Odd Lots - 46: Space Robots Are Helping Hedge Funds Invest

Episode Date: September 16, 2016

The most valuable commodity for investors is information, and hedge funds and asset managers are going to great lengths to get it -- even to outer space. This week on the Odd Lots podcast, Tracy Alloway and Bloomberg View columnist Matt Levine are joined by James Crawford, a former NASA scientist who founded Orbital Insight. Crawford's company uses satellite photos to do things like track retail sales by studying parking lots and track oil supplies by scanning global oil tanks. He explains how his company figures out what to look for and how to look for it, and how investors and governments use his information to make decisions.See omnystudio.com/listener for privacy information.

Thanks for listening to Odd Lots. Follow the show on Amazon Music for more future episodes or just ask, Alexa, play the Odd Lots podcast on Amazon Music. Today's show is brought to you by Vanguard. To all the financial advisors listening, let's talk bonds for a minute. Capturing value and fixed income is not easy. Bond markets are massive, murky, and let's be real. Lots of firms throw a couple flashy funds your way and call it a day. But on Vanguard, at Vanguard, institutional quality isn't a tagline. It's a big line. It's a very big. It's a firm. It's a few. commitment to your clients. We're talking top-grade products across the board of over 80 bond funds, actively managed by a 200-person global squad of sector specialists, analysts, and traders. These folks live and breathe fixed income. So if you're looking to give your clients consistent results year in and year out, go see the record for yourself at vanguard.com

slash audio. That's vanguard.com slash audio. All investing is subject to risk vanguard marketing Corporation Distributor. Put knowledge to work and grow your business with CIT. From transportation to health care to manufacturing, CIT offers commercial lending, leasing, and treasury management services for small and middle market businesses. Learn more at CIT.com. Put knowledge to work. Hello and welcome to another edition of the Odd Lots podcast. I'm Tracy Allaway, executive editor of Bloomberg Markets. My co-host Joe Wisenthal is away and so I have a replacement co-host for you. I'm actually really excited about this. It's Matt Levine. He is, of course, the Bloomberg View columnist and quite possibly one of the funniest most original financial writers out

there. So thanks so much, Matt, for joining us today. Thanks for having me, Tracy. All right, Matt, let's start this off with a thought exercise. Let's say you're an investor and you're interested in investing in the shares of a company. Let's say it's something like PetSmart. How do you actually go about figuring out how well PetSmart as a business is doing? Well, I probably go there with my dog and see what their stores look like. All right. That sounds like a very systematic approach. No, I probably read the financial statements, right? Look at the 10K. You would look at their publicly available earnings statements. You would try to figure out what's going on in terms of their revenue trends, things like that. If you were really fancy about it, you might go there with your

dog and try to see how many customers are visiting, but that's not really going to help you if you only go to one pet smart, right? Right. So what if you could do something different? What if you could actually see how many people were visiting a pet smart or pet smart across the country? What if you could see big data, as people like to put it, and figure out how well the Petsmart business was actually doing. Sounds like it would be pretty helpful. All right. So on today's episode, we are going to talk to a company that is helping investors do just that. And one of the ways they're doing it is by using satellite data. So having satellite imagery that looks at things like factory activity, things like how many cars are parked out in parking lots behind PetSmart's and Walmarts and other retailers around

the world to try to gauge how those businesses are actually doing. And I know you've written a lot about data on Wall Street, how people use it. So I think you might be into this topic. That sounds great. All right. So without further ado, let's bring in our guest for today. It is James Crawford. He is the founder and CEO of Orbital Insight. He's also a former NASA scientist. So Matt, you can ask him any robotics questions you might be into as well. All right, James, thank you so much. for joining us today. Hey, Tracy, very happy to be here. Maybe just to begin, you could walk us through what your company actually does, the kind

of technology you employ and how people like potential pet smart investors might find it useful. Sure. Over the last few years, there's been a tremendous growth in the number of satellites over our heads. And it's interesting, the enabler for that has been a lot of the same technologies that bring you cell phones. So the miniaturization of electronics, rapid reductions in the cost of launch. And so just a lot more images are being taken.

