How Traders Used Google Searches To See The Economic Recovery In Real Time

the pre crisis peak to this one. So I think the SUP peaked at um February on February fifteen, and then we saw the the new peak yesterday, August eighteen. And in a sense, it really feels like we've compressed this sort of gigantic cycle into an extremely short period

of time. Yeah, that's true. And I was looking at the latest fund managers survey from Bank of America and it showed that I think fund managers have completely flipped from thinking that we're in a recession to thinking that we're in the early stages of a fresh economic cycle. And if they're right to your point, it does suggest that we've just seen, you know, one of the shortest

recessions of all time. Yeah. I mean, you could make the argument that the recession recession in terms of the shrinking of growth was done by the end of March, when most data points started turning up. And while the overall level of economic activity is still very depressed, and of course unemployment rate is still above ten percent, so hardly time to be declaring victory. We have seen steady improvement on a host of economic data points basically since

end of March early April. That's true, but I also feel like there's something kind of weird going on with the data. Like there's the old stock first flow argument, which we're seeing everywhere, but particularly in p M I. So even when we get a big rebound in p m I s, it doesn't necessarily mean that we're getting back to the levels that we saw pre crisis. But you're also seeing just sort of weird indicators that are

happening simultaneously. And I think one of our colleagues pointed out a really good one recently, and that was intentions to buy a house surging at the same time as mortgage delinquencies, which I mean never happens in an economic crisis. Now it's really weird. But I think because of all the weirdness that we're seeing this sort of contrary indicators, because there's this weird gap between pieces of change which have been very fast and unexpected versus levels which are

still very bad levels. And then also just the fact that it's so compressed, there's probably never been more demand for sort of alternative real time data points and this feeling that the official economic data points that we get monthly jobs report, monthly retail sales report, they just there's not enough of them. They're not timely enough to get a sense of what's going on, given how fast the

changes have been both on the downturn and the rebound. Yeah. Absolutely, and I mean, just on a very simple basis, everyone wants to know what's going on with the recovery, right, and everyone's tracking to what degree the economy has reopened, and some of the most useful indicators for that are arguably alternative economic indicators like um like open table reservations, things like that. Yeah, totally. I mean, that's like one

of the things we've been watching the most. It's like open table they could keep track of people making reservations or doing in seeding dining, so if you want to sort of understand how behavior has changed or how people are doing different things. Um, due to the virus, that's been one of the sort of key data points, not something that people were really tracking before as far as I know, on a meaningful level. So I think that's

really important. I mean, I think obviously real time alternative data has never been more in demand than what we've seen over the last six months. But I don't think it's going away now. It's kind of another one of these things where real time data points of a range of things will sort of be part of the conversation for a long time, even if and when we get back to something resembling a normal economy. Yeah, I think

that's right. So today we're going to be talking all about alternative data, what it's showing, and more importantly, how investors actually use it in their process. And so we're going to be speaking with Ben Brightholtz. He's a data scientist at Arbor Data Science, which is part of Arbor Research and Trading. I've been following their stuff. They do some really interesting things with looking at Google search trends for lots of different keywords and trying to divine an

economic significance from them. So let's talk more about that. Ben, thank you very much for joining us. Yeah, thank you very much. Joe, happy to be here. So let's just start a big picture. What do you do? What is arbor data science? Talk to us a little bit about

your work. Sure, So over the years we've gotten more and more into essentially this idea of filling the gaps between latent economic data and the econ data that can be distorted like we've seen with unemployment data as of late, and also really trying to help our customers and the investment space in general deal with surveys that have been more or less leading indicators for quite some time. They've

kind of fallen flat on their face. And this is something that's taken place well before UM even the current episode we're going through now, looking back to the financial crisis, with really the polarization of the country and the world on a political space, and really the advent of social media has created really this bifurcation and in sentiment it could be republic and democrat or it can be more

or less group think. Based on UM, the use of Facebook, Twitter, we create all these small microcosms we essentially live within and that is ultimately distorted the ability of survey data, for example, to have this leading nature that it used to have really for decades um. And that's posed a significant problem for investors that are in putting this either on a subjective level or within their own modeling to then project board where they think financial markets will go

in the future. I have a really basic question, which is what's the difference between big data and a large set of data? So big data is such a misnomer and um nasty term, you know most I think big data is a term that's kind of slowly gone away that I think the initial idea is that it's it's

