Coincident Indicator for the Colombian Economy
Financial markets are very sensitive to macroeconomic data published on a daily basis by official and non-official institutions. For a developing financial market like the Colombian, a few activity indicators have a stronger impact on the fluctuations of asset prices, and therefore, financial agents are always looking for the best predictive models to anticipate the movements of GDP growth, inflation, and employment. Due to the nature of such indicators, DANE (the National Department of Statistics) publishes its several activity indicators with some lag. Quarterly GDP growth, for instance, has a publication lag of two months. In this project we propose a coincident activity indicator that mimics the evolution of the ISE (economic activity indicator), published monthly by DANE (with a lag of 2.5 months). We construct the inputs for this indicator by efficiently processing and leveraging the large amount of transactional data that our customers generate on a daily basis (credit card payments, utilities payments, POS purchases, credit payments, transfers among investment funds, etc.). Based on the economic activity of the merchants and firms involved, we are able to nowcast, at the industry level, the activity for the whole economy on a monthly basis. The modeling approach uses a Bayesian Model Selection framework for selecting the best predictive models. Live testing has shown, so far, an excellent fit for the global model.
Carlos Patino currently works at the largest Colombian retail bank, Bancolombia, as Analytic Capabilities Manager. This team operates transversally to all operational and strategic areas of the bank, and collaborates with existing analytics and data science teams in the adoption of Big Data technologies, and Machine Learning techniques, aimed at solving business problems, and at the creation of data and ML-based products that benefit the bank’s 9 million clients. In his role, Carlos develops ML models using Spark, Python and other tools such as CDH. Prior to joining Bancolombia, Carlos spent three years working as an Advanced Modeler for the Customer Strategy Management group at PNC Bank in Pittsburgh, PA. Carlos holds a Masters in Public Policy and Management from Carnegie Mellon University, and a Bachelor in Economics from ICESI University in Cali, Colombia.