Now taking the image though is only the first start, only the start, because once you take the image, it just sits in it on a disk somewhere until somebody actually looks at it. And we're getting to the point where there's way more images and there are people that want to stare at them. So to continue your pet smart analogy, if somebody were to deliver to you a million pictures of pet smart stores, you'd be a long time looking through flipping through all of them trying to decide how many cars were in each one of them or what was going on in each picture. So what we really do is we complete that supply chain, if you will. We take the

images from all the different satellite companies. We run them through artificial intelligence software to count cars. We can count trucks. We can count train cars. We can count ships. You can look at agricultural fields, see what the productivity is likely to be. And then we aggregate up, add up all those numbers and deliver an analysis of, you know, how different retailers are doing, what the corn yield in the U.S. is likely to be, or other interesting economic questions. And who are your clients at the moment for this kind of data and what data sets are most popular? Because you mentioned a whole bunch of different types of things just then, like agriculture, retail, manufacturing, economics. What's most useful?

So we've been going through a, you know, so we're a startup, we're a B round startup. So we've been going through a prioritization exercise with our prospects and with our customers. And the first thing we built was the one you alluded to at the beginning, which is the retail car counting. So we're now covering 100 U.S. retailers, providing daily updates on the number of cars we're seeing in their parking lots. Now, it's important to say we don't necessarily see every store of every retailer every day. In fact, we see a small fraction of each retailer every day. So you have to look at a moving average over time to get some statistically significant picture of what's going on in the different retailers. But we pull in a lot of images of retailers.

We've also been working on oil because there's so much debate right now about the price of oil, and there's so much volatility in the price of oil. And trying to understand just a basic simple question, how much oil is sitting in all the storage tanks. So all the oil that's been pumped out of the ground, but not yet refined. And that number goes up when too much oil is being pumped and there's not enough demand. It obviously goes down if there's a lot of demand. And so that's a really important, perhaps the single most important determinant of the direction of the price of oil. And it's not well known. It's pretty well known for the U.S.

But when you look across the world, it's not well known. So the other major product that we're shipping now is tracking the oil inventory, the crude oil inventory. And then a lot of the other things I mentioned are all things that we're working on is earlier phase products. that we'll have available, you know, in future quarters. So where do ideas come from for things to kind of look for account? Is that stuff that you and your team comes up with, or is that client feedback? It's really a combination of both. Our favorite thing to do, and we've done a fair amount of this recently,

is to get in a room with some creative portfolio managers from large financial managers from either hedge funds or mutual funds or other folks in different places, some from Wall Street, some from London, some from Hong Kong, and just brainstorm with them. It's like we ask them, what are your data gaps? What is it that you'd like to know about the world that you don't know? What would help you make better trading decisions? And then we run that up against what is feasible in terms of what satellites can see and what they can't see. And we've come up with a very long list of great ideas.

This is the challenge of being a startup is we have way more ideas for this than we have resources to actually build. but we're knocking them down pretty fast at this point and working down the list of the really top priority things we think we can measure. So retail, like store car counting feels like a thing that people have done for a long time. You know, hedge funds will send an analyst to their local mall, and obviously you can do it on a very different scale. Is there something that couldn't have been done at all before that is made possible by satellite technology?

that's like, that's just the totally new kind of data? It's a good question. In the case of retail, I think scale is tremendously important. When we do correlations with, say, SEC reported revenue, there's a tremendous increase in our ability to predict that with scale. And if you have just a few observations, if you just go down to look at your local store, I can tell you you're not getting a very good prediction

because a lot of these chains have, you know, thousands of stores in them. And so just working, at scale and the satellites that have launched over the last few years, plus the artificial intelligence to be able to count literally millions of parking lots is really critical. It qualitatively changes the value of the signal. The oil signal is one that would be very hard to do without satellites because these oil tanks are located in every country in the world. So they're in Singapore, they're in Hong Kong, they're all over China, they're all over Nigeria,

