Short-Term TSLA Price Movements - 2016

[mmd]

Think of it this way. Say you buy 10 call contracts at $190 for the next week, 1 contract=100 shares, so you buy right to 1000 shares. MM will want to stay neutral! Since they sold you calls, they're effectively short, and they immediately buy shares. How many? Exactly enough to offset option pricing movement. This site says delta is 0.524 for strike above, so they BUY 524 shares: CBOE - IVolatility Services

MM is now worm and cozy - if your options gain value, so do their shares. You sell them back your options? They dump shares immediately. At no point do they lose (or gain) anything as price changes.

How do they make money? By decaying your time-value of option and on the buy/sell spread.

Example, today's closing price was over $5 for option strike $190(next week expiry), and SP closing price was $190.43.
So you pay $5 for $.43 of the real value and $4.57 of time-value. MM is neutral, they just need to wait for time value to expire and they're good $4.57. Remember, shares they hold protect them from price movement.

Interesting. But your analysis only considers the case when SP goes up. What happens, if say the price drops to $100 tomorrow due to some bad news ? Up to $185, the MM is compensated by the decayed value of the call options the MM sold. How is the MM protected on the downside below $185? So then, they also need to buy well out-of-money put options, is it correct?

 

[erha]

I'm technical, and it's not obvious to me that "machine trained by a human is what Musk is looking for" as opposed to just "machine learning". Supervised learning is still machine learning (unless you meant rule-based systems, which are a poor approach to this problem because of their brittleness).

Why do you think rule-based systems are a poor approach? I think that is exactly what Tesla is aiming for.

Machine learning can obtain surprisingly good performance really fast, but the behavior in corner cases is hard to predict. And that is not a risk we can accept for autonomous cars.

 

[sub]

This is second time this week you annoyed me by attacking member that brings useful information to this forum.
I think you blocking Maoing is awesome solution, you don't have to read his posts, and rest of us don't have to read your attacks.

Interesting. But your analysis only considers the case when SP goes up. What happens, if say the price drops to $100 tomorrow due to some bad news ? Up to $185, the MM is compensated by the decayed value of the call options the MM sold. How is the MM protected on the downside below $185? So then, they also need to buy well out-of-money put options, is it correct?

I agree with this, if MM is functioning this way they would be adjusting their position continuously. Multiply that by how many positions in how many equities and they better have some very advanced software or be a magician. Also, if this were true there would be no such thing as open interest etc, if option traders were always buying/selling from a MM and the MM was always taking the other side of the trade, there would be no need for open interest and the spread size would not be related to the open interest. I call BS on this entire theory until proven otherwise.

 

[The Blue Owl]

Why do you think rule-based systems are a poor approach? I think that is exactly what Tesla is aiming for.

Machine learning can obtain surprisingly good performance really fast, but the behavior in corner cases is hard to predict. And that is not a risk we can accept for autonomous cars.

Rule-based systems are ineffective for this problem primarily because the input space is extremely large. Such systems (also known as fuzzy control systems) are appropriate for situations where the space can be divided into a manageable number of regions, which in turn allows for a manageable number of rules to be written to manage the transitions between a manageable number of internal states. Thermostats, anti-lock brakes, and many industrial control systems can be modelled that way because the possible inputs can be enumerated, and the state-space can be appropriately partitioned. Cars operate in an open environment where the combination of inputs and states cannot be practically enumerated and partitioned. You can implement a fuzzy control system for operating an airplane in cruising mode at altitude, but not a system to operate a fighter jet in combat (you'll need neural networks for that, and for now we use the human variety).

The only way to handle this that I am aware of is to treat this as a big data problem, not an algorithmic problem. That's where the modern machine learning methods shine. You feed huge amounts of streaming input data to the neural network in training, and you let it tune itself to the task (either unsupervised, with reinforcement learning, or supervised, with a human trainer, or a combination). The corner cases are by definition those that occur rarely, and that's why the system may not react correctly in some situations -- it hasn't seen them before (even then, the failure modes of a neural network will typically be more forgiving than those of a rule-based system). But the problem is not one of algorithm design, it's one of acquiring all the relevant data. Once you've identified a new corner case, all you have to do is to feed new streaming inputs from actual driving exemplifying the problem to the network and let it learn. (This is how human drivers learn, too, and why insurance rates tend to get lower for more experienced drivers.) Rule-based systems don't help here: if you don't know that a particular situation can occur, you can't write rules for it.

