VERY SHORT-LIVED THREAD: Your best #s for 3Q16 Delivery Numbers

[Ulmo]

Just missed the podium at 5th place.

Thanks for setting up the sheet.

Your estimate was by far the closest at the time you placed your estimate, message #5 of the thread, off by only -175, with the next best already placed estimate off by -407. However, at that time, the running average for the 3rd estimate was 24,548, only off of the actual result by 48, so it is possible that someone in your position could have looked at prior estimates to come up with a number. But that's not the best news: your estimate placed the running average at the CLOSEST it ever got, at -8 of the real number (which itself is only an estimate with accuracy of a hundred). If we had stopped the thread after you posted, we would have had the correct number, and that wasn't true again ever.

Looks like median estimate was 24,800 -- only about 300 off or 1.2%. And maybe closer after final numbers are in.

Pretty impressive!

Hmm. I forgot "median" existed. I'll calculate it, too:

According to my spreadsheet, of those in this thread, the median was 24,775, off by -275. It got closest by posting time when @mrdoubleb & @Bgarret placed their estimates with message #s 20 & 22 respectively, at median 24,555, off by -55.

If sorted by join date (and including 4 estimates from another thread), it was already pretty accurate by @tlo (joined 2011-12-09), only 4th in line after the first three, with median 24,434, off by 66. It didn't get that close again until @AudubonB joined 2013-03-24, with -68 off. The last time the median error stayed under 100 was off by -80 with join date 2013-08-24, bonaire, and then it just diverged above error 100. It started staying in the 200 error range around join date 2015-06-01, and rose above 300 error for join date 2016-05-22 and never returned, ending up at 24,816, off by -316 (so that other thread threw it off a little more).


I'm noticing that the idea of filtering out by when they joined came up with some interesting results that got the numbers to conform in different trends. I played a little by trying to weight them, but didn't get far with that. The plain average was off by 532 (including 4 from other threads). The closest I got was multiplying by number of TMC posts, which got the average to 24,399, off by -101. Another attempt was multiplying by number of days on TMC, which got the average to 24,318, off by -182. Another weird one was adding days on TMC + number of posts, which got me to 24,239, off by -261. But the best that can be done is simply discarding all estimates of those who joined TMC after 2013-04-05, getting an average off by 6, and a median off by -80, not far off from the best median of off by 66. In a pinch, the average never got off by more than 66, with it being -24 off, for join dates up to 2013-06-09. After there, it diverged, due to a guess of 31,600, by @uselesslogin. It would have stayed good until join date 2015-05-19 if that hadn't happened, and the median would have stayed good until 2013-08-30. Average stayed under 150 off in that case until join date 2015-11-12, and median stayed under 200 off except for one up until join date 2016-01-02.

 

[Crowded Mind]

Your estimate was by far the closest at the time you placed your estimate, message #5 of the thread, off by only -175, with the next best already placed estimate off by -407. However, at that time, the running average for the 3rd estimate was 24,548, only off of the actual result by 48, so it is possible that someone in your position could have looked at prior estimates to come up with a number. But that's not the best news: your estimate placed the running average at the CLOSEST it ever got, at -8 of the real number (which itself is only an estimate with accuracy of a hundred). If we had stopped the thread after you posted, we would have had the correct number, and that wasn't true again ever.

Hmm. I forgot "median" existed. I'll calculate it, too.

Next level fun facts right here. Love it.

 

[Yuri_G]

Tesla released Q3 2016 Production and Deliveries today at or before 11:09AM: Tesla Q3 2016 Production and Deliveries (NASDAQ:TSLA)

Tesla said they "delivered approximately 24,500 vehicles in Q3", so subject to revision, and the winners here could change.

The average estimates in this thread for Deliveries came to 25,026, which was 526 too high (2.1%).

1st place winner is @Yuri_G, estimated 24,555, off by 55 joined TMC 2012-11-08.
2nd place winner is @MikeC, estimated 24,580, off by 80, joined TMC 2012-07-19.
3rd place winner is @gene, estimated 24,411, off by -89, joined TMC 2013-02-11.