South America, Venezuela, the US, Europe, and you're not going to be a very, you're not going to be able to see them just driving past. It's when you can look from above and you can actually see the shadows on the top of the oil tank that you can actually get a sense of what's in them. People have in the past flown helicopters over the major oil fields in the U.S. to measure crude oil inventory. But you're not going to fly helicopters over all the oil fields in China or even all the oil fields in Europe and Africa and South America. So I think that might be an example of something that's just incredibly hard to do if you don't have this satellite coverage. potential clients come to you with ideas for data sets, that feasibility is one of the things

that you consider. Are there any other considerations that you have to take into account, like privacy or trade secrecy? Like if a hedge fund comes to you with an idea saying, for instance, they want to figure out the future price of hog futures. So they want to look in the backyards of every American family and figure out whether or not they're buying barbecue sets or something. like that. Would you do it? That's an interesting question. I don't think you could see that. So the legal limit for images in the US is 30 centimeter pixels. That's about the size at the top of your laptop. So you can tell cars. You can see, you know, how many cars are parked in people's driveways.

You can see how many cars are parked at a Walmart. You can't really tell whether it's a Ford Fiesta or a Mazda M3 or something. But you can roughly tell a car from a truck. I don't think you can tell whether somebody's got a barbecue set. Generally speaking, we work at such a broad level of aggregation. So we might look at, you know, how many questions similar to what you're asking, we might look at how many solar panels have been installed in all the roofs and all of Colorado and how fast has that grown over the last five years. That tends to be the level of aggregation that we work.

The imagery doesn't, it's hard to get enough imagery to work at a very low level. level of granularity and you have what you said, sometimes privacy concerns, although our resolution is so poor, I don't think that's a major issue. So usually the limitation is, do we have enough imagery and is the imagery of sufficiently high resolution to see whatever it is that we want to count? And then we count it, as I say, very coarse levels of aggregation for investors. But even if you're aggregating the data, if it's for something like, say, manufacturing activity in China, where the government publishes official statistics, it publishes PMIs that a lot of people mistrust.

I mean, I'm sure China doesn't necessarily want a bunch of satellites pointed at it, saying actually it looks like activity from your manufacturing sector is slowing, a lot more than official figures suggest. Does that ever come up as an issue? No, not really because the control of the satellites rests in the country that launched the satellites. So we are mostly using satellites. We're using satellites from all over the world, but the majority of them are flown out of either the U.S., Canada, or Europe, and a few other countries as well. But there's no – when you put a satellite in low Earth orbit, it necessarily passes over every square foot of the Earth, about every two weeks in each individual satellite.

So the government of China can't control this. The only restriction the U.S. government imposes generally is if there's areas where U.S. troops are in active combat satellites that are run by U.S. companies are not allowed to distribute the imagery of those regions. But that's, you know, incredibly small percentage of the world. So generally speaking, no, this is not a problem. It's a matter of providing visibility for everybody. And we don't single out any particular country, any particular industry. You know, we're trying to understand, you know, very broad trends and provide everybody a better insight into what's going on in the world. Is your product sort of reports and insight and analysis, or are you, like, in some cases, like, feeding raw data to algorithmic trading firms?

Like, are people coming to you for kind of, like, raw signals or for a higher level insight? It's actually some of both, and it tends to be, as your question implies, it tends to be more the quantitative firms that want the raw data. And a lot of the more fundamental firms, they are a little bit more interested in aggregated results, charts and graphs that actually give them some higher level insights into what the data is saying. How much do you actually charge for this data? Yeah, unfortunately, we don't give that out publicly, and it varies a lot by, in the case of retailers, for instance, by how many retailers the individual customer wants to track.