unstructured data. That's for example, you can find all this wonderful information on a Bloomberg terminal, all right, and it comes you can download it via a p I or access at via via your your Windows or your terminal, all nice, clean and easy to use, ready to input. And big data um to me. Uh, this day and age, especially with alternative data, has to do with more or

less unstructured kind of ugly data. So this, for example could be all just like us talking right now, or when you are all on TV, you have all of this this text, this closed captioning that exists out there, and let's say, for example, it's in fifteen second increments, and it can be ugly, it can be have plenty of errors within the data within the closed captioning um And essentially we have to use algorithms and different processes in order to take that unstructured data and make it

something useful and really turn it into something that's more or less numerical in order to benchmark against financial markets, econ data, overall sentiment and so on. So, you know, big data is kind of a word. I think that's somewhat going away. But to me, again, it means somewhat of an unstructured data set. So I'm thinking about what you scribed as the problem with surveys, and uh, you know, I think it's either the University of Michigan Consumer Sentiment

survey or the conference board one. There's one of these data points that we have it on the Bloomberg terminal, and it's like they say, it is now a good time to buy a washing machine? Is now a good time to buy a car. There's even one that's one of my favorites. Is now a good time to buy

a vacuum cleaner. But I guess what you're doing is you don't have to ask people is now a good time to buy a vacuum cleaner, because in if you know how to find the data, you can just look at searches for vacuum cleaners and that's presumably a lot more reliable than asking people into survey whether or now is a good time to buy a vacuum cleaner. Right, So, the within surveys, there's and there's a plenty of studies on this as of late showing that respondents will not

provide really honest answers relating to their financial hardships. So there's there's large gaps and you know, our things better now or worse? Are you going to spend do you have the money to spend here moving forward on a vacuum, on a new washing machine? And so on? And there's always been a gap, for example, example, between the web based responses and phone based and we saw this too

with the election. That's a whole another other topic, but um on a web based survey and individuals are typically much more honest than they are regarding financial hardship than they are on the telephone or basically being put on the spot. So the idea here between behind search activity and this is something that I think that has improved in most recent years, is yes, we can get ahead of this intention of consumers and we're not necessarily we're

not really going to lie to that little window on Google. Um. You know, we might lie maybe sometimes to our girlfriends or our boyfriends or husbands or wives. Um, but you know, what we put into that search window is really truly what we're seeking and what we're actually trying to query.

There's no no one really looking over our shoulder. So our belief is that search activity, um, really, over the past five six years has become kind of a great estimate or indication of the consumers intentions of what they plan to do. Am I going to buy a wash machine? Or if I'm in distress, what does it mean that by default on my credit card payment or I don't pay my credit card payment? Or what if I need to go out and search and find a bankruptcy lawyer.

These are the type of things we can pick up on, uh, you know, within this information to then create a kind of um, you know, overall look at the consumer. And this can be all the way from the you know up towards the United States, the complete um, you know, country level, it can be worldwide, and it can be drilled down all the way down to a metropolitan area UM. And again, the whole idea there is trying to get

the most honest representation of the individual. And I'll also say that the growth UM in the Internet and really access to the Internet, both mobile and on the PC, has been a big boon for search activity, so that you now have fifty of the world having Internet access and using it on an active basis. That's more than four and a half billion individuals, which has really doubled,

if not tripled, since the financial crisis. So I think early efforts of using search activity UM is, I know, a lot of it pre crisis kind of fell on its face and kind of faded away. Google used to have these curated indices um. I think they had twenty five of them, kind of showing how the economy, economy was moving um here and there. I think that that what didn't work as well because we didn't have the

ubiquity of Google searches and really Internet access. And as that improves, this type of information becomes that much more important. I think to the investing process. How much do you think the the unusual or the extreme circumstances surrounding the

coronavirus crisis are are distorting survey responses? And I asked that because again, I've seen a lot of criticism of the p m I s recently, and one of of things people are saying about those surveys at the moment is that respondents aren't really judging their experiences on a month to month basis, but they're sort of responding by comparing now to a period of relative normality. So everything's getting skewed. Do you think the unusual nous of of

our current circumstances might be skewing survey data as well? Yes, I think so. I think it's it's a combination. Like you said earlier with stock flow, it's what type of reaction have we had over the past couple of months. UM I think has been more reflective within the survey data, and we're seeing that breakdown between surge activity UM and surveys. And we also have this big group think are almost circular reference that occurs within a lot of the sentiment data.

