DO NOT POST YOUR VIN HERE -- THIS THREAD HAS BEEN DEPRECATED
The new VIN tracking thread is here: NEW BUYERS: Please post your VIN here
Click image for full-sized chart.
OK, you may have noticed something new about the chart. Since the VIN assignment rates (red dots) are clearly not following anything close to a linear path anymore, I decided to break the old trendline and start a new one. These two trendlines use the columns labeled "2-mo. Rate A" (old trendline) and "2-mo. Rate A" (new trendline). The old trendline will remain static and the new trendline will be updated to reflect new VIN assignment rates over time...at least until there's reason to break the line again. If the VIN assignment rate is all over the place (technically speaking) and a trendline stops making sense, I'll remove them and just go back to a scatterplot without trendlines. Questions?
Ordered S85 on 9/12, confirmed 10 minutes later.
VIN today, 23833
Can we do 167 before Sept 30th and get to VIN 24000?
Craig, I want to thank you for organizing the VIN data. The summary makes it easy to see what is going on and to look up specific numbers. I suppose it is time for me to describe how I analyze this data.
In an ideal world, Tesla would generate VINs sequentially, not skipping any numbers. Granted batching in production would shuffle things around a bit, but on balance over long enough stretches of time the pace of VIN creation would track with vehicle production. What one would like to know then is what was the last VIN created each day, and one could difference this series to get the daily number of VINs issued. The data on hand however is not complete (we do not know what was the last VIN on any given day), and there may be a process lag from when a VIN is first generated and when it is communicated to the customer. Moreover, there are voluntary customer reporting lags. So even if Tesla is sequentially issuing VINs with no gaps, our data on hand cannot give us a pristine view into that process.
Even so the customer reported data is not without merit. I think it is quite useful even if it is incomplete and is subject to some distortions about dates. While we may not know the last VIN issued on each date, we may approximate it as follows. Compute the cumulative max VIN for each date. Drop all VINs strictly less than the cumulative max VIN. If there is not VIN equal to the cumulative max VIN on a given date, one can interpolate between retained VINs. Let's call this interpolated series the cumulative VIN series. It is in fact an estimate of the last daily VIN series. The cumulative VIN series can be plotted as line along with all the other observations. All observations will lie one or below this line. The ones strictly below at least serve to validate the absence of substantial gaps. For example, there appears to be a gap from about 19150 to 19500. We cannot rule out that Tesla may have skipped some 350 VINs, but had there been a couple of observation in that range we could falsify that hypothesis. So all observations are helpful, but only the ones that advance the cumulative VIN series are needed for my estimation procedure.
Given the noise in the cumulative VIN series, it is best not to difference it daily or other short intervals. Two week differences or longer seem reasonable to me. Here's an example of a calculation. Suppose I want the weekly after for the four weeks ending 8/27. On 8/27 the max VIN was 21464. Now I need the max VIN for 7/30, but only have VINs on 7/29 and 7/31. Taking the average of max VIN s on those dates, I.e., interpolating, I get 18555. Differencing, I get 2909, but excluding the 350 gap leaves about 2560 over 28 days. The average weekly rate then is 640.
Two-week ending. Weekly rate.
There does appear to be a slow down in the end of the month. I suspect that in August there was a push to issue lots of VIN and perhaps prioritize batches for maximal delivery in Q3. This would explain the spread of VINs that came in well below the cumulative VIN curve. Perhaps not only were cars batched for production, but also communication to customers may have been delayed accordingly so as not to raise expectations of quick turnaround. Whatever the case, the slow down in VINs in September need not indicate that physical production has declined, but merely that production may be catching up with aggressive VIN issuances in August. That is after the easily expedited batches are rolled out, the odds and ends need also be done. If this interpretation is correct, then we may likely see the pace of VINs pick up again. If we had a longer time series to analyze we may discover some intra-quarter seasonality and other patterns that can help us better understand VIN generation, car production, and ultimate deliveries.
All the best to those who have a passion for VIN statistics!
Finalized 15. august, europe
Vin number recived today btw
It is clear that Elon is trying to hide the production rate.
I want to put you at rest and all the other investors. Elon mentioned that it is cheep to ship to China since containers come fully load to the USA and return empty and thus he fills them.