Okay, we are going to take a short break for a message from our sponsor. We'll be back in one second. Put knowledge to work and grow your business with C.I. From transportation to health care to manufacturing, CIT offers commercial lending, leasing, and treasury management services for small and middle market businesses. Learn more at CIT.com. Put knowledge to work. Okay, we're back with James Crawford. He is the founder and CEO of Orbital Insight, and we are talking satellite data and analysis, how investors on Wall Street can actually use it. you know James I just asked you about the cost and this gets to I think one of the issues that people sometimes have with these sorts of businesses which is that we're ultimately talking about

proprietary data that sometimes you have to pay a lot of money for that isn't accessible to mom and pop or your average retail investor and people sometimes think that that's unfair how do you respond to those so I guess I would say that having spent a lot of time with the financial investors, the hedge funds, as well as the mutual fund guys in the general financial community, that the number of different data sources these guys are working from in modern investing is really pretty impressive in terms of what they get from social media, what they get from folks like 4Square, from the credit card companies, from us. And I think it's difficult overall for individual

to compete with that on a retail name by retail name basis. And I think that's probably unfortunately, or fortunately, depending on how you look at it, going to be more and more true going forward, that individuals are going to be primarily either in broad indexed funds or in funds that are managed by people that actually do aggregate up enough investment capital that they can pull in.

It's pretty rich collection of data. Because the folks we work with, It's not like they use, you know, SEC reports plus orbital insight data. They'll be pulling in literally dozens of different data sources to create mosaics of information to inform their thinking about these investments. Is your client-based, like, does it skew to sort of quantity sophisticated hedge funds or like the big mutual fund complex is also using your data along with other things? Actually, actually both.

We've got a really nice mix of mutual funds as well as quant hedge funds and fundamentals. We actually have by count, we have more customers on the fundamental side, but we get more revenue from each of the quant funds, just because they typically, if they buy the data, will buy every single name. James, can you give us an example of when your technology or analysis really surprised you or a client, like where it found something really counterintuitive or something that you weren't expecting? I can give you one recent example where it wasn't really something. It was a little bit of a surprise.

I was surprised at how clear it was. So we had heard some people say anecdotally that they felt like the internet, the online shopping was affecting the lower grade malls more than the grade A malls. The grade A malls are typically the ones that will have, you know, shopping is an experience where they'll have roller coasters in the mall and the mall will have

100 stores and they'll have Santa there at Christmas and so on. And whereas the B-C-D-class malls are more like strip malls. So we had heard this speculation, but only is a speculation that the Internet was really hurting the lower-grade malls, whereas not really affecting the larger malls. And when we actually aggregated the data together for the last five years for U.S. malls, it just immediately jumped out that the overall car counts year on year kept going up for the for the top rated malls and and we're gradually sinking for the C&D class malls and it's like wow I guess I guess those guys there when they were when they were thinking about that they know what they're they know what they're thinking about can you this is sort of a dumb question but can you can you tell me a little

more about the process of like going from here's an idea that we want to know how to count to having the sort of software to count it so you know you talk a lot about cars at mall parking lots, which is sort of a, you know, colored rectangle on a big black rectangle. Whenever I read articles about this kind of thing, there are always these beautiful geometric pictures. And then the caption says, these are, you know, terrorist fields in China. And I sometimes wonder, how do you know?

And, like, is there sort of, like, is this a sort of primarily, like, the software can kind of figure out what stuff is? Or is this human analyst looking at pictures and trying to teach the software how to match patterns? Or is it sometimes human analysts getting on a plane and saying, you know, what is this thing that we're looking at and trying to go look at it in person? Yeah, I actually think that's a great question. So we tend to start, let's say somebody comes in and they want to track the development of wind farms in China, just as a random example. So we would start just by pulling up the images and let's see what they look like from space. What do these wind farms look like from space?