So we all look to the equity market. We all know that we can use the equity market essentially forecast where consumer a confidence will be for the next month UM, and I think a lot of that's feeding into some of the more rosy consumer confidence numbers as well as the UM eyes and again that's some somewhat of distortion UM and why we seem to't like to rely on the search activity for the most part. So let's talk a little bit more about that search activity. How do

you take how do you get the data? First of all, what does Google make available? And then how do you present it in a form so that it's usable because there's obviously seasonality factors the you know, you can't just look at searches for a vacation and see whether they go up or down because people don't vacation at the same uh at the same pace all year round. So how do you get the data from Google? What's that

process like? And then what do you do to actually put it in a format such that it's not just OLiS for investors, like just describe it overall? How work? Sure, so we are able to access just like anybody else via Google Trends, which there is an API to be able to grab that information, and what we do is we avoid using the specific search terms. So if we're just going to say wash machine or vacuum UM, that

will include specifically that exact term UM. And we know that there can be multiple variations of those actual text terms, and so we want to pick up on that. The beauty is Google curates and creates two different types of groupings of search activity. And they do this for you know, each and every country essentially, which is going to take care of the major language barriers and issues that we'd

run into as well. And so that is they create categories, which there are roughly und forty plus different categories, everything from accounting services all the way out to urban transportation which would be things like uber and lift. And then they have topics and that can be anything from inflation or those talking about disinflation, or gold bugs or bitcoin um.

And that's going to then be more encompassing and based on their mapping of a numerous new it could be hundreds, if not thousands of thousands in certain cases, of different search terms and phrases that then get housed underneath those individual UM topics. We can I stop you and ask you a quick question right there, Sure the data the year able to draw. Just to make clear, is that the granular within those hundreds or thousands of terms, you're able to get data for each one of those. You

could see beyond just the sort of general category. Yeah, so we can drill down. There are ways to drill down within the individual categories. We understand what the actual searches are within those categories, but in order to create a more encompassing indication of what the consumer business is looking for or thinking about, we do then pull in that search trend. Essentially, that's going to be an aggregation of all of all those searches underneath a given topic

or underneath a given category. And like I said that, one of the greatest things about the way that Google set this up is that you are then able to say, let's look at urban transportation uber and LIFT, and let's look at it not just here in the US. Let's go to UH somewhere like Germany, let's go to Australia, or let's go to Japan UM and they take care of, fortunately a lot of the language barriers in that urban transportation that is translated into into um, you know, Japanese

or um you know whatever is being German, and so on. UM, so getting into the course of how we then digest and use that information. Like you said, is there's a high degree of seasonality. Of course, it could be like with clothing with back to school, or can be accounting services coming into March, April and October. UM. So we do decomposition where we'll we'll break down each individual topic or categories search activity into three components, and that is

it's overall trend component. You think of it as like as kind of a slower moving average trend of that search activity. And then we have the seasonality that we're able to then strip out. And then we also have

this thing we call the residual or the shock. What's interesting about the experience that we've seen here in UM with COVID nineteen is we were never so interested in the shock component and the very quick um shifts in search activity either positive or negative until COVID hit, when we saw it's just substantial breaks from these trends and

what would be expected by seasonality. That can be anything from the searching for you know, physical policy news, economic news, how individuals are searching on the line, then for groceries UM and making those type of consumer staples purchases. But getting back to it, the idea is to break it down into those three components that we get idea of what is the you know, the long term trend UM

and shift really potentially in search activity. How does that relate then to to what we're seeing within financial markets and overall economic data. And then what are these shock components and regarding those big distortions or shifts away from those underlying trends, what does that have to tell us about how things may be abruptly changing in the near

term UM and what that can mean? I mean, of course for potential volatility UM and equity markets, uncertainty in general UM from the consumer base UM and so on. And so what we do is we pull down those three pieces of information that then gets used within our written content as well within our own models and our clients models UM and so on. So correct me if I'm wrong. But the data that you're using is mostly public data. If investors all have access to the same data,

