Although governments usually lag behind businesses in adopting new management techniques, many are starting to pay attention to the value of big data. Of course this may be because analytics has had a greater impact on the field of microeconomics and policy makers have traditionally relied more on macroeconomic analysis. We have been a leading company in applying big data to macroeconomic analysis, and we have published a number of papers on Mexican tourism, social mobility, and an awad-winning paper on new techniques for predicting macroeconomic indices. In these papers, we have demonstrated that bank card transaction data is a resource that can provide insight into economic activity in a more timely fashion and with greater granularity than traditional methods.
The Impact of a Natural Disaster
To apply this data source to yet another new area, we collaborated with United Nations Global Pulse, an innovation initiative of the United Nations Secretary-General, with the mission to accelerate adoption of big data for humanitarian action. A case study was designed to determine what descriptive statistics we could derive from bank card transactions that could measure the recovery time of an area hit by a natural disaster. For this, the team studied behavior in Baja California Sur (BCS) in Mexico when it was hit by hurricane Odile in September 2014. The damage attributed to hurricane Odile included 11 dead, 135 injured, thousands left homeless and considerable infrastructure damage totaling approximately US$1 billion.
The purpose of this analysis was to provide the raw input needed for more informed policy decisions for future events. Once we know what data is relevant, with this means of data collection we can gather the information in near real-time as natural disasters develop, improving our ability to respond rapidly.
A challenge to most economic studies is that they are not laboratory experiments that you can rerun, nor are they even clinical trials with a control group where you can study what would have happened to some individuals if there had not been a hurricane. Using Bayesian time-series models, analytics has come a long way in addressing this problem with synthetic controls. We used a R library “Causal Impact” to estimate what would have been the economic activity in BCS if the hurricane had not occurred and compared that to the actual activity to determine the amount of time needed for an affected region to return to a baseline. Google developed this library with the purpose of measuring the impact of ad campaigns: the number of clicks counted vs. the counterfactual number of clicks there would have been without a campaign. But since its publication, the library has found wide use in social statistics. With it, we built a normality model using both BCS’s past economic patterns and data from states with similar patterns.
We also subdivided the population to be studied into 3 income groupings, by gender, and by geography. Since geography in the raw data was subdivided into 341 postal codes with just 131 of those producing meaningful data, the geographic areas were grouped to be more manageable into 8 clusters using the analytics k-means algorithm.
Much of the behavior observed is what one would expect during an emergency, although now we can be much more precise about the magnitude of changes in behavior. The study was able to identify which areas were hit the hardest by the drop in economic activity and the relative time it took to recover. Some of the obvious effects that were quantified were that right before the hurricane spending shifted to buying food and gas, poorer income groups prepared less, electronic transactions plummeted during the hurricane due to widespread power outages, and in the aftermath the mix of transactions switched to cash withdrawals. Less obvious insights were that women engaged in twice the number of preparedness transactions (perhaps this one was only less obvious to male observers), women’s spending took longer to return to the baseline, and lower income groups returned to normal patterns of spending faster than higher income groups. See our publication for more details.
Overall, the average recovery time was observed to be around 2 weeks. When broken down further by location, recovery times varied from 2 days to more than 1 month (for the towns located on the south coast where the hurricane struck with its highest intensity).
We believe policy makers should consider the implications of these observations. They will need to determine whether poorer citizens are not preparing because they do not have resources or they do not have access to the warnings about impending danger.
“This type of real-time quantitative data on how people prepare for disaster could be used to inform proactive, targeted distribution of supplies or cash transfers to the most vulnerable, at risk populations”, Miguel Luengo-Oroz, Chief Data Scientist at UN Global Pulse.
Revelations about a gender gap could indicate preparation measures should target women, as microcredit does, since they are engaging in stockpiling beforehand more than men. Real-time transaction monitoring in the aftermath could speed supply response and prevent price gouging. And of course insurance can more accurately estimate economic losses with this data.
As electronic payments migrate from cards to mobile devices, such as BBVA’s Wallet app, they will only become more ubiquitous and the accuracy of this type of analysis will improve. Even if businesses lead the way in finding new uses for this data, it won’t end there. This was the first study of how to use transaction data to improve humanitarian aid, and further work will allow highly targeted responses that were previously infeasible.
“With this project, we have created a replicable and evidence-based approach to understanding vulnerability. New insights can help authorities improve community resilience, which benefits the vulnerable and is also good for business continuity. Displacement by floods and earthquakes, hurricanes or fast-moving epidemics all represent a risk to economic sustainability”, Elena Alfaro, CEO of BBVA Data & Analytics
Find the whole story here