Is it clear when a human looks at them where the wind farms are? If it's not, then, yeah, we may have to get a wind form expert on the phone and have them look at the image and explain to us what these things look like from above. So we get into the point where we have a few humans who are able to actually find the object we want to count. Then we build what we call the labeled training set, where we have the humans actually go in and click the mouse on this thing that we want to find, whether it's trucks or windmills or solar panels or whatever. We build a labeled training set, often consisting of hundreds or if we're doing this,

if we want to get really high accuracy, thousands of images. And from that, the machine vision algorithms can take over and they can learn inductively from those examples what this object looks like. And then we can count, you know, a billion of the thing, which is what we did with Carr, as we trained it on a few thousand images, and then it counts. We've now counted 3.7 billion cars. It's a process that takes some work, yeah. What's the accuracy rate on cars?

Like how many shopping cart return areas or, you know, painted squares get counted as cars? Do you have any idea? Yeah, we've actually gotten to be quite good at that because that's been something we've been working on now for a couple of years. So we're about 95% accurate on cars. So if you train up the analysts and manage to pinpoint things, as accurately as possible, there are still limitations to the value of that data, right? So you can see how many cars are parked at shopping walls or at pet smarts or wherever,

but you can't actually tell how much people are spending once they're inside. That's right. There's always what we call confounders, things that confound you when you try to do the analysis. So, right, so not knowing exactly how much people spend, not knowing whether they actually spend anything at all are our major confounders. confounders for in the case of retail. Another confounder is if you have a multi-layer parking garage at some malls, for instance, you can only see the top level. And if you have somebody going into a mall, you don't necessarily know which store they're going to. They may park by the Macy's, but not actually shopping the Macy's, just walk straight through it and shop at some other store.

That's what my dad does. Yeah, there's a variety of confounders. And that's why you don't get, you know, If our data shows an increase in car counts, it doesn't necessarily always mean there's an increase in sales. It simply gives you, basically, it loads the dice in your favor if you're an investor. What do you think the future of this business is? Like in five or ten years, is it going to be an absolutely massive industry? Or does the fact that most of this is proprietary data, sometimes expensive, does that limit its ability to scale up?

No, I don't think so. I think there's a, I think it's tremendous opportunity. And the main reason we think that is that the availability of imagery is only going up. So within one to two years, we expect to have daily imagery of the Earth at maybe three to five meter or pixel. So pretty coarse, but still the whole Earth every day. Five years out, the kind of time frame you're talking about five to seven years out, we expect to have, you know, reasonably good resolution, you know, a meter or less per pixel of the whole world every day. And then we can track not only retail traffic, we can track mining, we can track manufacturing, we can track the car manufacturers, we can track ports, we can see port work stoppages,

we can see approximate imports and exports in different countries. We can basically track the physical aspects of the economy at that level. And that becomes valuable not only for investors but also for governments, for non-governmental organizations, for other Fortune 500 companies that are trying to plan their supply chain and understand what's going on in the economy that surrounds them. We've also been working a lot recently with non-governmental organizations. We've been working with the World Bank on poverty mapping so that we can help them understand where poverty is and where it isn't and how it's changing because a lot of these places

they can only do surveys, poverty surveys, once a decade. And obviously our world is changing faster than that. So we actually think over time this becomes a foundation of economic analysis, of understanding the physical aspects of the world to track economies for all kinds of purposes. And I do think that over time the individual, what we call signals, the individual signals like just car counts or just truck counts do get cheaper and then what people are actually buying and actually using is an aggregation of all these different things that we're able to count going forward.

In that future, are you sort of still the analysis layer or at some point our hedge fund saying, I want the data, I want just millions of images and I want to make my own proprietary signals and my own proprietary analysis of it? I don't think there would be the mileage for them to take the images. I think some of the quant funds are already looking at a pretty granular level at the counts. So some of the quant funds will actually take, you know, address date count and that's essentially the data that we give them and they work from there. So I could see that going forward but the analysis is something that we use the same analysis

routines for government customers, investors, insurance companies, energy companies, you know, all of our customers we use the same image analysis and a lot of the same data analysis routine. So I think there's a lot of economies of scale for us in that. But at the same time you're right, some of these quant funds especially get pretty granular in the data that. take from us. Okay, I think that's a good place to leave it. James Crawford of Orbital Insight. Thank you so much for joining us. Absolutely. Thanks for all your great questions. All right, Matt. That was your first episode of Odd Lots. How did you find it? I thought that was fun. I've always been interested in these sort of imaging and proprietary data