0th Place Winner was @winfield100 who joined TMC 2013-02-16 who guessed a range of 23,000-26,000 for which the very middle was exactly correct at 24,500, who did not post in this thread, but posted at https://teslamotorsclub.com/tmc/posts/1759232/ and is in my mind disqualified for not playing (by not posting here) AND for posting a huge range.

For Production, which Tesla said was 25,185, the average of 8 estimates here was 25,133, which was off by only -52. @Ameliorate (joined TMC 2016-05-22) won at 25,200 (off by only 15). He was also the only one to post a guess for In Transit, which was not asked, but his estimate was 4,500 and the correct answer was 5,500.

My Google Spreadsheet is where I calculated this: Your best #s for 3Q16 Delivery Numbers


The perfect TMC join date cutoff for join date sorted running averaging point was 2013-04-05 when @vgrinshpun joined, join date running average off by 6, which his own guess was not far off by 400 (only 1.6%), but two more good join cutoff points were 2013-08-24 (@bonaire) at running average off by 142, and 2015-04-01 (@luca) at running average off by 180, which was the earliest 2015 join date, so in a sense, your 2015 cutoff point for an average of 180 was not far off point. From the 180 valley of error, the error raised up steadily to a mountain of error at 532 at the end (join date 2016-09-04).

The prior time the running average error (regardless of direction) was off by that much was for a TMC join date of 2012-02-09 (but not due to their own failure, being only -263 off), still suffering a low prediction off by -3,750 from join date 2011-04-03, but the blame doesn't go just to that low estimate, because if he had not placed an estimate, it would have been wildly off by much more from the earliest two join dates to play, 2009-11-03 and 2007-08-20, who, after all, were only off by 503 and 1,211 respectively, not that bad, considering the range of guesses, but for which the running average was off by the greatest (!). The cumulative average just hadn't gotten sensible, yet (back in the less than 6 guesses range), so I left the (well known) names out. Feels kind of like looking back in time by looking further in spacetime in the Universe.

Does this qualify me as an analyst, now? It'll be interesting to see exactly how many more cars Tesla sold when they release the Q3 results.

 

[AudubonB]

Well done, Yuri! As per the thread's strictures, I'll remember to be lenient with you when moderating.



...for your FIRST reprehensible post...


And now, also as promised, I'll be shutting this thread - as soon as I can figure with this still-new-to-me platform.

Thanks to all - well played, everyone.

 

[3Victoria]

Jeeze, it really works. Next time we should email our estimates to Ulmo so that the estimates are secret and we don't get follower bias.

Next lottery numbers

 

[3Victoria]

Oh, when analyzing it might be useful to throw out 'outliers', say those 2SD from the mean (or median).

 

[Turing]

Seems like we should start including in transit at end of quarter numbers in estimates. Big difference between someone who estimates tesla will get 24.5k deliveries by shrinking in transit down to 2-3k. Vs someone who thinks they get 24.5k while maintaining 5k in transit.

 

[austinEV]

Crazy that the average was so close. Wisdom of crowds indeed.

 

[pz1975]

BUMP.

With the actual Q3 delivery number being 24,821, the real winner of this contest was:

Atlantis - with a guess of 24,832.

Congratulations!

 

[EinSV]

BUMP.

With the actual Q3 delivery number being 24,821, the real winner of this contest was:

Atlantis - with a guess of 24,832.

Congratulations!

Also median estimate was 24,775 -- a difference of only 57 vehicles or 0.2%. Impressive!!

 

[EinSV]

Also median estimate was 24,775 -- a difference of only 57 vehicles or 0.2%. Impressive!!

Oops used @Atlantis's numbers instead of Tesla's.

Correct numbers are 24821-24775=46 or 0.185%.

Basically dead bang nailed it.

And holy smokes was @Atlantis close!

 

[Ulmo]

Thank you for the updated actual numbers. I updated the spreadsheet. From my quick look at it, it looks like the group average (and mean) are even closer to the actual than the estimate. Absolutely fantastic.