Forecast reconciliation with clustering structure: application to stock prices


Raffaele Mattera, University of Rome, Italy


This paper discusses the use of forecast reconciliation for time series characterized by unknown hierarchies. A relevant example is the stock market, where common stocks aggregate to the market index in accordance with a hierarchical structure that needs to be uncovered. To apply reconciliation to the financial domain, a novel forecasting framework combining forecast reconciliation and clustering is proposed. The application to Dow Jones Index highlights that better forecasts of both the index (top-level series) and the individual stock prices (bottom-level series) can be achieved with this new approach. A discussion about possible future research on the relationship between reconciliation and clustering as well as their application to financial forecasting is also provided.