That's why Google has driven its autonomous cars millions of km trying to put them in all conceivable situations, and they've only barely scratched the surface. That's also why Tesla's continuous collection of ever-increasing amounts of real-world driving data is a big deal.

There are other problems of similar nature where rule systems have proved to be inadequate. Language translation is one striking example. All early attempts based on expert systems failed miserably. The advent of deep-learning algorithms, coupled with the exponential increase in computing power and the availability of huge data sets, gave us human-level performance. Another example is the game of Go. It was inconceivable just a few years ago that a program would play at professional level, but Google pulled it off.

I don't know what Elon meant when he made his remark, and I'd rather not speculate. But if I were to bet, I wouldn't bet on a rule-based approach.

 

[Julian Cox]

I'm technical, and it's not obvious to me that "machine trained by a human is what Musk is looking for" as opposed to just "machine learning". Supervised learning is still machine learning (unless you meant rule-based systems, which are a poor approach to this problem because of their brittleness).

I'm also pretty sure when he says 80-90% accurate he is not thinking of elephants, but rather, say, the Interstate covered in sludge, faded or missing road markings, road signs painted by vandals to show 85mph instead of 35 mph limits, etc. There are plenty of tricky situations in day-to-day driving on North-American roadways which would make it very difficult to achieve three nines reliability.

Not really the place to have a debate about this. Just wanted to point out that a statement of Machine Learning being 90% of the puzzle does not translate to Autopilot being 90% accurate. One is not the other. Machine learning is one of a number of techniques to get to 99.999% accuracy and these techniques can be additive, not alternatives.

Right now Autopilot 7.1 Beta is much better than 99.9% accurate (could already be three 9s) in highway situations otherwise it pops up a warning alert and hands control back to the driver. Musk has described difficult situations as ones in which the car is going too fast to simply stop within the radius of its ultrasonic sensors if something intrudes and too slow to be set in a relatively disciplined highway situation with very few variables to handle. Basically 20mph - 50mph in residential or commercial environments. Here I can sort of see that Machine Learning per-se may not necessarily do as well at perceiving the impending danger of exceptional never before encountered events as well as a human might be able to. Generally humans get a fear response to 'something odd happening'. Machines tend to work on pattern recognition 'I recognize that and it's on my 90% comprehensive list of bad things to avoid' - which isn't good enough.

That said I think Tesla is on the right track in a way that Google and all the rest is not. Tesla seems to be setting out to square the disparity described above by documenting in data what a normal journey on a familiar road should look like to the vehicle's sensors. This is much more human-like. With sufficiently comprehensive mapping data the sensors ought to be able to do what a human does: See something unfamiliar and deal with it cautiously. This is a reverse of the imperative to catalog every possible anomaly that the car could ever encounter - and I think a much more attainable goal.

It is also worth pointing out that Tesla's current network learning fleet is not so much a machine learning fleet as a human teaching fleet.

I also think that given the above operating scheme, Tesla has a brute force option at its disposal for dealing with fully autonomous cars that are too 'frightened' to move: Route the camera to a live OTA network operator to take a look at the unfamiliar situation and make a judgement call. Like, yes, the bridge in front of you has in fact collapsed. Turn around.

 

[The Blue Owl]

Not really the place to have a debate about this. Just wanted to point out that a statement of Machine Learning being 90% of the puzzle does not translate to Autopilot being 90% accurate. One is not the other. Machine learning is one of a number of techniques to get to 99.999% accuracy and these techniques can be additive, not alternatives.

That's different from the quote from the article. In that quote, Musk explicitly refers to accuracy:

Musk has become skeptical that machine learning can solve the harder problems involved with autonomous driving, saying that to be viable the system would need to achieve 99.999% accuracy, while he currently sees machine learning reaching 80% or 90% accuracy.

Also:

Machines tend to work on pattern recognition 'I recognize that and it's on my 90% comprehensive list of bad things to avoid' - which isn't good enough.

That's not how deep neural networks operate. That's how rule-based systems operate.

I take your point about additivity, in principle. I was referring to rule-based systems in particular, but you didn't bring those up in your original post. It was my own speculation (although, based on your remark above, it may have been what you had in mind).

