What are you buying today?

[BG2008]

The best part about IB is if you call in to their "trade desk", you'll probably get the order in right as the internet login gets fixed.....in about an hour!

 

IB's help line is absolutely atrocious.  If you have a client with 8 figure accounts and you put them on hold for 20 minutes.  They are bound to leave in frustration over time.  But I have also just invested a ton in learning how to use their algos. 

[bathtime]

DBX

[RichardGibbons]

Traditional insurances had AI/ML for decades. They called it statistics.

 

No, they didn't, and I'm not quite sure why you'd pretend these are the same thing.

 

On a practical level, ML uses historic data (usually entire dataset, let's ignore train-test-split for a second) to make generalizable predictions. Statistics draw inferences from a sample. To make statistics ML, just up your sample size to make it population size  ;). I can walk you through the same path to show you how AI is basically the same thing (starting with perception). I'd love for you to explain what is it that makes you think I'm pretending?

 

Generally, AI/ML fields borrow heavily from statistics. Sure they are different if you take the purist approach (e.g., ML predicts based on passive observations and AI implies agent interaction with the environment to maximize chances of goal achievement).  Sure, other fields are contributing (EE, CS, etc.) and some of the latest algorithms don't come from the field of statistics but at the core, these are all statistical methods (see the assumption with any algorithm that is available today). The reality of things is that what changed the field are three things: 1) availability for computing power (AWS, GCP, etc.) and 2) data, lots of it 3) fusion of different methods (e.g., TensorFlow).

 

Feel free to let me know what insurance-specific algorithm(s) Lemonade has that is not rooted in statistics that are not available to Progressive, Geico, etc. I say insurance-specific because I'm sure Lemonade, by virtue of being new (i.e., no cultural or digital transformations necessary), can rapidly deploy a bag of algorithms to help with processing (robotics process automation), translation (nlp/nlg), etc. So is the premise that they are an efficient back office?

 

I have a math degree, two computer science degrees, and like, 10 AI courses under my belt, so I'm not really someone who you can throw technobabble at to try to obfuscate the issue.

 

Your argument is basically, "hey, these two techniques can both can be used to analyze data, therefore they are the same." 

• Hey, a human and a lump of coal floating in a bucket of water are made of roughly the same stuff, so to anyone but a purist, they're the same.
  
• A pie chart and the algorithm for a self-driving car are both just derived from data, so those are the same thing. So, anyone who can make a pie chart should be confident they can create a self-driving car!
  
• Basic addition and partial differential equations are just about number manipulation--these are all just mathematical methods--so anyone who can add 2+3 ought to be able solve PDEs.
  

And I'm not sure why you'd expect me to know about Lemonade's proprietary algorithms, or why you think me being unable to share such algorithms so would provide any evidence of anything. Like, it's obvious you know that you said something silly. Why would you double down on the silliness?  It's OK to say, "yeah, they're not really the same thing. I really just meant that they're both ways of manipulating data."

[LC]

I lean towards Infoisone's observation that in many cases, AI & ML methods are essentially statistical methodologies re-branded and applied to a a large dataset (and for sure, it is applied in some novel ways). Look at the prototypical KNN algorithm: it is essentially a combination of OLS and the classification problem. The advancement is the computing power (and thereby application to "large data"), but not the methodology. Most AI/ML is simply a marketing term to MBA-educated senior management.

[Cardboard]

This is the "What are you buying today?" thread.

 

Go argue elsewhere Gibbons.

 

Cardboard

[clutch]

I agree with the overuse of "ML/AI" as a marketing ploy. I work in the software industry and I see many companies (including ours) put "ML/AI" on marketing and PR materials but under the hood, it's just traditional computational techniques.

[fareastwarriors]

Added to more $GRIF at 75. Small. Just few thousand $s.

[patience_and_focus]

I lean towards Infoisone's observation that in many cases, AI & ML methods are essentially statistical methodologies re-branded and applied to a a large dataset (and for sure, it is applied in some novel ways). Look at the prototypical KNN algorithm: it is essentially a combination of OLS and the classification problem. The advancement is the computing power (and thereby application to "large data"), but not the methodology. Most AI/ML is simply a marketing term to MBA-educated senior management.

