How did you do for 2018/2019 regular season?

Home for all your discussion of basketball statistical analysis.
Post Reply
vzografos
Posts: 24
Joined: Thu Sep 06, 2018 10:42 am

How did you do for 2018/2019 regular season?

Post by vzografos » Tue Jun 11, 2019 10:00 am

So if you guys are predicting the outcome of regular season games (win/loss and perhaps line or probability) how did you perform this season?

In terms of accuracy I have been tracking some sources. Here are their results:

BETFAIR exchange: 66.75%
FIVETHIRTYEIGHT CARMELO: 65.77%
COVERS.com 66.91%
ODDSHARK: 67.07%

Also some more tracking here: http://www.thepredictiontracker.com/nbaresults.php
But I think that includes playoffs as well.

Can you do any better? Perhaps reach towards 70%?
Let me know 8-)

Crow
Posts: 6250
Joined: Thu Apr 14, 2011 11:10 pm

Re: How did you do for 2018/2019 regular season?

Post by Crow » Wed Jun 12, 2019 6:10 am

That is a pretty tight distribution. Are these results self-reported or independently verified and is the verification independently verified?

What would that work out to in terms of expected rate of return if all predictions were bet equally after house vigorish?

Does a lot of bettor loss come from not playing the predictions evenly? I was told by one bettor previously that playing a lot of bets per night (like 5-7 ) was unusual or perhaps not wise. I don't know if others agree, disagree or "it depend''. I haven't ever taken it seriously. Not saying that was smart, but I haven't.

Assume you bet regularly? Have you had problems with payouts or the law? Or bad streak liquidity?

Crow
Posts: 6250
Joined: Thu Apr 14, 2011 11:10 pm

Re: How did you do for 2018/2019 regular season?

Post by Crow » Wed Jun 12, 2019 9:43 am

Different test but we have a thread on team win projections for season.

vzografos
Posts: 24
Joined: Thu Sep 06, 2018 10:42 am

Re: How did you do for 2018/2019 regular season?

Post by vzografos » Wed Jun 12, 2019 12:26 pm

Crow wrote:
Wed Jun 12, 2019 6:10 am
That is a pretty tight distribution. Are these results self-reported or independently verified and is the verification independently verified?

What would that work out to in terms of expected rate of return if all predictions were bet equally after house vigorish?

Does a lot of bettor loss come from not playing the predictions evenly? I was told by one bettor previously that playing a lot of bets per night (like 5-7 ) was unusual or perhaps not wise. I don't know if others agree, disagree or "it depend''. I haven't ever taken it seriously. Not saying that was smart, but I haven't.

Assume you bet regularly? Have you had problems with payouts or the law? Or bad streak liquidity?
These results are publicly available and can be verified independently.
If you have the patience to record the data manually or even better build some script to download historical data.
I have been scraping various websites with information, predictions and statistics and been tracking predictions for the last season.

As you can see the market (betting exchange) is not that efficient and some "expert" websites, such as Oddsshark can do better. Of course the best performance comes not from empirical information but from machine learning which can do around 70% (at least for 2018/19).

Now accuracy is one thing. The other is probabilities. How close you can predict to the actual probability of the event (see Brier score).

There are two things required in order to make consistent money long term from "betting" (let's call it sports-trading from now on and not gambling). A good prediction algorithm/method that you can predict better than the market (accuracy and Brier score) and very important, a Trading Strategy (call if money magagement if you wish). I think that is more important.

There are many Trading Strategies and wagering on all predictions equally is not a great one, unless your predictions are so much more accurate than the market (doubt that). So more often than not equal betting show a consistent loss. You can do a bit better and use Kelly-related strategies or something more advances like porfolio-management theory.

Actually, once you start treating sports-trading like any other trading (e.g. stocks, forex etc) there are many similarities. By far the best strategy is to spread your wagers across different events, where you have a positive edge, in order to maximise your return and minimise your risk. How you spread it, well that is a little bit complicated and it involves a bit of math, but it takes a lot of the risk and volatility out of sports-trading.

Regarding law and betting. Well I am in Europe so we have a bit more lax regulations than you guys. Betting and online betting is legal throughout and in some countries (e.g. UK) it is not even taxed :)

vzografos
Posts: 24
Joined: Thu Sep 06, 2018 10:42 am

Re: How did you do for 2018/2019 regular season?

Post by vzografos » Wed Jun 12, 2019 12:29 pm

Crow wrote:
Wed Jun 12, 2019 9:43 am
Different test but we have a thread on team win projections for season.

yeah I noticed. I would be very interested when the next season begins to see if we can do some early predictions on per-game results and who is going to make it to the playoffs.
I generally stop when the regular season stops and do not bother after the playoffs. Reason is that there is a completely different set of statistical data for the playoff season (yes teams tend to play differently after the regular season) and I haven't got round to getting the playoff data from stats.nba.com

eminence
Posts: 141
Joined: Sun Sep 10, 2017 8:20 pm

Re: How did you do for 2018/2019 regular season?

Post by eminence » Wed Jun 12, 2019 3:57 pm

For fun I tried for a stretch this season and went 91-30 (moneyline). Just my own intuition for each game, no real analysis done. No money here, so no money made unfortunately, lol.

vzografos
Posts: 24
Joined: Thu Sep 06, 2018 10:42 am

Re: How did you do for 2018/2019 regular season?

Post by vzografos » Thu Jun 13, 2019 4:47 am

Image

So here is a graph showing the prediction accuracy for each of the different providers. The accuracy is accumulated after each game, with a total of 1230 games in the regular season.

Observe the noisy nature between 0-200 games, which is quite expected for any type of accumulated statistic.
Also, notice the drop between 200-400 after which all curves begin to correct and finalise at their respective numbers. I think this may be due to the early uncertainty about the season, since team & player average statistics take some time to stabilise.
Finally, notice the high-correlation amongst all these curves. I wonder if this is because of sharing of information between predictions or because this is the nature of the prediction variables.

In any case. Something to think about. I would be really interested how to avoid that dip between 200-600 games.

Mike G
Posts: 4429
Joined: Fri Apr 15, 2011 12:02 am
Location: Asheville, NC

Re: How did you do for 2018/2019 regular season?

Post by Mike G » Thu Jun 13, 2019 12:30 pm

That dip at 30-50% of the season is likely due to these providers' insistence on regressing to their prior expectations. As the season wore on, it became unavoidably clear the Clippers and Bucks and Blazers were actually doing this well, and others were going to fall below pre-season predictions.

vzografos
Posts: 24
Joined: Thu Sep 06, 2018 10:42 am

Re: How did you do for 2018/2019 regular season?

Post by vzografos » Thu Jun 13, 2019 1:11 pm

Mike G wrote:
Thu Jun 13, 2019 12:30 pm
That dip at 30-50% of the season is likely due to these providers' insistence on regressing to their prior expectations. As the season wore on, it became unavoidably clear the Clippers and Bucks and Blazers were actually doing this well, and others were going to fall below pre-season predictions.
Difficult to say.
Actually the data from COVERS is I believe consensus data from users of the site. I really think it is market sentiment. Now how everyone is doing their estimates is hard for me to say.

More than anything I find the correlation very interesting

Post Reply