companies for some of the reasons you alluded to. I write a lot about insider trading. And one thing I always wonder about is why people get so upset about the lack of a level playing field between retail investors and professional investors. And as James said, you know, there are so many sources of data that professional investors can rely on. You know, flying helicopters over oil fields, flying satellites over oil fields, it seems silly to worry about any one source of data. Yeah, there was one thing he said that kind of depressed me where it was basically like retail investors don't have any hope of competing with the big guys in terms of information flow. And so everyone, you know, mom and pop are just going to increasingly herd into passive investing

and index funds. I think that kind of worries me a little. Oh, not me. I think that's, I think that's been true forever. I think if it's your job to invest and you have the tools to invest, you're going to be better at it than someone who's doing it as a hobby or something they just, you know, do in their spare time. I don't think that you have to hurt to passive funds, though. I mean, you can hurt, you know, if you're a mom and pop investor, you can invest in the mutual funds that it sounds like use his data, you know, alongside the hedge funds. So, you know, the point is not passive versus active or retail versus professional, really. It's that, you know, there is a professional management layer that decides where to invest money

and as a retail investor, you can have access to it. I guess where I get uncomfortable is the differentiation or lack of differentiation between access to information and intelligent analysis. Like one guy can get ahead just because he has better access to information flow, whereas another guy falls behind because even though he's super, super smart, he just doesn't have that data. I think that's what depresses me about the whole thing. I hear you, but there's a lot of data in the world, right? I mean, there's more being generated all the time.

And it seems a little hopeless to say we're going to give everyone all the same data at the same time. Because, you know, I understand what you're saying, but what would you do with all this satellite data, you know? Would you analyze it if you were a retail investor and you had to work a day job? Yeah. Alongside my day job at Bloomberg, I would be looking at satellite images for hours and hours and hours on end. That sounds totally like something I would do. All right. Matt, thank you so much for joining us today. That was really good fun.

Yeah, thank you. Hopefully we'll have you on again at some point. I love that. All right. I'm Tracy Allaway. You can find me on Twitter at Tracy Allaway. And I'm Matt Levine of Bloomberg View, and you can find me on Twitter at Matt underscore Levine. Thanks for joining us, everyone. Take care. Put knowledge to work and grow your business with CIT. From transportation to health care to manufacturing, CIT offers commercial lending, leasing, and treasury management services for small and middle market businesses. Learn more at CIT.com. Put knowledge to work.

The news doesn't stop on the weekends. Context changes constantly. And now Bloomberg is the place to stay on top of it all. Hi, I'm David Gurra. Join us every Saturday and Sunday for the new Bloomberg this weekend. I'm Christina Rafini. We'll bring you the latest headlines, in-depth analysis, and big interviews. All the stories that hit home on your days off.

And I'm Lisa Mateo. Watch and listen to Bloomberg this weekend for thoughtful, enlightening conversations about business, lifestyle, people, and culture. On Saturday mornings, we put the past week's events into context, examining what happened in the markets and the world. That on Sundays we speak with journalists, columnists, and key political figures to prepare you for the week ahead. Join us as soon as you wake up and bring us with you wherever your weekend plans take you.

Watch us on Bloomberg Television. Listen on Bloomberg Radio, stream the show live on the Bloomberg business app, or listen to the podcast. That's Bloomberg this weekend. Saturdays and Sundays starting at. 7 a.m. Eastern. Make us part of your weekend routine on Bloomberg television, radio, and wherever you get your podcasts. What separates good leaders from transformational ones? I'm Jessica Chen, and in season two of Leading By Example, we'll sit down with executives like Grace Chen of Bertie Gray to find out.

It's important to understand where you spike, but also really acknowledge where you don't and find people who can fill those gaps. Listen to leading by example executives making an impact on the IHeart radio app, Apple Podcast, or wherever you get your podcasts.