Zindi and Rand Merchant Bank (RMB) joined forces in a 9-month endeavor known as the RMB CPI Nowcasting Grand Challenge. The objective was to leverage alternative datasets for predicting CPI one month ahead, a critical need for financial institutions to guide lending and investment decisions. The competition garnered participation from over 1500 data scientists across 86 countries, showcasing global interest and talent in the field, and generated valuable models and brand awareness for RMB.
Problem:
Accurately predicting short-term CPI fluctuations poses a significant challenge for financial institutions. Traditional methods can fall short in capturing the complexities of economic dynamics, and a crowd-solved data science challenge is a novel way for organisations to explore alternative approaches, while also exploring the data talent available in diverse data science communities.
Solution:
The winners were Tom Wetherell (1st place overall), Arno Pienaar (2nd place overall), Puja Pande (top-performing South African), and MG Ferreira (most innovative model). Contestants drew on diverse alternative datasets and pioneering machine learning techniques to nowcast CPI one month into the future, and the challenge provided a platform for participants to benchmark their models as well as discuss their approaches. Winning strategies ranged from incorporating petrol price and exchange rate data to leveraging satellite imagery and historical CPI trends for modeling.
“I really enjoyed the RMB challenge. I recently completed a big CPI model for a client, and this was a good way to benchmark that model, with room to apply the latest techniques,” explained MG Ferreira. “As always, the competition itself and the interaction with other competitors is amazing, I always learn a lot from that. What makes this competition maybe a bit special compared to others on Zindi, is the large amount of data. I could pick and choose which data to use from many sources — it was truly a big data challenge.”
Impact:
For RMB, the challenge represented a chance to look at new ways of predicting CPI, and a way to discover and connect with the deep pools of data science talent in the Zindi community. The models developed during this competition are already being implemented at RMB to supplement their existing CPI predictions at the bank.
“Our collaboration with Zindi was successful for RMB in several ways,” said Jon Cornfield, Head of Markets Trading and Execution at RMB. “We’re already testing the winning models against our in-house CPI predictions and we expect to use these models going forward. We appreciated the hybrid approaches and innovative data sources used by the winners, and we look forward to the competitive edge these models will lend us.”
“Secondly, in terms of talent, we were delighted with the breadth and depth of data talent this collaboration with Zindi unlocked — from the many universities and students involved, to the skills demonstrated by Zindi users all over the world, this was a great opportunity to position RMB as an employer of choice in the data and AI space. We’re looking forward to further interactions with this newly unlocked community of practice.”
The RMB CPI Nowcasting Grand Challenge demonstrated the power of collaboration between the financial sector and Zindi’s data community. By harnessing innovation and expertise, the competition not only advanced RMB’s CPI prediction capabilities but also stimulated a vibrant ecosystem of knowledge sharing and talent development. Moving forward, the winning models promise to enhance RMB’s predictive capabilities, underscoring the value of Zindi’s community in driving innovation and excellence in AI.
See more:
www.rmb.co.za/rmb-nowcasting-challenge
zindi.africa/competitions/rmb-cpi-nowcasting-challenge
Source : https://zindi.medium.com/zindi-and-rmb-reinvent-cpi-nowcasting-with-community-first-machine-learning-c0d6acf21e5d
