February 11, 2020: Simon is buying Taubman. How to quantify the synergies?

Usually the first considerations when a company acquires another is how to realize economies of scale, cross-marketing opportunities, and areas of common improvement. If one knew which shoppers visit each location and quantified the cross-visits between malls then they could make educated decisions about which malls to focus on for improvements, tenant acquisition, or cross marketing.

With traditional measuring methods this is hard to achieve, but cellphone location data is the perfect fit for such an exercise.

But not all visitors are created equal. To get the true picture one needs to go a step further and adjust visits by the purchasing power of the visitors. Hence at Advan we computed the income of each device using its shopping behavior, and then adjusted the visits and cross-visits by income. Here are our findings.

First, looking at traffic alone, here is the breakdown of traffic by state. Florida has the highest number of visitors for both Simon and Taubman, and the highest number of cross-visitors: 27% of Simon shoppers also visit a Taubman mall. California on the other hand (which is the next largest Simon state that has Taubman presence) only sees 3% cross-visits:

The picture is similar, but more informative, when taking into account the income of visitors. We see that Florida is still the largest state, with $590 billion of annual purchasing power across all visitors, of which 28% ($168bn) also visit Taubman, and California is now #2 in Simon presence at $430bn purchasing power with only $16.5bn of that cross-shopping at Taubman:

Of course the full details can be gleaned only when breaking down the above by individual mall, where one can identify which visitors, and which specific cohorts of visitors (high income/low income, millennials / baby boomers, etc) are visiting multiple locations, and make educated decisions on where to invest and how to expand the efficacy of each mall.

January 31, 2020: Monitoring the coronavirus effects

Markets have reacted very strongly to the potential effects from the coronavirus epidemic in China. MGM, Las Vegas Sands and Wynn Resort stocks have been hit particularly hard, dropping 20% since the news, mostly on fears that their Hong Kong and Macau locations will see materially reduced visitors during the annual Lunar Year celebrations. But are the actual effects on business and travel what the market expects or is this an over-reaction?

To analyze the effects we decided to look at a few key metrics:

  • Macau and Hong Kong casino traffic. We mapped and measured the traffic on the 5 largest MGM, LVS and WYNN properties in Macau
  • Macau and Hong Kong airport traffic
  • US and Global airport traffic

Let us examine the data in detail.

First, the casinos in Macau and Hong Kong:

Which is materially lower than last year's traffic:

In the first week since the news of the virus, we see that traffic was fairly unchanged, and even increasing in the Las Vegas Sands locations. However, starting in the second week the traffic is starting to fall, and instead of the increase usually seen in the week after the Chinese New Year we see a prolonged slump, with the difference being on the order of -20% instead of the typical +20%, although the large increase after the New Year's is limited to about 1 week.

Second, the Macau airport:

Versus last year:

A similar story unfolds in the Macau airport, although the moves are more muted; the increase after the Chinese New Year was small last year and the drop this year is on the order of 10%.

Then we look at the Hong Kong airport:

Versus last year:

This supports the same results as the Macau airport, and in fact is a bit stronger; instead of a 10% expected increase after the New Year's, we see a drop of 10%.

And last, the US airport traffic and how it compares to international airport traffic:

The US airport traffic is the blue line and the Global traffic is the red line in the graph:

The effect so far has been failry muted when it comes to global or US airport traffic. The measurements are within the margin of error and do not indicate any panic or concern. Of course things change, and fast, and we are monitoring the situation daily, but so far any effects appear to be concentrated in and around China.

December 03, 2019: Black Friday is Holding Strong, but not everyone benefits equally

Department Stores, Restaurants, Malls, all continue growing their traffic on Black Friday on a year over year basis. The growth rate has decelerated a bit, but it's still in the healthy double-digit numbers, from an already high base.

Only Hotels showed negative growth, but less so than in prior years -- a weighted average of -1% vs -3% and -2% in 2018 and 2017 respectively.

Below are a couple of graphs that illustrate the traffic. Do not be fooled by the simplicity of the graphs, the data underlying them consists of 3 Trillion data observations across 2 Million geofenced locations pertaining to 1,400 companies, all normalized using Advan's custom algorithms for reliable year over year comparisons. Simple is hard to do!

Sector performance: This shows how many more people visited a sector on Black Friday than they did the year prior.

As we can see all Consumer sectors (which include Apparel, Clothing, Malls, Supermarkets, etc) grew in 2019 between 8% and 25% versus 2018. Their growth rate is down between 20% to 40% since 2018, but it's growth nevertheless. Certainly much better than the meagre 2017 situation where traffic was falling.

