, I have been running a program to go through the Nash google sheet as well as my own compilation of queries, looking for the active ones for that day. At the end of the day I mark the queries as having won or lost, and store them off in a spreadsheet.
So far I have stored off around 520 queries/results, and around 201 of them were winners and 320 were losers. I am still engaged in looking for patterns and trying to determine the best way to weed out the bad queries from the good ones.
But what seems kind of striking to me at this point is that 520 is a pretty large sample size. The question I have started asking is, how big a sample size would I need of this, before I could be confident in keeping all the queries I have, just as they are, and then simply fading them rather than playing them? I don't think 520 is a big enough sample size to justify doing so, but what sample size would justify it? 1000? 5000? 10,000?
If I stored off 1,000 of these query results, and they were still losing at a 60% clip, would that sample size justify simply fading them?Last edited by pip2; 02-08-15, 10:46 AM.Comment -
I'm doing a somewhat similar thing. First, I'm backtesting all the trends on the spreadsheet for 4+ seasons ('14 YTD, '13,'12,'08,'06). I'm only thru the first 100 trends, but that backtest shows really good results (60%+) for all the previous full seasons, but '14 is only 523-495 so far. I'm keeping a running list of trends that seem to obviously be weak, and I will use this growing list to filter my plays going forward. I'll be curious to see if the numbers stay flat for '14 as I work my way thru the remaining 2/3 of the trends. It seems statistically unlikely, but Nash13 may be right to look very suspiciously at how the trends project into the future.
In the meantime, I've been using the analyzer software to play all non-conflicting trends with small units. This should theoretically be better than the raw backtest, because many conflicting plays are eliminated. Since 1/27 those trends have gone 99-114 (was actually doing OK until the last two days, which have gone 11-34).
NHL has been considerably better, going 79-56, +18.5 units since 1/27. I did a 3 season backtest there that was also positive, but I didn't do '14 YTD, which I will probably do once I finish with NBA. It takes a significant time commitment to comb thru each season for each trend, so I'm just plugging away as time allows.
*I've liked the NHL more all along, as my overall impression of the NBA trends is that we may have too many smaller trends, which allows for more volatility.Last edited by Cutler'sThumb; 02-08-15, 11:40 AM.Comment -
One impression I get from reading down my saved list of results, is that in one of those days, say, when there are 10 winners and 15 losers, or maybe even 20 losers and 5 winners, there are some names frequently associated with those rare winning queries: Nash, Hiyahya, and Jmon. Their names might be associated with other winners that are simply labeled as "NBAXX" as well.
It isn't that simple, because they have a lot of queries in general, so they also have some losing queries associated with their names, but I wonder if I might also fix my query libraries by just weeding out everything that wasn't approved/assembled by one of those guys...Comment -
i 100% aggree with both posts above. just as in other terms they would say: "it is not about what you have done in the past, it is about what you can do in th future."
so there is no point in finding a 70% ATS trend which will go forward 50/50 or even worse from now on.
major point: what can we do about it?
i see the trend sheet as a pool. not every trend is good or value, some are overfitted other are just sound. i declare statistical and logical criteria and after that i take all things together with things the queries don't show. and then i play the games.
so far i am YTD up 40 units. if that stays like this, i am ok. as long as i am not loosing money everything is fine.Comment -
I agree with you but along with being fine if I make money, I am perfectly fine if the queries I currently have can consistently lose at a 60% rate, because for me that isn't materially different than winning 60% consistently. My problem is that I am not sure of how big a sample size I need to have in order to declare that the 60% rate is consistent enough to bet on..Comment -
I can't agree with you here.
If query is good, that means that the logic behind it was correct and it brings profit.
If query is bad, it just means that the logic didn't prove itself and that the query shouldn't be relied on, but it doesn't mean the opposite, so even if in the short run fading it should be good, I wouldn't be as sure about it in the long run, as I would be about a query that is good to begin with.Comment -
How about if you had 100,000 query results saved and 60,000 of them lost over the past couple of years? Wouldn't you feel confident about fading them? I would. But what I am wondering is where is the number between 100,000 and 500 where there isn't much of a chance that the 60% figure is just variance?Comment -
I don't know.
For example, if I assume that good rebounding team does good against bad rebounding team, if it comes off a loss where they lost the rebounds battle (just an example of course) and I see that numbers support me and I use it, what changed in the league, that suddenly it should just stop working?Comment -
I'm playing with numbers trying to quantify why the overall numbers this year are so different from past years. Thru the first 100 trends, I have a total record of 553-532, 51.34%. I'm sure there would be many conflicting plays if we looked at each day and the win rate would go up, but this win % is still a significant departure from past seasons.
One of my suspicions is that the smaller trends are hurting the number. Small sample size from the past means it is much easier/more likely for a trend to experience a correction to the mean (especially if the trend is overfitted). So I decided to eliminate all trends that had samples less than 150 going into the season. This eliminated 65 of the first 100 trends. I then added in three other trends from my working list of trends that were on my "weak" list (I had 11 others, but they were in the 65 I already eliminated). This took the record to 248-205, 54.75%.
I'll keep working on the backtest, but I think there may be an important lesson here: beware small trends. They can be good, but they can also turn quickly, which can be hard to catch when we're dealing with this many trends.Comment -
OK, so here are my picks/units for today, fading today's queries:
mil/bkn O 2 units
atl L 2 units
mia L 1 unit
ut O 1 unit
ut L 2 units
ind W 8 units
phi L 2 units
phi U 1 unit
sa U 1 unit
okc W 2 units
okc U 2 units
Let's see how this goes...last night my query library went 6-16, which is actually 16-6 if you are fading them.Comment -
Nice work Cutler. One thing I have come to believe about the sample size is that it might be good to not consider it by itself, but as a function of time. For example, two queries could have the same sample size of 200, but one goes back 20 years to get that 200, while the other goes back 2 years to get the 200. So the sample sizes are not really the same. Are you factoring that into deciding which queries to eliminate?Comment
, but you obviously figured out what I meant)
