Cities are physical hardware that withstand overlapped dynamics, resulting into very complex systems. Any change in one of its boundary conditions produces positive and negative effects. Traditionally, urbanists have tried to measure those effects -mainly through surveys- in order to get information whether the management decisions were producing the desired results. Digitization has brought two major innovations for city managers: the opportunity to open new participative channels, and the possibility of gathering massive amounts of evidences to support decision-making.
At BBVA Data & Analytics we believe that this increasingly complex world needs systemic approaches to measure consequences of any events. We are particularly obsessed with orienting debates from subjective opinions to objective evidences. Our research in that domain led us to the development of methods and techniques that provide insight into the commercial activity of a region from the descriptive capacity of financial data (e.g. bankcard transactions). We are particularly good at measuring the economic impact of exceptional circumstances being natural disasters or large events.
In this article we describe how we measure and characterize the impact of urban management decisions on the city space, based on the commercial activity digital footprint. We explain how we measure the expected commercial activity of an area and compare it to the reality. We believe that local authorities can use this type of observations to optimize positive effects and mitigate the negative ones in the decision making process.
This is Gran Via (literally “Great Way”) the upscale shopping street also known for its nightlife located in central Madrid, Spain.
The area suffers from traffic congestion and cars occupy a space that pedestrians are claiming back. Like other cities around the world, Madrid is evaluating sustainable solutions that mitigate the density of car traffic while promoting economic activity.
Last Christmas season, the city council adopted traffic calming measures in Gran Via (one of the busiest streets of the city) in benefit of pedestrians. During a few weeks until early January, the access was restricted to one lane in each direction and a speed limit of 30 km/h was enforced. The measure aimed at encouraging smoother pedestrians flows and indirectly at improving the shopping experience. It was also the test for a more global strategy to reduce the negative effects of car congestion in the center. Unsurprisingly, that traffic calming strategies provoked heated debates in the media about its economic implications.
As part of our Urban Analytics initiatives, we analysed approximately 7,000 daily bank card transactions to extract facts on expenditures in the Gran Via area.
What was expected: the prediction model
First, we inferred what would have happened during Christmas period without traffic restrictions. Practically, we used Bayesian Structural series models previously used to measure the economic impact of Hurricane Odile that hit the Baja California Sur peninsula in Mexico in 2014. In that case study, the model predicted the level of expenditure expected if the hurricane had not hit the area and we were able to measure the time necessary to recover normal activity.
Applied to the Gran Vía traffic calming strategy, we were able to compare the commercial behavior of the area before the restrictions with the commercial behavior of similar areas (in terms of time series correlation) where no action was taken.
Our approach took a few steps:
- The time series of expenditure is calculated for the Gran Via area, and for all the postal codes in Spain, from January 2014 until today.
- We identify the postal codes with a high correlation in the last three years in their temporary patterns of consumption with respect to the Gran Via area. For each area we extracted temporal sequence of 1,000 daily points. Eventually, 200 areas were selected to define the expected activity model.
- From a minimum correlation threshold, N times series corresponding to N postal codes are selected to train the model that defines the time series that best fits to the pattern of commercial activity of Gran Via.
- The resulting model defines the total expected expenditure in the Gran Via area in ‘normal’ conditions, by analogy to what happened in the N postal codes with a similar behavior.
Then, we compared the expected expenditure in the Gran Via (according to the prediction model) with the expenditure actually recorded during the Christmas period of 2016, measuring the difference between the two time series.
Measuring the impact: comparing the prediction model with reality
There were traffic restrictions in three periods. The first period, from December 2nd to 11th registered a small increase of +1% of spendings. The second, between December 16th and 18th, obtained +2% variation of spendings. In contrast, the last period, from December 23rd to January 8th, recorded a decrease of -13% in spendings.
Last Christmas, we observed an overall decrease of spendings of -8% in comparison with the expected activity in Gran Via.
Typically, any economic impact needs to be characterized. For instance, we analyzed the evolution of three main commercial categories of the area, namely fashion, restaurants and hotels. The results helps highlight the impact more specifically:
- We observed a global decrease of 17% in spendings in fashion -the dominant activity by income- over the three periods.
- However, the restaurants -the dominant activity by number of businesses- increased their sales by +5% over the two first periods.
- The hotel category also obtained positive variations during the three periods, with an increase of +11% in spendings.
As a brief overlook, when we compared the real activity of the Gran Via area during the Christmas Season with the expected activity of the prediction model, we obtained the following measures of economic impact:
Although in absolute terms, the commercial activity registered in the Christmas Season of 2015 and 2016 reached similar levels, the turnover of the last Christmas was below the expected according to the statistical model used. Indeed, the same model applied to the whole city does not detect the decrease that is registered for the Gran Via area.
The practice of Urban Analytics provides this type of insights on the commercial pulse of a region from the descriptive capacity of financial data (e.g. bankcard transactions. The data source we work with -anonymized digital economic footprints- allow us to describe both: short wave effects (events that, during specific hours or days, have temporary effects in the normal pace of an area), and long wave effects (slow changes in the profile and behaviour of the citizens and visitors, or trends that transform the commercial fabric of a neighbourhood, changing its character). We believe it is our obligation to open and share this knowledge with local authorities that must take complex decisions and with society to adapt to changing circumstances.