how are they using that to actually generate outperformance? How do they differentiate how they're using the data versus how another fund or another investor might be using the data? Right? So I mean that's that's the question we get. We probably get the most most is since we do deal mainly again with with public forms of data, there's plenty of alternative data that is private and the credit card space and spending UM and so on. Is we try

to uncover data we think that is underutilized UM. And in this case with all of our dealings specifically with fixed income portfolio managers, pension fund managers UM and the like, the use of search activity on a broader scale, on a country by country, even a metro by metro level, we believe has been under appreciated UM and non internalized

the extent that it could be. Now, like you said, once, think something like this gets over used or gets used as a key benchmark potentially to filling the latent gaps between economic data. Potentially some that alpha creation could UM evaporate UM and that would mean we have to move on to some additional data sources for this time being. In all our communications, the front offices of of investment managers, banks and so on have not been heavy users of

the search activity. I think that early uses of it prior to the crisis and during the crisis kind of fell flat. Again. Maybe the ubiquity of actual Internet usage and those young too old that we're using Google was not there as of yet. And what we've seen over the years, really since two thousand eleven two thousand and twelve, search activities ability to fill the gap and really take the place of surveys has improved markedly year after year UM.

And that's something we can measure UM statistically and VR modeling for essentially those turning points as to when maybe search activity loses its flare loses its ability UH to then forecast and now cast via g d P retail sales inflation UM and the like. But we're not there yet.

So obviously the demand for this data, and you mentioned maybe search data is sort of relatively newly being incorporated into investment processes, but for years we've been hearing about satellite looking at parking lots in Walmart, or satellite looking at trained or credit card data that's been out there

is a thing for a while. How intense is the search basically for new data sources, either on the bi side, the investor side, or you as sort of a data vendor so to speak, to just constantly be coming up with something that's relatively underappreciated. What does that process look like the use of alternative data within the investment world. You know, really the investment world was very late to using alternative data UM, you know, compared to healthcare, even education,

UM and the like. And we initially saw this, you know, in our routines of going out to big banks, for example, and discussing with their teams, UM, you know, how they're utilizing alternative data. It was almost always in the back office. So it could be UM, you know, anything to do with their customer relations. It could chatbots in terms of creating natural better language, natural language processing and ployment data for that. It could be trade matching all kinds of

different things that were done in the back office. They were trying to basically bring in machine learning, bringing better data to create better predictions. And it could have to do again with their customers, which customers to call and not call, who's going to potentially provide the best UM,

best avenue for new business UM and so on. But what we've seen, i'd say, you know, starting roughly inen, we started to see a UM with the advent of more alternative data available via numerous vendors, the increase in transfer to the front office has happened rapidly UM. And I would say that now with COVID nineteen and the inability for econ data to keep up with the actual UM happenings of the economy UM, and really the needs of investors to understand that just what's going on with

how rapidly things are changing. UM. The demand is just intense, and so it calls to us and and calls I know too many of our competitors in similar, similar alternative data providers. UM has just shot to the moon and you can see that again. Bloomberg of course offers some

of this alternative data. There's plenty of other repositories to grab it, but I would say that the degree of interest is increased tenfold UM since it's it's beginnings in what's been your favorite alternative data set during the crisis, Like what has either surprised you or what has been most useful in judging the direction of the overall economy. We've been benchmarking a lot. The mobility data that's available via Apple UM and to cart lab is another one.

Google and Benjy benchmark in that off of search activity, and I've been absolutely shocked at how well. Search activity has been able to predict two things, and that's been retail sales on a month over month basis and also

inflation on a month over month basis. A lot of our kind of point forecasts looking forward, based on what we believe are the most unique and important search activity have done very well UM in predicting the rebound in May, for example, the heavy damage done to transportation, energy, UM

and apparel within March and April. To CPI for example, we had UM noticed the heavy degree of rebound in all three of those categories, in particular UM within apparel, which ultimately lead to rebound in overall apparel spending in May, which then ultimately translated to higher inflation UM it was

reported in June. And so the search activity that we've been able to use most utilized, which I think Joe featured in a CHARTUM a number of weeks ago, has to do with a series of key categories, and that can be everything from beauty and fitness, which is we found to be a highly leading indicator UM as well as just the general public searching for economic news and physical policy news revolving around welfare and unemployment and jobless