 

[The Blue Owl]

That said I think Tesla is on the right track in a way that Google and all the rest is not. Tesla seems to be setting out to square the disparity described above by documenting in data what a normal journey on a familiar road should look like to the vehicle's sensors. This is much more human-like. With sufficiently comprehensive mapping data the sensors ought to be able to do what a human does: See something unfamiliar and deal with it cautiously. This is a reverse of the imperative to catalog every possible anomaly that the car could ever encounter - and I think a much more attainable goal.

It is also worth pointing out that Tesla's current network learning fleet is not so much a machine learning fleet as a human teaching fleet.

I don't know how I missed this the first time. Yes, this is what I am talking about, and what I also described in my response to erha. This is what machine learning does. It doesn't mean humans are not involved in teaching it; that's just supervised machine learning. We agree on the substance (just not on the terminology).

 

[MitchJi]

The article tries to be fair and balanced to provide some reasons why uptake could be slow and pointed to charging infrastructure. This is a non issue because infrastructure will easily scale with the number of EVs on the road. But they missed the biggest challenge which is scaling up battery production capacity.

So to get to their 2 mb/d oil displacement, cumulative EVs need to reach 50 M. Growing at 50% means annual EV production needs to hit 25 M. At 60 kWh per EV, this requires 1500 GWh/year capacity, or 30 Gigafactory equivalents. Reallistically, about 50 gigafactory scale factories need to be on line at some stage of ramp up by the critical year for oil impact. From this battery supply perspective, 2023 seems unrealistic. Since development time is about 5 years for such a plant, 8 years to launch 47 additional gigafactories is an extremely tight timeframe.

Seems pretty obvious to me.

In your question you had [indiscernible] should be corrected, like the -- so the 30% savings is not just due to logistics. Logistics is a big factor. We are --

It's not even the biggest though.

Logistics [indiscernible] the fact that it's just go to one station to the next instead of going from multiple entities to multiple entities. But really when you get to the kinds of scale that we're talking about, you really get to design customer equipment that's much better at processing each step. And you really get to design the machine that makes the machine, not just do so with off-the-shelf equipment. So it took -- everything about it is going to get a whole lot better. That's why we think the 30% number when the Giga Factory is at full production is a conservative number. Yeah. And then, yeah. So.

Two things slowing the building of GF1:
It's the first one.

Tesla has limited capital, and has not yet demonstrated it's a cash cow.

 

[Bgarret]

Sorry about the cross post, but I thought this was a great contrast of the bull and bear cases (unwittingly) by Bloomberg. On the bear side is Corey Johnson, noted Tesla baiter. On the other side is a spokesman - also from Bloomberg - who put out the analysis earlier this week about electric cars causing an oil glut in 2023. Watch and laugh. Corey's "electronic" cars, Lithium demagaguing and snarky "Tesla doesn't meet forecasts" vs. .....Science and fact. Love the look on his "colleague's" face every time Corey butchers the interview.

enjoy.

Rising Electric Vehicle Sales Concern Oil Industry | Watch the video - Yahoo Finance

 

[BornToFly]

Sorry about the cross post, but I thought this was a great contrast of the bull and bear cases (unwittingly) by Bloomberg. On the bear side is Corey Johnson, noted Tesla baiter. On the other side is a spokesman - also from Bloomberg - who put out the analysis earlier this week about electric cars causing an oil glut in 2023. Watch and laugh. Corey's "electronic" cars, Lithium demagaguing and snarky "Tesla doesn't meet forecasts" vs. .....Science and fact. Love the look on his "colleague's" face every time Corey butchers the interview.

enjoy.

Rising Electric Vehicle Sales Concern Oil Industry | Watch the video - Yahoo Finance

Man is Corey an idiot. He still spews nonsense such as "Tesla looses money on every car, and that is not a sustainable business model." Anyone with a half a brain knows Tesla MAKES 25% profit margins on its vehicles, and that business model is extremely profitable. They are spending a lot on the Gigafactory, store expansion, super charger expansion, and new model development (which is why the company as a whole is not profitable YET), which will eventually bring in serious revenue as they take over more and more of the car market. You have to invest capital to start and grow a new business. It is pretty hard to be profitable on day 1, especially when your product requires an expensive set of factories to build.

 

[Oil4AsphaultOnly]

Now that the Model X is confirmed for the Geneva Motor Show (Opening on Tuesday), can we expect some positive catalyst from that?

Meanwhile, on the macroeconomics front, back at the G20 summit:“This is a moment where you’ve got real economies doing better than markets think in some cases.”

