When indicators stop working

When indicators stop working


At a recent event, I joined with a group of experts in matters such as politics, economics, the environment and public health in Latin America. We were debating the general outlook in Latin America. Particular attention was given to recent changes in Peru and Colombia. We began to talk about the recent presidential polls and violent demonstrations in Colombia and how that could be affecting consumption. We concluded that when looking at consumer confidence indicators and indices, they did not correlate with actual consumption data. In particular, the ICC – the Consumer Confidence Index, developed by the National Department of Statistics from Colombia – stopped indicating the actual intention to consume, and therefore lost its usefulness. The reason? As consumers were so used to living in uncertain circumstances, they wouldn´t change their consumption behavior even under extreme conditions.  

But even if that assumption was true, I wanted to validate the hypothesis mathematically: the ICC index wasn’t valid anymore and therefore there was no correlation between its outputs and real market data. For this purpose, I used the Pearson correlation coefficient, comparing monthly smartphone shipments to Colombia (Canalys data) against monthly results from the ICC index. Just as a clarification, Pearson values (r) move between +1 and -1, so if r=0 there is no correlation at all, while r=+1 means there is a perfect or direct correlation and r=-1 indicates that there is a perfect inverse correlation. Values over +0.8 and under -0.8 indicate significant correlation, direct and inverse respectively. Once I compared all monthly values, I found that r was equal to 0, which meant there was no correlation at all between the above-mentioned variables.

But when looking at the above graph, apart from realizing it looks like a bow tie, I started wondering if I was comparing the data the right way, as smartphone shipments tend to anticipate consumer behavior at least a month in advance. For that reason, I moved the shipment data one month later, and eureka! The new result was r=1, hence there was a perfect direct correlation between ICC monthly results and Canalys smartphone sell-in data! The first conclusion I drew was that there is a correlation between the ICC index and real market data. The second conclusion was that smartphone shipment data shows great potential to predict consumer behavior a month in advance (given supply conditions are relatively stable), even during uncertain times.

It is possible that this correlation would have been different had I taken as a reference a much longer period (preferably 10 months or more). But, in the current circumstances, I didn’t find it possible to compare today’s consumers to the those of previous years. Our needs and priorities have significantly changed because of the pandemic and lockdown, so the data did too. It is certainly interesting to explore the usefulness and applications of sell-in data as we start to monitor how it can provide more insights into the future.