¶ Consumer spending during unemployment: Positive and normative implications

AEA link

¶ BibTeX

@article{ganong2019consumer,
  title={Consumer spending during unemployment: Positive and normative implications},
  author={Ganong, Peter and Noel, Pascal},
  journal={American economic review},
  volume={109},
  number={7},
  pages={2383--2424},
  year={2019},
  publisher={American Economic Association 2014 Broadway, Suite 305, Nashville, TN 37203}
}

¶ Abstract

Using de-identified bank account data, we show that spending drops sharply at the large and predictable decrease in income arising from the exhaustion of unemployment insurance (UI) benefits. We use the high-frequency response to a predictable income decline as a new test to distinguish between alternative consumption models. The sensitivity of spending to income we document is inconsistent with rational models of liquidity-constrained households, but is consistent with behavioral models with present-biased or myopic households. Depressed spending after exhaustion also implies that the consumption-smoothing gains from extending UI benefits are four times larger than from raising UI benefit levels.

¶ Notes / Excerpts

Potential teaching note:

The behavior we document is inconsistent with theories based on benchmark rational models of forward-looking but liquidity-constrained households because liquidity constraints are unable to explain why households fail to save in anticipation of predictable income declines. However, the path of spending we document is consistent with models where some households exhibit present-biased or myopic behavior (Laibson 1997, Campbell and Mankiw 1989).

Interestingly, the myopia seems to extend not just to benefit exhaustion but benefit arrival, too.

The most remarkable evidence of excess sensitivity comes from New Jersey, which has unusually generous UI benefits. UI payments there begin quickly, such that many workers receive their last paycheck and their first UI check in the same week. This induces a sawtooth pattern in average income, which falls as some workers become unemployed, then rises by 10 percent in the month with the extra check, and then falls to a stable level thereafter. Spending around onset follows the same sawtooth pattern: falls, rises, and then falls to a stable level, suggesting that much of the extra check is spent immediately, even though households know their income is likely to fall sharply in the following month.

That’s surprisingly strong evidence against consumption smoothing, unless these households just have low $\beta$? Oh, yeah, that’s their myopia story actually.

. At exhaustion, we estimate that about one-third of agents are the low β types.

reproduce the “spender-saver” behavior in Campbell and Mankiw (1989), albeit with a micro-foundation relying on heterogeneity in β

Oh, actually, $\beta$ is their one-month-ahead discount, and $\delta$ is the time-consistent exponential discount factor. So in the intro, they say variance in delta can also match the data with a slightly worse goodness of fit. So basically, they think the best story is something like short-term impulse control, rather than generalized time discounting.

“Standard model” has monthly exponential discount 0.9898 with no myopia, so their $\beta=1$. With variance in beta, they estimate exponential discount to be .9951, with two groups with $\beta=0.522,0.899$.

(their model suggests most households have only a small spending drop at exhaustion, and only a small share are very myopic. But volatility of data means they can’t actually test that against the alt hypothesis that most households slightly myopic.)

Response to income is stronger than expected, but increases to income can’t help us distinguish between hypotheses:

• rational expectations with liquidity constraints (Kaplan and Violante 2014, Carroll et al. 2017)
• behavioral econ, self-control problems (Laibson 1997)
• people following rule of thumb (Campbell and Mankiw 1989)

The low beta people mean that duration of benefits is important to consider along with level of benefits.

We find that the welfare gains from improved consumption-smoothing due to extending the duration of UI benefits are four times as large as from raising the level of UI benefits in a generalized Baily-Chetty formula.

They cite (Schmieder and von Wachter 2017, Kekre 2017, Kolsrud et al. 2018) for literature on optimal path of benefits.

Florida UI lasts only four months.

Also, they find a spike in job finding at benefit exhaustion, but other papers have documented similar.

With regards to their comments on credit cards:

Second, it would be interesting to further explore our finding that households do not seem to borrow much during unemployment. For example, households only borrow an average of about $20 per month on Chase credit cards during unemployment, despite having large unused credit lines. Because unemployment is a mostly temporary shock to income, the rational buffer stock model predicts a large increase in credit card utilization during unemployment. The absence of credit card borrowing that we observe among unemployed households is particularly striking against the backdrop of widespread credit card borrowing by US households overall (Laibson, Repetto, and Tobacman 2007).

I wonder if that’s just a measurement issue? Like, people just aren’t using credit cards from Chase? My intuition from experience is that the people who use credit cards extensively in that way (rather than just as a security buffer for transactions or for rewards bonuses) often have multiple credit cards from different institutions.