benefits welfare and unemployment. Unemployment itself has been a highly leading indicator. And then also UM, one of the things we picked up on very early was the incredible drive for home improvement that really began in the final weeks

of March. UM. And what we had seen was this effervent search activity UM, you know, looking across all the major metros and all the major states of the United States, heavy degree of need for our need, a desire to place appliances, to paint their homes, to get a new roof,

new side, new carpeting, UM. And this is something that really took place ahead of the Hares acting signed on March seven, It began really two weeks before that, which I think was a leading indicator that the consumer would be stronger and potentially spend more UM than those uh, the naysayers. And then that we had expected UM to see given the calls for a recession and potential depression, given the full stop to the economy. And it's really striking just this week, UH, we've seen home depot and

lows post extraordinary sales. Home improvement has just been one of the monster stories of this recovery. How much spending and how sustained that's been I just want to drill

a little bit further down. I mean, it's clear that like okay, if someone identified that trend at the end of March and I saw what was going on, there were huge investment opportunities because like again, like I said, home depot lows, it's that are huge beneficiaries that their stocks about extraordinary runs due to this, uh desire for people that like renovate and fix things in their home

while they're working from home and so forth. How then, do in your clients and when you talk to them, how do they actually make a decision by or sell based on the data and the context that you're giving that What is the the you know, that's sort of the last mild question, so to speak. They can get the data from you, but then how are they actually using it to form of you and take a risk both on a subjective then also on an algorithmic basis.

We have many, many clients that are effectively now casting, and so they're now casting either econ data, the econ environment, and then as well the financial the impact on the on the actual financial market in terms of producing their own actual forecasts of where things will be one week to six weeks to twelve weeks later. So the search activity UM is one that we found provides a lead time. UM. That's more you know, kind of medium term as opposed

to ultra high frequency short term. So uh, you know, within the searches, just like survey data, UM, we're not going to be able to help someone UM if if effectively make a decision for that day. You know, what is the next twenty four hours of economic activita? People buying more watch machines, they buying more cars, m Are they buying more apparel? It's not That's not exactly how it works. It's a more immediate term, medium term focus of UM. Varying lead times typically from one week to

eight weeks. So we have things, for example, like apparel UM that will have a lead time of days to a week UM, and then we'll have things like building materials or roofing that will have a lead time of seven to eight weeks. So then what our customers and our clients are doing is taking that information in understanding those lead times and then either in putting it to their own subjective decision making process in order to affect

their decision. It could be a risk management one in regards to their actual book or their position, determine if there's something that could be disruptive to their position, or it could be on the flip side, someone that's actually UM, you know, using on a more tactical basis, that is, and in putting it to their own now casting forecasting process and that coming up with their own conclusion UM of how will that supports or doesn't support their their

general idea. But UM with this data, along with a lot of the natural language processing data that we work with, UM does not have a high frequency basis. This is something that's more medium term, if not long term in nature. Then last thing like where do you see what's the next big thing for you? In terms of I just thinking back to when you said, Okay, at some point the search data will get more used, the it will get more modified, the alpha from having access to it

will theoretically diminish. What are the next frontiers in terms of data that you think are interesting and potentially still underappreciated or underutilized at this point. So I think the advent of mobility data, for example with discard Dicart labs that's able to zero in on specific retailers and look at the actual foot traffic UM that's occurring coming to them going away from them. It can be also down to you know, parks UM specific locations within different metros

or rural areas. I think this mobility data, which we don't have a high degree of historical data to work with, is something that moving forward will become more and more of the leading indicator that think individuals will seek for. Unfortunately, you know, Apple, Google and the cart labs have you know, they sell this data, so it's not necessarily publicly available.