 

Although I would agree with Infoisone and LC on parts of it, for example many times what is branded as AI/ML is actually just "old" statistical techniques like KNN, I will disagree on other parts. Sure, machine learning started with its roots in statistics and borrows some concepts from it. However, on the theory itself it has long evolved to be distinct from how theoretical statistics has advanced (see statistical learning theory and PAC theory here http://www.econ.upf.edu/~lugosi/mlss_slt.pdf and deep neural network theory - https://www.pnas.org/content/117/48/30039). And many modern techniques are inspired from this new theoretical framework (SVM, Deep learning). Both on theoretical side as well as practical technique development side increasingly the practitioners are also diverging. They think of problems differently, attend different conferences, etc, etc. The goals/objectives of techniques coming out in these fields are also different.

 

The AI/ML field reminds me of early days of statistics where a lot of practitioners came from varied fields (Fisher - was he a geneticist using statistics or other way round?) and helped grow it to become a mature and distinct field of its own. Statistics is an important component of AI/ML, but so is optimization, computational complexity, etc (http://pages.cs.wisc.edu/~andrzeje/lmml.html).

[RichardGibbons]

This is the "What are you buying today?" thread.

 

Go argue elsewhere Gibbons.

 

Cool! The most hypocritical comment of the day!

[lnofeisone]

I have a math degree, two computer science degrees, and like, 10 AI courses under my belt, so I'm not really someone who you can throw technobabble at to try to obfuscate the issue.

Congrats on your accomplishments! Between the two of us, we have a math degree, 3 computer science degrees, 1 electrical engineering degree, and a physics degree. I teach AI/ML at a university (historically, non-online  8)) and worked for an insurance company. So now that we pointlessly settled that (and really credentialed our mutual "technobabble") do we really need to go through the false equivalence that is the next 3 lines you wrote? On a more cordial note, I found this particularly hilarious "Hey, a human and a lump of coal floating in a bucket of water are made of roughly the same stuff, so to anyone but a purist, they're the same." Got to give some love to kinetics and thermodynamics.  Call me in 100 years. Pretty sure we will all be lumps of coal floating in a bucket. Though maybe if I turn myself into a diamond, I'll sit there on a shelf for a bit longer or sink to the bottom of the said bucket  ;D.

 

For my curiosity, forget prop algorithm(s) that Lemonade has, what AI/ML technique in your mind is not rooted in statistics? In fact, let's say for a second that Lemonade (like Capital One in the past) found a way to stratify the broad population into smaller segments. And now they have to make inferences. So, back to statistics.

 

More importantly, and probably more pertinent to this forum:

 

1) Today, Lemonade ratios are declining but are still above the industry average (59.6% for 219, 61.6% for 2018 - I'll agree upfront that these numbers aren't totally accurate as Lemonade doesn't cover everything P&Cs do). So, for now, they are converging to average.

 

2) Let's peel off some of that sweet, sweet, AI/ML magic. Lemonade's largest markets are CA, TX, NY (around 70%). All 3 of those markets clock in net loss ratios that are typically below the industry average with premiums above the industry average. As a fun fact, in California, they have a pretty high justified complaint ratio. Imagine what it takes to get a millennial to complain and take it to the state. By the way, few companies just above and below Lemonade have 2 star ratings and some very scratching remarks, as per Gooogle.

 

3) They are currently ceding 75% of their policies. Curious how their reinsurance fees will hold up as more data comes in.

 

4) Aside from my belief that they are simply converging on the weighted average of the rations of the markets they operate in, I'm genuinely curious what general set of AI/ML algorithms differentiates Lemonade from Progressives of the world? What makes you believe that the latter can't figure these algos out? The latter are sitting on plenty of data, can afford to acquire new datasets, and hire an army of data scientists to get through the data. Cloud is not really a differentiator anymore. I agree, Lemonade is willing to try things that others haven't (e.g., behavior analytics) at the production level but at its core, it's still a test-and-learn shop.

 

I don't have a high conviction in the timing of this short (hence such a small short). I do think it's a nice platform that beautifully obfuscates a traditional insurance company. Probably should take this to the Lemonade thread...

 

[cameronfen]

Added 1/3 short position in LMND after seeing my small short be very much worthless. Not sure what Motley Fool sees in this that is so transcending of AI/ML that Geico or Progressive don't already have or can't buy.

 

I don't know the company well, but, having worked as a tech person in an insurance-related company, insurers often don't seem to have the culture to adopt market-changing technological solutions. I agree that something like this that seems to be the obvious strategy for tradition insurers. But it isn't necessarily something that those traditional insurers can actually execute.

 

Traditional insurances had AI/ML for decades. They called it statistics. I haven't seen anything revolutionary (e.g., Tesla was the only EV for a while) out of Lemonade and they aren't price competitive if I have a car + rent/own. Throw in challenges with renters in big cities, at minimum there will be turbulence in the next few quarters as older policies start to roll off. Just my 2 cents and I was wrong on LMND stock before.

