story

How much money will you earn (and spend) next week?

by Chuck Howard, Dale Griffin and David Hardisty
Posted on March 09, 2020

Setting a budget is commonly portrayed as a quick and easy fix for many of our most pressing financial problems. Need to reign in your spending? Set a budget! Want to get out of debt? Set a budget! Have to save more money? You get the idea.

But if setting a budget is so easy, why is it that consumer debt has reached record highs and consumer savings remain perilously low, even while most people are setting budgets? (US Federal Reserve 2020; Zhang & Sussman 2018). Maybe budgeting isn’t quite as easy as first thought.

The budgeting process

Budget setting requires predicting future income and future expenses. The accuracy of these predictions is important because budgets influence our spending (Howard & Lukas, 2019), and if our predictions are optimistically biased then we will end up spending more (and saving less) than we anticipated.

To test our hunch that budget setting can be a challenge, we began by studying consumers’ expense prediction accuracy. In a series of studies that includes data from a budgeting app in the UK, a nationally representative survey of US consumers, and a field study conducted with members of Canada’s largest credit union, we found that consumers display an expense prediction bias in which they persistently under-estimate their future spending. We also found that this bias occurs in large part because we tend to believe that our future spending will be fairly typical (e.g., groceries, gas, and rent), but almost all of us also encounter atypical expenses in any given week or month (e.g., dinner with our in-laws, a replacement part for our car, or a new dress for our child’s dance recital). Fortunately, this insight suggests a simple solution: prompting the Credit Union members in our field study to consider three reasons why their expenses might be different from a typical week completely eliminated the expense prediction bias.

Income prediction bias in the gig economy

We are now in the process of extending our work on budget setting for expenses by studying the accuracy of people’s income predictions. Our research in this regard is motivated by the rapid growth of the “gig economy,” which is defined by temporary, freelance work like driving for Uber. One important and previously unstudied aspect of gig economy employment is that it involves variable and perhaps unpredictable income. We therefore want to understand if gig economy workers mispredict their future earnings, and if so, in what direction and why. It could be the case that people with variable income under­-predict their future income, much like they under-predict their future spending. However, our hypothesis is that gig workers with variable income display an income prediction bias in which they over-predict their future gig income.

“Think about what an atypical week looks like in your own life. Is it more or less busy than usual?”

To understand the logic behind our hypothesis, consider the case of a part-time Uber driver Dave. In a typical week, Dave makes $300 from Uber, and he therefore includes $300 per week from Uber in his household budget. Given that Dave’s Uber income will fluctuate from week to week, how accurate will his budgeted income be at the end of the month? To answer that question, think about what an atypical week looks like in your own life. Is it more or less busy than usual? If you were an Uber driver like Dave, would an atypical week leave you more or less time to drive? Our prediction is that Dave’s atypical weeks most often leave him less time to drive than usual because the events that define these weeks – a car repair, staying home with a sick child, or attending a cousin’s wedding – tend to reduce time for work rather than create more. Therefore, if Dave bases his Uber income prediction on his income from a typical week he will end up earning less than he predicts over time.

Results thus far

To date, we have run four studies to test our hypothesis that gig economy workers over-predict their future income because their predictions do not account for atypical events. In Study 1 we asked 157 Amazon Mechanical Turk (AMT) workers to predict their AMT income for the next week, and to tell us what thoughts came to mind while they were making their prediction. Later in the survey, we asked them to report their typical weekly income from AMT. Consistent with our hypothesis, predicted income and typical income differed by less than $1 on average. Moreover, we found that 65.3% of participants referenced typical outcomes when making their prediction.

In Studies 2 and 3 we measure income prediction accuracy. To accomplish this, we’ve been recruiting Uber drivers and AMT workers online, asking them to predict their income at the start of the week, then checking in with them at the end of the week to see how much they actually earned. These studies are still in progress, but the preliminary results support our hypothesis that gig economy workers tend to over-predict their future gig income, and that this bias is both statistically significant and economically meaningful: AMT workers have over-predicted their weekly income by an average of 11.6%, and Uber drivers have over-predicted by an average of 60.2%.

“How can we help people improve their income prediction accuracy? ”

Above and beyond understanding the magnitude of the income prediction bias and why it occurs, a key goal of our research is to understand how to help people improve their income prediction accuracy. To make sure we are on the right track, we ran the following preparatory study with 260 AMT workers.

Participants were randomly assigned to predict their AMT income in one of four conditions. Participants in the control condition were asked to spend some time considering their AMT income for the next week, and then to make their prediction. Participants in the intervention conditions received prompts to consider different types of atypical outcomes. Finally, as in Study 1, participants were asked to report their typical weekly income from AMT later in the survey. Our expectation was that participants in the control condition would predict future income equal to their typical income (replicating Study 1), but that participants in the intervention conditions would predict future income lower than their typical income. Consistent with this expectation, predicted income in the control condition differed from typical income by only 0.7%, whereas predicted income in the intervention conditions was 8.2% - 12.8% lower than typical income.

Next steps

The studies discussed above provide preliminary evidence that gig economy workers with variable income tend to over-predict their future income because they fail to consider atypical outcomes when making their predictions. The key limitations of these studies are that they include only two types of gig worker, they only examine prediction accuracy for one week, and they do not measure potential downstream consequences of over-prediction such as over-spending or under-saving. To overcome these limitations we have partnered with a personal finance app to run a large-scale longitudinal field study that will measure income prediction accuracy in several different types of gigs over several weeks, and allow us to measure the costs of over-predicting income and the benefits of improving prediction accuracy. Ultimately, our goal is to provide guidance so that online banks and financial apps can develop tools to help consumers budget more accurately and hence keep their finances under control.