Forecasts of Doctor Visits for Flu: Simple Conservative Methods Beat Google's Big Data Machine Learning Model

Abstract

Purpose: Katsikopoulos et al. (2021) found that the recency heuristic's forecasts of week-ahead percentage of doctor visits associated with influenza symptoms-reduced forecast errors by nearly one-half compared to Google Flu Trends' machine learning model. This research note examined whether the accuracy of forecasts could be further improved by using another simple forecasting method (Green & Armstrong, 2015) that takes account of the observation that infection rates can trend, and did so in a conservative way (Armstrong, Green, and Graefe, 2015) by damping recent trends toward zero.

Methods: Out-of-sample forecasts from alternative models were compared using multiple error measures.

Findings: Damped trend models reduced absolute forecast errors by 13% relative to forecasts from the recency heuristic, by nearly 12% compared to forecasts from linear regression models, and by roughly 54% relative to Google Flu Trends forecasts.

Limitations: Rolling estimation of damping factors, multiple step-ahead forecasts, and forecasts of time series with many zero observations such as for smaller US states were not tested.

Implications:The findings provide further support for the superiority of the simple and conservative no-change and damped-trend (or combined) models to machine learning models for making time-series forecasts for complex uncertain situations.

Note:
Funding Information: I received no funding other than my university salary.

Declaration of Interests: No conflict of interest.

Keywords: complexity, forecast error, predictive validity, uncertianty, simplicity

JEL Classification: C1, C5, I1, Z18