 

I’m not really interested in Lemonade, but fairly interested in AI.  I agree that traditional ML is basically rebranded statistics.  All the methods from KNN to kernel SVM and even maybe XGBoost could be found in Elements of Statistical Learning or some updated similar text.  However, I think AI really evokes deep learning.  I don’t know if lemonade uses NLP, computer vision or Reinforcement Learning or even more exotic things like Graph Neural Networks—I kind of doubt they do very much (even though they could benefit from all the above except maybe RL because control theory  techniques are still often better irl) and I bet much of what they say is marketing.  But there has been a bit of a divergence between AI and statistics in the last 8 years and basically it’s been deep learning.  Statisticians use deep learning sometimes, but not to the extent that it’s used in AI which is almost in entirety and almost all the cutting edge research here is done by CS people and not statisticians.  Just my 2 cents. 

[willie2013]

ROOT as a speculation (calculated gamble) significantly below the IPO price with a chance to grow within and disrupt the car insurance business.  Been tracking it and the chart turned up on decent volume.

[kab60]

Bought some Analogue Holdings (HK), pretty good business at a fat discount to NAV. Will write it up.

[Spekulatius]

AZO

[RichardGibbons]

I have a math degree, two computer science degrees, and like, 10 AI courses under my belt, so I'm not really someone who you can throw technobabble at to try to obfuscate the issue.

Congrats on your accomplishments! Between the two of us, we have a math degree, 3 computer science degrees, 1 electrical engineering degree, and a physics degree. I teach AI/ML at a university (historically, non-online  8)) and worked for an insurance company. So now that we pointlessly settled that (and really credentialed our mutual "technobabble") do we really need to go through the false equivalence that is the next 3 lines you wrote?

 

The only reason I wrote it was because you were trying obfuscate the issue by saying a lot of technobabble. I viewed it as attempt to obfuscate the issue, and intimidate anyone who was scared of buzzwords into shutting up.

 

Also, I don't think those things are false equivalences. It's just that they're more more clearly ridiculous because they aren't cloaked in technobabble.

 

That said, I'm fine with not arguing about whether statistics and AI are basically the same thing. And for all of Cardboard's hypocrisy, he's right. It's way off topic for this board.

 

More importantly, and probably more pertinent to this forum:

 

1) Today, Lemonade ratios are declining but are still above the industry average (59.6% for 219, 61.6% for 2018 - I'll agree upfront that these numbers aren't totally accurate as Lemonade doesn't cover everything P&Cs do). So, for now, they are converging to average.

 

2) Let's peel off some of that sweet, sweet, AI/ML magic. Lemonade's largest markets are CA, TX, NY (around 70%). All 3 of those markets clock in net loss ratios that are typically below the industry average with premiums above the industry average. As a fun fact, in California, they have a pretty high justified complaint ratio. Imagine what it takes to get a millennial to complain and take it to the state. By the way, few companies just above and below Lemonade have 2 star ratings and some very scratching remarks, as per Gooogle.

 

3) They are currently ceding 75% of their policies. Curious how their reinsurance fees will hold up as more data comes in.

 

4) Aside from my belief that they are simply converging on the weighted average of the rations of the markets they operate in, I'm genuinely curious what general set of AI/ML algorithms differentiates Lemonade from Progressives of the world? What makes you believe that the latter can't figure these algos out? The latter are sitting on plenty of data, can afford to acquire new datasets, and hire an army of data scientists to get through the data. Cloud is not really a differentiator anymore. I agree, Lemonade is willing to try things that others haven't (e.g., behavior analytics) at the production level but at its core, it's still a test-and-learn shop.

 

I don't have a high conviction in the timing of this short (hence such a small short). I do think it's a nice platform that beautifully obfuscates a traditional insurance company. Probably should take this to the Lemonade thread...

 

FWIW, I suspect that you're right about Lemonade. I do buy SAAS companies that are breaking traditional business models, and Lemonade is a "don't buy" for me because at a P/S of 53, it's too expensive considering its growth rate and questionable moat.

 

(Keep in mind for all my comments that I've literally spent a total of 5 minutes in my entire life looking at the company.  It's easy to know that I don't want to buy it, and then I don't spend more time.)

 

On 1 and 2, I think there's a reasonable chance you're right. Insurance is a pretty commodified business, and it's unclear to me what they could be doing differently that would provide a big boost over what other people are doing with statistics. Maybe they have found something, but the odds are against it, I think.

