To the untrained eye, this spring saw the return of Americans eating out and shopping at their favorite restaurants and stores at nearly the same level as pre-COVID times.
According to Google mobility data, which tracks people's locations with Google apps on their cell phones, retail and recreation visits in April, as well as the summer months, were roughly 5% off from where they were in January of 2020. In other words, Americans were back to the dining and shopping routines they had before the pandemic.
But the D. E. Shaw group, the $60 billion quant giant and one of many users of Google's data, found through its own analysis and manipulation of the data that this isn't the case. The recovery "has been more muted than the raw data suggests," a new research paper from the firm reads, and the promising bounceback since April "has essentially been nonexistent."
Alternative data, like Google's geolocation insight or the many supply-chain-focused datasets, is a great tool for top investors, but does not automatically turn into blockbuster returns. Funds have to clean, analyze, and dissect the info for it to be usable, leading to arms race at many firms for top data-science talent.
D. E. Shaw, the secretive fund started by computer scientist David Shaw, reveals in the report how it manipulates raw datasets like Google's geolocation info — offering a look inside how one of the smartest buy-side firms in the world makes its secret sauce.
"The proliferation of alternative data has run ahead of the ability to process all of it in a meaningful way," Kris Dawsey, a global markets economist in the D. E. Shaw group's discretionary macro trading unit, told Insider.
"This is not to say that these data sources are not high quality – they're reflecting what they are intended to reflect."
Adjusting for seasonality, without history
The analysis done on the Google mobility data was for the firm's discretionary macro team, which manages billions and is run by human traders, not machines.
The biggest adjustment the firm made to the mobility data was adding in seasonality filters, which show that the retail foot traffic of January 2020 is a low bar to compare to. Stores typically see a lull in activity in January following the holiday shopping season.
"January is consistently close to the low point of the year in non-seasonally-adjusted activity. In other words, in the absence of a pandemic, we would have expected mobility during the rest of 2020 on average to have been substantially higher than during the GMD baseline period," the report reads, using an acronym for Google mobility data.
Taking into account seasonality fluctuations is not novel, but the Google data D. E. Shaw was manipulating did not have enough of a history to determine the impact of changing weather on consumers' behavior. So the firm built out its own seasonality factor to compare temperature levels with state data on shopping and dining.
D. E. Shaw calculated an increase in temperature of 10 degrees is equivalent to just under a 3% bump in retail visits without any other extenuating factors.
The additional data analysis provided context for the firm to better understand the Google data. The result is a recovery that's not nearly as promising as what the data first displayed.
"Our seasonally-adjusted measure suggests that mobility is still down substantially (about 15%) relative to the pre-COVID baseline, presenting a more pessimistic picture of overall normalization that has taken place," the report reads.