But I think as they build a larger and larger track record in order to benchmark that against anything begun search activity, survey information of how consumers are operating, where they're moving, and what they're doing. I think that is more or less the kind of the cutting edge and leading edge of understanding the consumer and now and then how they're interacting with retail, interacting with people around them

using urban transportation UM and so on. And obviously in this environment of COVID nineteen with how much we were not moving around in March and April UM, I think it'll be critical UM here moving forward to get a better grasp on how much of a revival um economies. There's economies are seeing and how mobile people have become or it will be one of the fun ones we didn't talk about, but I know it's uh, it's definitely um, you know, fringe too, just because it's it's such a

strange space. Is that the Twitter sentiment is one that's become I think, more and more useful UM in terms of gauging actual investor sentiment. It's been pretty wild to watch the number of economists and even formal central bankers that have popped up on Twitter that use it pretty voraciously. We even have like Christia Freeland, Um, you just took

over a finance minister in Canada. There's just such noteworthy individuals, and it's it's become something that's become more and more predictive, I think, not necessarily a direction of equity markets, but more or less a gauge of uncertainty UM and you know, financial market volatility. So we've built a lot of algorithms too. It's just like another thing where it's kind of like because I know people were like interested that ten years ago,

but it wasn't enough interest. There weren't enough people on there for Twitter or social media to be representative, but sort of like kind of like search where you can actually get a big enough cross section that it's meaningful exactly. So that's the what's absolutely wild is the number of people provide providing original content and the speed by which

they are actually tweeting has accelerated just demonstably. So we saw this incredible crescendo in tweeting and Twitter activity in fin twitt through really the middle of March, and it's just held there ever since. With this COVID you know, pandemic everyone at home, UM and really grasping for information. So it's been fun to be able to break down all the different opponents of Twitter, which we do into

is based on clustering prior to the financial crisis. We break it down into primables, bears, primatists, UM, economy US UM and the like UH, and then able to grab out you know, how are they feeling about liquidity in the market, how are they feeling about the equity market?

COVID nineteen UM, I had the consumer UM and so on, And it's amazing pulling in the information, like you said, prior to just three or four years ago, its ability to actually get ahead of and forecast, you know, volatility and maybe a little bit of financial market direction is improved significantly. So it's it's it's been an interesting space to dabble into. Ben. That was great, Ben bright Hole,

I really appreciate you joining us. This feels like such a big area and there's such a clear explanation of how it all worked. Thank you for coming on. All right, Thanks Joe be so much fun. That's really interesting. Yeah, I thought that was great. You know, I do feel like just from us from a media perspective, we you've never used alternative real time data as much as we have over the last six months, and so I thought it was great to hear how it's actually collected and

then how it's actually used to put into an investment process. Yeah, it's funny like thinking back to this now, but I remember in I guess it would have been February telling someone about how we were tracking movie bookings in our theater bookings in South Korea because of the COVID outbreak there, and the person I was telling it to just thought

it was like so unusual and so amazing. But of course now everywhere around the world and especially in the US, people are looking at all sorts of those kinds of things, from restaurant bookings to the mobility data that Ben was talking about. Um, it's kind of become normal. Yeah, And I'm really fascinated by the sort of you know, the speed with which sort of alpha deterior rates. So you can imagine the first person who really discovers that search

indications for certain terms has some predictive value. There's a lot of money to be made in that. But look, I mean we're talking about it on the podcast that Ben's active, and pretty soon you have to figure that will be table stakes that people will be searching for the next, the next thing, that that is a process that will essentially never stop. Yeah, I think that's right.

But also I think what becomes clearer from speaking with Ben is that understanding the data, how it's collected, and how you can actually apply it is really really important. So even with something like the mobility data, it's very useful at the moment, but I think it's benchmark to early January or something like that. So it's really good to be aware when the summer comes around that the benchmark that you're comparing the data to might not be

you know, completely applicable to warmer weather. So there's all these quirks in each data set that you really have to get to know. Yeah, totally. I mean even with the Google Day it a just having the sort of experience to adjust for seasonality it takes. Those are all things that if you were to say, if I were to just look on Google trends and look at that vacations, it would be hard for me to get much signal unless I really like understood the data and had experience

working with it. Mm hmm, yeah exactly. All right, shall we leave it there? Yeah, okay, this has been another episode of the ad Thoughts podcast. I'm Tracy Alloway. You can follow me on Twitter at Tracy Alloway and I'm Joe Wisn't Thought. You can follow me at the Stalwarts, and you should follow our guest Ben Brightholtz on Twitter. He posts tons of interesting charts from the uh the Arbor research work that they do. Follow him at Ben Brightholtz.

Follow our producer on Twitter, Laura Carlson. She's at Laura M. Carlson. Followed the Bloomberg head of podcast, Francesco Leavi at Francesca Today and check out all of our podcasts at Bloomberg unto the handle at podcast I for listening.