Skip to main content
Open Access Publications from the University of California

Judgmental Time Series Forecasting: A systematic analysis of graph format and trend type


In many areas like economics, finance, and health, people make judgmental forecasts looking at previous time series data. In such efforts, either tabular presentations or graphs are utilized, where graphs can be in different formats like bars, lines or points. Different presentations may cause certain biases stemming from bottom-up processing. To delineate such perceptually driven biases in judgmental forecasting, we investigated the effect of graph format (line, bar, point) and trend type (upwards, downwards, flat) on judgmental point forecasts when no domain information was provided. Bringing together perspectives from graph processing, visualization and forecasting literatures, our major goals were to determine which graph formats lead to more accurate forecasts and whether bar graphs lead to mean reversion bias or within-the-bar bias in forecasts. Additionally, we wanted to determine whether asymmetric damping observed in sales forecasts of downward vs. upward trended series were confounded by graph characteristics. We found that forecasts in line and point graphs were less biased than those in bar graphs; forecasts based on bar graphs depicting trended data exhibited mean reversion bias. We also observed a general positivity bias in forecasts for all trend types in line and point graphs. This implied trend following forecasts in upward trends and mean reverting forecasts in downward trends revealing an asymmetricity in the absence of context as well.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View