He also mentioned that he intends to open a factory in the EU and two more in the US. All this indicates a very healthy demand. This is what counts most.
At any rate the Q3 is almost there and we will hear what Elon wants us to hear.
VIN 23753 23sep
Finalized 13aug, Norway
VIN 23844 got it today 24/9-2013
finalized 15/9-2013, Norway
I don't think Elon is trying to hide anything... he said he wants to be at 800 at the end of the year. Earlier in the year there was a report that for August TM raised the order of frunk liners to 650 - that is exactly were he is now. Stabilizing the rate at 650 now is not unexpected... you can not push, push, push... we'll see when he goes for 800 later in the year.
24185 today Norway
If you try to extrapolate the weekly VIN assignment rate directly into a weekly build rate, you will cause yourself heartburn. The problem is that European cars need to have VIN's assigned well in advance of production, while the North American cars are scheduled into production almost immediately. I assume the long lead time for EU cars is necessary to manage parts inventories and optimize logistics.
I don't see any obvious gaps in the VIN sequence. Although this data set is missing VIN's between 19130 and 19500, there are about 12 reported in this range in the TMC EU Deliveries spreadsheet. The only VIN gaps that I've seen for sure were after the US Signatures and after the Canadian Signatures.
So when you see weekly VIN assignment rates pushing 700, you have to know that it's just a block of EU cars being pre-assigned. The actual factory build rate has got to be closer to 600. And as my Aussie friends would say, "No worries, mate. She'll be right."
Can you please attach a link to the EU Deliveries spreadsheet/thread?
Here's the link to the EU spreadsheet. Hopefully it will work.https://docs.google.com/spreadsheet/ccc?key=0AkhbjTlSwONrdEdXaWJNXzZqV1Z...
Out of curiosity, looking at the EUDelivery sheet and also the map I found on TMC of reservation holders there are a lot of people with P#### or S### where I get that the S### were the signatures, but currently I don't see that I got a sequence number. I have the internal RN number and that's it. I guess I'll get a VIN assigned at some point, but did the sequence numbers only happen for serious pre-orders before the car started shipping and configuration was made available or what's with that?
RN# is just an assigned reservation # when you put down your deposit. It has no bearing on when your car will be produced.
RN# will be replaced by the VIN# when the design is confirmed and
finalized. If you're in the EU, then there'll be quite a few weeks of silence until you're invited to confirm your design.
VIN# assignment will be your first glimmer of hope that you're getting much closer to factory receiving your order.
Tesla quit assigning reservation numbers for EU customers back in early June. After that date, you are given an order number. They did the same thing in North America near the end of February.
New target: VIN 25,000
Will it be posted during next week?
@Benz... should be a breeze at this rate
cfOH, can you start a thread with vin#, delivery date, and country (state if US)? I'm wondering if Tableau software can produce some interesting insight... or you tried already or it doesn't interest you? this thread is one of the best by the way. Thank you.
24176 - 9/24
Holy moley! What an asplosion of VINs! ;-D
Apparently, Tesla has a single guy sitting at a computer issuing VINs and he was working only part-time last week; today he caught up.
And yes, I'm kidding about that.
I'm guessing Tesla has end-of-quarter-itis and is pumping out all the VINs it can for whatever stat it helps come reporting time.
@JMH: If I understand the method you're using, I think it underutilizes the data we have and relies too much on data (observations) that happen to be at the end-points of the period you're using to get a rate. What I like about regression methods is that they use ALL the data and don't give undo weight to any one period (either long ago or more recently). I think that's why the 2-month rate estimate is pretty stable...it uses a LOT of data to get the estimate. Of course, the downside is that it'll be slow to show a dramatic shift (either up or down), but then I don't think a chart will tell us something important has happened sooner than news reports will.
Thank you Craig! Your VIN tracking is one of my favorite TM forum threads.
I wonder if Tesla has hit a production rate plateau at around 660/week due to a supplier or factory constraint?