So should we expect more of the same from the Dow/Nasdaq in the first week of Mar as this past week? I really don't expect Tesla to announce anything (Musk's tweet about white seat support notwithstanding ) until the end of March, so all short-term price action would be driven by the rest of the market and continued de-coupling to crude oil prices.

 

[BornToFly]

Now that the Model X is confirmed for the Geneva Motor Show (Opening on Tuesday), can we expect some positive catalyst from that?

Meanwhile, on the macroeconomics front, back at the G20 summit:“This is a moment where you’ve got real economies doing better than markets think in some cases.”

So should we expect more of the same from the Dow/Nasdaq in the first week of Mar as this past week? I really don't expect Tesla to announce anything (Musk's tweet about white seat support notwithstanding ) until the end of March, so all short-term price action would be driven by the rest of the market and continued de-coupling to crude oil prices.

My absolutely meaningless thoughts - I expected a rise on the news before Model 3, just not 5 weeks out. I was thinking more along the lines of 2 weeks out. I hope this isn't similar to a runner/cyclist that sprints for the line too soon. What I'm hoping is that everyone is expecting a rise before March 31st, and now that it is happening, nobody wants to be left out, so longs and short term traders hoping to capitalize on the run up are reluctantly getting on board sooner than they planned, but are getting on anyway. Shorts are getting squeezed a little because fewer are willing to sell right now. I am 50%/50% on whether or not there will be a drop after the reveal. If it looks great, there won't be. But if the Model 3 is a little funny looking in order to get the drag under 0.2, and there is some criticism after the reveal, then I see a short term dip. Hopefully there will be another climb after that brought on by the large number of pre-orders that come in, even if it is a little funny looking....

 

[Zhelko Dimic]

Interesting. But your analysis only considers the case when SP goes up. What happens, if say the price drops to $100 tomorrow due to some bad news ? Up to $185, the MM is compensated by the decayed value of the call options the MM sold. How is the MM protected on the downside below $185? So then, they also need to buy well out-of-money put options, is it correct?

What I described is a simplified principle of options issuer hedging, not actual execution. I don't know details of actual execution, and they may very well be trade secrets and implemented differently by different trading organizations. Here are couple more things that I know.

Trading companies have some serious programming talent. They pay really, really well, and expect a lot. My data is old, but when I was looking into it 10 years ago, main focus has been on crunching insane amounts of data super fast. I would not be surprised that they have real-time or near real-time insight into their cumulative position.
But even if they don't, there are ways to manage exposure issues through organizational process. For example, you can request that all individual traders maintain close to zero delta position and never stray more than xxx, whatever xxx is.

Above doesn't addresses your question, but I would not be surprised that there are diff. levels of hedging and there is compliance and risk mgmt. group that is handling issues like one you described. Some portion of profit could be spent on advanced hedging that guards against when markets move very far overnight. Banks are in the business of managing risk, that's their core excellence, supposedly.

To conclude, I've described how things were done during the the dawn of options trading, many many years ago. Nowadays, trading companies have mathematicians, rocket scientists, brilliant developers, to execute better methods, and I'm certain they've solved any issue we can think of to significant degree.

Would it hedge them against whole market waking up 10% red? That I don't know, but I'm certain they can handle couple percents, and even if they lose on particular company, that's likely price of doing business.
I just thought of a metaphor that kinda works and I like it . Think of MMs as car insurance company. Now TSLA, NFLX, etc will be high risk drivers (high IV), so insurance premiums are high. Occasionally, such driver ends up in a ditch (TSLA moves 5% overnight), but on the balance, MMs always charge as much premium as they need to comfortably cover all of their risks and live large

- - - Updated - - -

I agree with this, if MM is functioning this way they would be adjusting their position continuously. Multiply that by how many positions in how many equities and they better have some very advanced software or be a magician. Also, if this were true there would be no such thing as open interest etc, if option traders were always buying/selling from a MM and the MM was always taking the other side of the trade, there would be no need for open interest and the spread size would not be related to the open interest. I call BS on this entire theory until proven otherwise.

OK.

 

[RobStark]

Two things slowing the building of GF1:
It's the first one.

Tesla has limited capital, and has not yet demonstrated it's a cash cow.

All companies have limited capital even cash cow Apple.

The difference is the number.

Tesla has proven again and again its ability to access the capital markets,when needed, on favorable terms.

 

[Papafox]

What I described is a simplified principle of options issuer hedging, not actual execution. ...