 

And if they have found something, how big of an of advantage is that "something" likely to be? I'd bet that whatever it is wouldn't be transformational, like improving margins by ten percentage points for an extended period of time.

 

On the other side of the analysis is that they likely haven't squeezed out most of the economies of scale yet. As they grow, they'll get a boost from that. On the other hand, at this scale, they can also be more picky about their business, so maybe that all balances out.

 

On 4, In terms of why at traditional insurance company might have problems with a tech startup--which is what I was talking about originally--it's not the tech. It's the culture.

 

(Here I'm generalizing my experiences over the whole industry, which might not be fair. But it happens again and again, where old school companies can't adapt to new tech models despite all reason seeming to indicate that they should be able to do so. And that's why I view my experience as more of a case study than an anecdote.)

 

So, I worked in an non-tech insurance-related business that did well enough to become the gorilla. A young tech company came in with about one hundredth of our resources, and started attacking us.

 

We were able to describe the threat in detail early on, but were ignored ("Don't worry about them. They're nothing, and will go bankrupt soon because they're doing unprofitable underwriting. Focus on the ball.").  Then they captured one of our customers ("It's just an aberration. This will prove that their business model doesn't actually work. They'll go bankrupt faster.")

 

We came up with technologies to counter them, technologies that we could've built in a few months with a few decent developers. But of course, those technologies needed to be put through the planning committee, and their value compared against all the other technical projects.  How long does that take?  Well, maybe a year to through the process, just to get get developers on it, and then the developers suck. So years to get something useful, and by then everything's moved on.

 

And, over the course of about 15 years, the tech company has been eating most of the customers.

 

The big problem is a culture built on conservative math and squeezing pennies out of operations fails against a rapid iteration, testing and failure tech model.  Pretty well every company says technology is a competitive advantage, but in the insurance companies I've see, technology is a cost center--it's there because it's necessary to compete. The company isn't run from a perspective of "tech is all we have, let's quickly cycle, failing 60 times in order to find that one solution that works and crushes everyone else."

 

If, as an individual, you have a project that fails in a tech company, it's something to learn (because it's set up so the failures aren't super-costly). If you fail in the insurance company, you've seriously impacted your career, and might never get that promotion. Why take that risk, when analyzing every decision to death, getting written buy-in from everyone, and proceeding at a slow, measured pace is much more likely to lead to successful personal outcomes? Then if you fail, everyone else fails too, and everyone's knows that there's nothing different you could have done.

 

The culture really matters.

 

(All that said, I still wouldn't bet on Lemonade.)

[rosemontseneca]

LMT, BAESY

[CassiusKing1]

Why do I feel like I'm in a scene of Good Will Hunting?

[SafetyinNumbers]

Added more ATTO today

 

Tiny market cap but big company.

 

They reported a strong Q3 and will likely again in Q4. Goldman, one of the three analysts didn’t update estimates post earnings last month so the forward EBITDA consensus is way too low. I think he’s either embarrassed or indifferent (ATTO reported $45m in EBITDA and GS was at $18m). Either way, consensus 2021E EBITDA is only $158m, while the three estimates are $114m, $174m and $185m. For 2022E, the consensus is $171m with Goldman at $138m and the other estimate at $204m.

 

Throwing out Goldman’s stale estimates, ATTO is trading at 3.7x EV/EBITDA.

 

Competitor Concentrix (CNXC) was just listed on the Nasdaq after spinning out of Synnex and it trades at ~9x EBITDA making, ATTO quite accretive for an acquisition. Slide 31/32 in their analyst meeting deck make a pretty good case to buy Atento to solve for growth in emerging markets and for accretive acquisitions.

 

https://ir.concentrix.com/static-files/6c895513-f519-46ce-9566-5c33ca93a8dc

 

I think this deal happens within two years which will be after ATTO management gets margins up to its target of 15% and the stock price is much higher.

 

At 8x EBITDA on the consensus 2022E number of $204m, yields a target of $70 vs the current price of $10.45.

 

Lots of room to be wrong in between those numbers and still be happy. In fact, I think the EBITDA estimates are too low so I see upside beyond that.

 

 

[Gregmal]

Trimmed the GEOS and some more CRSP, paid down some margin and bought a little GS.

 

Any news for GEOS to trade up 30% since that day of forced selling?

 

Only news I saw was the $6 to ~$8 move which in a round about way screamed, "ALWAYS TAKE ADVANTAGE OF FORCED SELLERS!"... I've trimmed position down again to about half; I think the rest I'll layer out of in the 8s. I would not be surprised to see a sale of the company though. The buyback was actually intentionally, or unintentionally, brilliantly timed as well.