Yes I understand your point. Regression should give you an unbiased and fairly robust estimates. In fact both methods yield about the same rate when applied to the full two month range. So there is not much difference either way, except my approach can detect changes in rate. My approach was also motivated to try to make sense of a reasonable error structure, and differencing is one standard way to arrive at less correlated errors in case one is interest in time series modeling. Sure you can throw all the data into a regression model but with all the dependency of observations, you're not really gaining much information in the statistical sense and your standard errors of estimation will be severely underestimated. Even so, as I said your estimates of rates are probably not terribly biased which should suffice for the purpose at hand. I do wonder if you have tried adding a quadratic term or using some sort of smoother like loess or splines. Doing so would give you a way to estimate changes in rate by differentiating the quadratic fit or differencing the smoother fit.
On a data analytic note, it does appear that Tesla generate VINs in batches and works them through in production. This is pretty apparent in the last two days. I've done some closer analysis of day to day changes in cumulative VIN and have seen this quite a bit. There are days where the cumulative VIN surges ahead followed by a longer stretch with very modest gains if any at all, a pattern consistent with generating a block and spreading it out randomly over several days. It may be possible to try to reconstruct these blocks. That is if the gain in max VINs is slight, it may be reasonable to attribute VINs on that day to an earlier block when it was first generated, but before it was communicated to the customer. The point of such an exercise would be to gain some insight into how Tesla is operation which could aid in a better appreciation of VIN data. So with this most recent block of VINs what we should look for is how they work through that batch and when will they need another block.
Finally, it would be helpful to do a gap analysis of VINs. Basically sort all the VINs in numerical order and compute differences. If we truly have a random sample of all VINs with no gaps, then the differences will have a geometric distribution. Large departures, then, provide evidence contrary to our null hypothesis and suggest that VINs have in fact been skipped. Gaps above a certain critical threshold can be identified and potentially subtracted out of our estimates of rates. Also it is possible to bencmark VIN rates against reported deliveries to show the gaps are negligible over the span of a quarter. Gaps are perhaps the biggest threat to the credibility of VIN data. So doing a proper analysis of this issue and making informed adjustments may be the best way to shore up credibility. (Is there a more complete dataset available online?)
Well the analysis JHM and cfOH are discussing is WAY above my pay grade. But I encourage you to continue to post VINs and the VIN assignment dates. Thanks
Not sure we can make it to 25,000 by the end of September...may be close.
JHM: I could do a more sophisticated analysis, but, frankly, the data aren't good enough to warrant it. If I had 3 or 4 times more VINs, I might start worrying about the distribution of residuals and other statistical minutiae, but with these rather meager data "crumbs" we're getting, anything fancier than OLS regression is probably just putting lipstick on a pig (to be technical ;-). And no, I don't think more complex or higher-order forms would be particularly useful, if even appropriate, because the phenomenon we're mapping out is rationally controlled by people, not exhibiting some natural process where we're trying to maximize the predictive caapabilities. Heck, if I wanted to do that, I'd start looking at something really fancy like spline methods. And wouldn't THAT be a waste of time. ;-D
@craig (time to Channel Brian H, in a good way) - "data aren't good " - wow! Someone who uses 'data' as a plural noun! A rare sight to behold. Kudos.
Tesla assigns RN #'s (reservation number). I got one in July before the VIN. So they still do it.
One physical constraint: I believe the frames must be stamped with the VIN before going thru the Paint Center. So fundamental "issuance" is not done in arbitrary blocks.
What's your explanation for the collapse of the numbers for a while recently?
Could it be China?
Time for more VINs and a bump.
@nickjhowe: How else would you say it?? ;-)
@Brian H: I posted earlier in this thread (see third comment on page 1) my thoughts on why I think the rates had leveled off. But, honestly, they're just guesses.
Also, I'm pretty sure VINs are mostly issued sequentially -- first-ordered-first-issued -- but production batching then starts to generate out-of-sequence deliveries. Plus the whole EU regulation/shipping/inspection stuff further delays individual cars from getting to their owners. That's one of the reasons I don't even try to track actual deliveries...way too messy.
mmmm. Splines are pretty.
Thanks for all the VINS. Please contribute your recent VIN assignment.
Thanks again cfOH for the time and expertise to put this info together.
Still need new VINs to keep the chart/info relevant/useful
Reservation September 4th
Confirmed September 5th
Changed order - included "Winter" option on September 10th
VIN 24256 on September 24th
Well, they just surged again. Faskinatin.
I would like to add our VN to the list if I could just read it. It is so small on my monitor - I can't recognize anything (with the exception of a table and a graph. Is there a trick?