Understanding the Expense Prediction Bias

Download data and study materials from OSF

Principal investigators:

Chuck Howard

The University of British Columbia

Email: chuck.howard@sauder.ubc.ca

David Hardisty

University of British Columbia

Email: david.hardisty@sauder.ubc.ca

Homepage: http://davidhardisty.info/

Abigail Sussman

University of Chicago

Email: Abigail.Sussman@chicagobooth.edu

Homepage: https://www.chicagobooth.edu/faculty/directory/s/abigail-sussman

Melissa Knoll

Consumer Financial Protection Bureau

Homepage: https://www.consumerfinance.gov/data-research/cfpb-researchers/melissa-knoll/

Sample size: 1288

Field period: 03/05/2016-06/17/2016


Past research suggests that consumers tend to under-predict their future expenses, a phenomenon we’ve labeled the expense prediction bias. Building on research in cognitive psychology showing that prospection and retrospection differ in terms of content and experience, we theorize and demonstrate that the expense prediction bias is driven in part by temporal asymmetry: a general tendency to mentally represent the future as more typical than the past.


Hypothesis/Research Questions:

Increasing perceived atypicality of future expenses will increase expense predictions.

Experimental Manipulations

Participants were randomly assigned to one of three conditions. In the control condition they were asked to recall and predict their expenses for the past and next week. In the typical condition they were asked to list three reasons why their expenses would be similar to a typical week before making their prediction. In the atypical condition they were asked to list three reasons why their expenses would the different from a typical week before making their prediction.

Outcome Variables:

Expense prediction

Number of expenses listed

Amount of each expense listed

Financial slack

Willingness to spend on an optional expense

Willingness to pay for a loan

Intention to save vs. repay debt vs. spend

Summary of Results

Expense predictions in the atypical condition were significantly higher than in the control and typical conditions, providing support for our primary hypothesis. Our exploration of downstream consequences of the bias did not reveal any notable results.