Student team: Sydney Son; Andrew Thvedt; Louisa Ong, Ariel Luo, Jen Woo
Challenge: Improve on previous modeling efforts to forecast COVID-19 infection rates based on daily time series data. Evaluate how to forecast daily COVID-19 time series data in one geography (e.g. New York) based on daily COVID-19 time series data from both that geography and other geographies (e.g. South Korea, Italy).
Problem Statement: Given the unprecedented, fast-moving, health & economic impacts of COVID-19, a more dynamic forecasting approach was needed to leverage fast-changing external data and adaptive predictive models to inform an organization’s financial outlook. The objective was to generate a solution that harvested daily external signals around virus and social policy impact across countries, along with economic data related to the impact on goods and services at multiple sector and geographic resolutions–taking in the latest data from countries experiencing impacts and combining this with the organization’s historical financial data to forecast potential “shocks.”
Envisioned Outcome: The team will deliver three elements:
- Summary deck of why the models used are recommended
- A prototype/proof-of-concept modeling using forecasting / machine learning algorithms; and
- Visualization of data through either a dashboard or chart(s) in Python notebook(s)
Data: John Hopkins University Covid-19 data, Oxford Policy data, US Census data, and Google Mobility data
Solution: Students developed short as well as long-term COVID-19 forecasting models using an epidemiological SEIRD (Susceptible Exposed Infected Recovered Deceased) modeling approach. The team creatively acknowledged the trajectory of COVID-19 varied not only by region but also within a state such that it needed to be accounted for in the model. They allowed for flexible, region-level customization that integrates non-traditional epidemic curves (given the nature of COVID-19_ that capture different waves). This project involved prescriptive and descriptive understanding of COVID-19 for the 50 biggest cities and 50 states across the U.S., a tailored COVID-19 forecasting model for each region, a mobility forecasting model, and an interactive website with visualizations that could enable businesses to easily understand the trajectory of this pandemic for their specific city or state. This proposed solution would enable organizations to quickly respond by providing the capability to flexibly and continuously adapt their financial forecasting models in order to provide timely guidance.