Zhelko, I agree with you that the market makers have no ill-intent when they hedge their options sales. They're looking to hedge their position, I understand that. What I suspect, and I believe it becomes apparent on days when there's low volume and the close of monthly options, is that the market makers are working to push the stock price closer to max-pain. Granted, if that push places the holdings of the MMs out of balance, then they're not going to keep pushing, but by using creative timing of buys, sales, etc., perhaps the MMs can be simultaneously moving their holdings into an equilibrium while exerting force to approach max-pain, in an optimized manner. As you said, these firms employ some very bright and creative individuals.

 

[Bgarret]

Now that the Model X is confirmed for the Geneva Motor Show (Opening on Tuesday), can we expect some positive catalyst from that?

Meanwhile, on the macroeconomics front, back at the G20 summit:“This is a moment where you’ve got real economies doing better than markets think in some cases.”

So should we expect more of the same from the Dow/Nasdaq in the first week of Mar as this past week? I really don't expect Tesla to announce anything (Musk's tweet about white seat support notwithstanding ) until the end of March, so all short-term price action would be driven by the rest of the market and continued de-coupling to crude oil prices.

Not sure on this....there could be some interesting action around the invitations to March 31st...and remember, everyone seems to be talking about the Model III prototype unveiling, when the possible bigger story is a "Fully Functional Death Star!...er....Gigafactory".

Notice how the dialogue is all about the little car and not as much about the 20% completion of what, when completed, will be the largest industrial building in the world that will double the global production of Lithium-Ion batteries. Hmmmmm.

There are still more Rabbits and tweets to come out of Elon's hat. Stay tuned.

 

[Zhelko Dimic]

Zhelko, I agree with you that the market makers have no ill-intent when they hedge their options sales. They're looking to hedge their position, I understand that. What I suspect, and I believe it becomes apparent on days when there's low volume and the close of monthly options, is that the market makers are working to push the stock price closer to max-pain. Granted, if that push places the holdings of the MMs out of balance, then they're not going to keep pushing, but by using creative timing of buys, sales, etc., perhaps the MMs can be simultaneously moving their holdings into an equilibrium while exerting force to approach max-pain, in an optimized manner. As you said, these firms employ some very bright and creative individuals.

I'm not sure if we landed on exactly the same page, but close enough. Importantly, Max Pain is real. It always exists, but varies in strength, depending on the type of the week, and is no match for news, momentum, etc. While real, it can't be sole trading consideration, just an augmenting info.

I find it ironic that this is a force screwing up open interest, but the sequence is initiated by holders of that open interest, when they try to cash in.

 

[Johan]

So, I suppose this may have been posted elsewhere on the forum but it's a nice little data based clip that fits @jhm's narrative perfectly.

 

[hobbes]

And another one - interesting as it suggests a move to inhousing chip design:

Tesla hires yet another chip architecture titan out of Apples PA Semi, feeding the rumor that it plans to design its own silicon | Electrek

Not sure if people here aware of this but:

Elon Musk hires Apples alloy expert to lead materials engineering at both Tesla and SpaceX | Electrek

Basically, Tesla will be the only car company with likely the best alloy technology for it's cars. Space-age I would say! Inconel was just the beginning. Not sure how this may have influenced this weeks stock action.

 

[MitchJi]

This is insanity:
Oil industry sees Paris climate deal as chance to innovate - Business Insider

HOUSTON (Reuters) - If a crisis is a terrible thing to waste, the oil industry sees the Paris climate accord not as a death knell, but an opportunity to innovate and even grow.

The move is a shrewd one for an industry that has been on the defensive for years on climate issues, constantly fending off attacks that its products have contributed to an unhealthy rise in global temperatures.

While the landmark emissions-reduction agreement among 195 countries late last year was seen as a defeat for fossil fuel producers, executives and oil ministers sounded a clarion call this week at their first major meeting since the Paris talks for more research into how carbon capture technology can be cheapened and perfected.

The hope is that this and other technologies could sharply cut oil and natural gas emissions, protect the industry from the ramifications of climate change legislation and ensure developing economies still have access to inexpensive energy.

"If you could eliminate all of the carbon dioxide from fossil fuel combustion, then you could use those fuels as long as you want," Robert Armstrong, director of the MIT Energy Initiative, said in an interview on the sidelines of IHS CERAWeek, the world's largest gathering of oil executives

If you believe in the tooth fairy and you are willing to tolerate the health effects.