Drilling down to individual industries, the picture becomes clearer:

  • Department Stores and Restaurants lead the growth, followed by Malls ("REITS") and Supermarkets.
  • Growth is generally more muted than in the prior years.
  • Clothing, Shoes and Accessories are still growing, just not as much as they used to anymore. 1% vs last year's 4%.
  • Apparel is flat, vs a 1% growth in 2017 & 2018.
  • Hotels are slightly down -- people stick around with family apparently.
  • Airports show 0.5% growth (note, this is only on Black Friday, not the whole week of Thanksgiving), but note that
  • Gas Stations ("Integrated Oil Companies") grew 1.7%. It takes some driving to go to all these stores!

It is important to keep in mind that the growth patterns above are not uniform across companies. In fact, on average the Consumer sectors are down vs 2018. This means the big department stores / retailers / malls are dominating the growth (so that overall growth is positive) whereas most of the small chains are hurting (so that on average, traffic is down on a company by company basis):

The same story unfolds on an industry basis, where the Department Stores and Malls are overall slightly down on average, meaning the little guy is hurting, whereas the high trafficked stores see their traffic growing each year.

November 13, 2019: Miles Driven and Trucks Miles Driven: macro indicators

One of the interesting things you can do with access to anonymized, large scale and accurate cellphone location data, is to measure traffic patterns. Consistent with the Big Data theme, we decided to compute this across the whole US, to measure Total Miles Driven.

Total Miles Driven is estimated by the Federal Government (aka "FRED") in a monthly report, which is generated with about a 35 day delay (e.g., the September total miles driven is published in early November). The information is useful because it is an indicator of:

  • Gas Demand
  • Insurance Claims (all else being equal, the more people drive, the more accidents they will have, and the worse the insurance companies will perform, as the typically charge a fixed fee for insurance, independent of mileage)

There are two versions of Miles Driven published by FRED: seasonally adjusted, and unadjusted. We worked with the unadjusted, because it is really hard to replicate the seasonal adjustments (the government does not provide detailed steps of how they perform the adjustments).

The results are as close as it gets! The correlation between Advan's Miles Driven to the Federal Government's is 0.92. But Advan's data is available T+1 (i.e., with 1 day delay), so more than a full month ahead of FRED's. Talk about having an unfair trading advantage!

But we did not stop there. The beautiful thing about the breadth of the data available, is that we can also figure out which cellphone devices belong to truck drivers. Or for that matter, which devices live in a given Cebsus Block Group or Zipcode. Or which ones have a certain income level; or which ones buy or service their cars in a Ford dealer, or a Mercedes dealer; and so on. The miles driven can then be computed only for these devices, to generate custom insights that not even the government, or anyone else for that matter, has access to today, but will be unable to live without in the future!

Here are Advan's Trucks miles driven sourced from (the devices belonging to the drivers of) about half the commercial (class 8) trucks in the US:

The possibilities are endless, so free up your imagination and let us know what's the next interesting insight you want us to compute!

November 08, 2019: Freedom: priceless

It comes in many forms: Speech. Movement. Draw within the lines. Wait, scratch that last one.

We didn’t like that last one either. When it comes to measuring foot traffic, trade areas, migration patterns, demographics, cross traffic, and in general get behavioral metrics from mobile location data, yesterday’s state of the art was to get canned results on some buildings or venues someone else has decided you might be interested in.

What’s the reason for the limitation? In two words: Big Data. When you are dealing with billions of data points you can’t just write a query to extract any information you want because it will be very slow and very costly (sometimes in the hundreds of thousands of dollars).

It’s even worse in our case because we have 6 trillion cellphone observations over the last 5 years! And growing at a rate of over half a trillion every quarter. Gulp.

But the ability to draw any area, be it a building, a retail location, a factory, an oil field, a neighborhood, you name it, and get instant foot traffic measurements for it was making us loose too much sleep. A sure sign we had to do something about it.

So after several months of head scratching, experimentation, optimization, design, testing, design again, and all the other sleep-reducing activities favored by software engineers, we built a tool to be able to do just that. If you thought we would have shown it to the whole world the very next day, you would be right, but... once a nerd always a nerd. It took us 2 full years to open it up to our clients. Well, better late than ever...

So finally last month we announced REveal, a self service website where the user gets full control of the data.

  • Draw it:

  • Recite your favorite poem, and voila! True Home Trade area:

  • Want to find out where those living in a building you are considering buying work?
  • And how has that changed over time?
  • What is the true trade area of this Mall?
  • Where do the visitors live and work?
  • Are more people moving into this allegedly hot neighborhood, or is it all hype?