 

Sold the rest of this today. Also did the unthinkable. Shorted some Tesla. ~1% position.

[chrispy]

AMT-near a low for the year if we exclude the few days in March

[fareastwarriors]

Trimmed the GEOS and some more CRSP, paid down some margin and bought a little GS.

 

Any news for GEOS to trade up 30% since that day of forced selling?

 

Only news I saw was the $6 to ~$8 move which in a round about way screamed, "ALWAYS TAKE ADVANTAGE OF FORCED SELLERS!"... I've trimmed position down again to about half; I think the rest I'll layer out of in the 8s. I would not be surprised to see a sale of the company though. The buyback was actually intentionally, or unintentionally, brilliantly timed as well.

 

Sold the rest of this today. Also did the unthinkable. Shorted some Tesla. ~1% position.

 

So you're the reason Tesla is finally down today.

[willie2013]

KBAL a Michael Burry coattail

[Value418]

Added more ATTO today

 

Tiny market cap but big company.

 

They reported a strong Q3 and will likely again in Q4. Goldman, one of the three analysts didn’t update estimates post earnings last month so the forward EBITDA consensus is way too low. I think he’s either embarrassed or indifferent (ATTO reported $45m in EBITDA and GS was at $18m). Either way, consensus 2021E EBITDA is only $158m, while the three estimates are $114m, $174m and $185m. For 2022E, the consensus is $171m with Goldman at $138m and the other estimate at $204m.

 

Throwing out Goldman’s stale estimates, ATTO is trading at 3.7x EV/EBITDA.

 

Competitor Concentrix (CNXC) was just listed on the Nasdaq after spinning out of Synnex and it trades at ~9x EBITDA making, ATTO quite accretive for an acquisition. Slide 31/32 in their analyst meeting deck make a pretty good case to buy Atento to solve for growth in emerging markets and for accretive acquisitions.

 

https://ir.concentrix.com/static-files/6c895513-f519-46ce-9566-5c33ca93a8dc

 

I think this deal happens within two years which will be after ATTO management gets margins up to its target of 15% and the stock price is much higher.

 

At 8x EBITDA on the consensus 2022E number of $204m, yields a target of $70 vs the current price of $10.45.

 

Lots of room to be wrong in between those numbers and still be happy. In fact, I think the EBITDA estimates are too low so I see upside beyond that.

 

Starting tor read about these guys.  Seems to be a fair amount of the business in Brazil, which is going to be challenged, right?

[SafetyinNumbers]

Added more ATTO today

 

Tiny market cap but big company.

 

They reported a strong Q3 and will likely again in Q4. Goldman, one of the three analysts didn’t update estimates post earnings last month so the forward EBITDA consensus is way too low. I think he’s either embarrassed or indifferent (ATTO reported $45m in EBITDA and GS was at $18m). Either way, consensus 2021E EBITDA is only $158m, while the three estimates are $114m, $174m and $185m. For 2022E, the consensus is $171m with Goldman at $138m and the other estimate at $204m.

 

Throwing out Goldman’s stale estimates, ATTO is trading at 3.7x EV/EBITDA.

 

Competitor Concentrix (CNXC) was just listed on the Nasdaq after spinning out of Synnex and it trades at ~9x EBITDA making, ATTO quite accretive for an acquisition. Slide 31/32 in their analyst meeting deck make a pretty good case to buy Atento to solve for growth in emerging markets and for accretive acquisitions.

 

https://ir.concentrix.com/static-files/6c895513-f519-46ce-9566-5c33ca93a8dc

 

I think this deal happens within two years which will be after ATTO management gets margins up to its target of 15% and the stock price is much higher.

 

At 8x EBITDA on the consensus 2022E number of $204m, yields a target of $70 vs the current price of $10.45.

 

Lots of room to be wrong in between those numbers and still be happy. In fact, I think the EBITDA estimates are too low so I see upside beyond that.

 

Starting tor read about these guys.  Seems to be a fair amount of the business in Brazil, which is going to be challenged, right?

 

Brazil (~40% of revenue) has their highest EBITDA margins at 16%. Both revenues and costs are in BRL so it has hurt on a linear basis already in Q2/Q3 results. The USDBRL averaged 5.38 both of those quarters and it’s currently at 5.06 (but volatile) so if that holds there should be a positive currency impact next year, all else being equal.

 

Given Q3 margins were above last year I think it’s safe to say they have been up to the challenge.

 

 

[Gregmal]

Starter in ALCO and MAA, shorted some ZM. Added to BYND puts.