And on and on. The possibilities are truly endless.

As recently as 10 years ago it would be unfathomable to think that you can crunch 2 Petabytes of data (that’s 2 Million Gigabytes!) in seconds and get detailed answers to any question you can think of. Progress is sometimes pretty close to science fiction!

- Spock: Over and out.

October 31, 2019: There's no Season like Earnings Season

It is everyone's favorite game to perfectly forecast earnings by looking in the rear view mirror. We are trying really hard not to fall into the same trap ourselves.

Here is why: it is very easy to make random predictions, some of which come true, and then look back and cherry-pick the ones that you were right. Really, a monkey can do it. If you give them a banana for each correct prediction they would get pretty fat, fast!

What really distinguishes a good forecasting method from random chance is, the percentage of times you get it right versus the ones you get it wrong. If you are correct more often than not, even 51% correct vs 49% wrong, then you have something to say. Even if you are wrong more often but you have higher conviction (and therefore make more money) when you are correct, that also has value.

Advan's Machine Learning algorithms take the foot traffic data we compute and forecast top-line revenue. We get it right about 60% of the times. 2 out of 3. If that doesn't sound impressive, consider that many quantitative funds can build profitable algorithms from a mere 51% advantage. And in our own "paper trading" backtests, the performance of a long/short neutral portfolio constructed using Advan's foot traffic data has a Sharpe ratio over 2.

Considering we do not claim to be experts in either Portfolio Construction nor Machine Learning, these performance results are pretty good, if we may say so ourselves. The average Hedge Fund has Sharpe under 1 (not to ding Hedge Funds, actual trading is much harder than paper trading).

Having said that, we can't resist the urge to brag about individual hits. Just this once:

Texas Roadhouse (TXRH):

The consensus estimated revenue in Q3 2019 was $649.2mm. Advan forecasted $651.32 on foot traffic growth of 6.1%. The actual revenue reported on October 28th after the market close was $650.42mm. The stock closed at $50.17 on the 28th and traded up 20% the next day!

Advan's Year over Year reported traffic and TXRH top-line revenue have correlation of 0.7 over the last 8 fiscal quarters; Quarter over Quarter traffic and top line revenue have correlation 0.96. This correct forecast wasn't an accident.

October 22, 2019: Myths and Truths about Retailer performance

Is Gamestop's traffic up? Is Jimmy John's traffic growing more than Subway's?

Every day some new analysis of cellphone location data portrays to measure the exact foot traffic in one or all of these, and every day we emit a collective gasp at the incredulous claims.

Let's get this quickly out of the way: Gamestop traffic is trending down; Jimmy John is down too and that trend has not changed for 3 years straight; it's also worse than Subway's downward trend, except for some bright, but inconsistent, spots in 2019.


Jimmy John's (blue) vs Subway (pink):

But that's only the beginning of the story.

First, it is a disservice to the reader to portray that any dataset, and in particular cellphone location data, can estimate within a fraction of a percentage point the actual traffic of a company. With extremely detailed geofencing work, taking into account the hours of operation of every single location, and after testing hundreds of normalizations versus the actual revenue data and versus our partner's (Consumer Edge) credit & debit card transaction data, our research team at Advan can come close. We strive to be approximately right, instead of precisely wrong, as Warren Buffet said.

Second, the actual claims we have been hearing are completely off the mark. Gamestop's traffic for example -- and we have nothing against the company, these are the fundamentals talking -- is down. About 6% down year over year in Q2 2019 in fact, and not looking better in Q3. There are no two ways about it (if you insist on exact numbers, then 6.04% down, but remember, this is approximate).

Jimmy John's is doing better than Subway in 2019 (comparatively speaking; Subway has 13x the traffic), but that trend has started running out of steam in the first 2 weeks of October. Here's hoping it's just a small aberration. More worryingly though, if you go back to 2018 and 2017 the 2 chains' traffic is changing at the same rate, and that rate is downward trending in both cases. Not a bullish sign.

So please, do not believe everything you read without confirming how the analysis was performed. Consider placing more weight on analytics performed by the experts in crunching and normalizing location data for financial performance. Data is good; but incorrect and misleading data is worse than no data.

About Advan

Advan is the leader in the Big Data geolocation space, enabling participants in the financial industry to analyze foot traffic data across multiple sectors, including consumer services, energy, technology, healthcare, REITS, financials and others. Advan derives its datasets using multi parameter models that analyze cellphone location data crossed with curated geofenced areas.

Top tier institutional investors spanning from quantitative hedge funds to fundamental asset managers have been the main consumers of Advan’s products.