Read on Twitter
Here’s a thread about my first paper (I discovered today this @AEAjournals tweet!). The topic is the estimation of the elasticity of consumption to transitory shocks. https://twitter.com/AEAjournals/status/1335968447705075713
In short, I develop a robust version of the Blundell, Pistaferri and Preston (2008) estimator of this elasticity that lets consumption depart from a random walk. With the same data, I find the elasticity to be large, significant, and in line with natural experiments findings.
With more details: why is it a subject? The problem, to measure the impact of a transitory shock on consumption, is that transitory shocks are not usually directly observed (in survey data, people report their overall income, not saying what is transitory, what is permanent).
There have been two main ways to get around that. One is to use natural experiments of transitory shocks (e.g. tax rebate, lottery gain). The other is to put more structure on the survey data and derive restrictions that identify separately the effect of transitory shocks (BPP).
Now, the two methods yield very different results: most natural experiments find significant and large responses of consumption while the BPP estimator implies the elasticity is not significant and small.
Note that both results seem to have been influential at the same time, in different subfields. The natural experiment findings are at the root of the search for macro models capable of generating large average MPCs (e.g. Kaplan and Violante 2014).
The BPP result justified setting the wealth effect of transitory shocks to zero in later structural estimations (e.g. Blundell, Pistaferri and Saporta-Eksten 2016).
(To add to the mystery of the difference across methods, a paper by Kaplan and Violante (2010) implements *a version of BPP* in simulated data and finds the estimator of the elasticity to transitory shocks to be very robust.)
What I do is that I develop an estimator that relaxes the BPP assumption that log-consumption evolves as a random walk. Indeed, I show that this assumption is not necessary and can lead to a downward bias.
When log-consumption is a random walk and transitory shocks are persistent, there are two instruments that can be used to identify the elasticity to transitory shocks, but one of the two depends on both the current and the past values of the transitory shocks.
But if log-consumption is not a RW, this extra instrument is endogenous because past shocks do affect current log-consumption growth, most likely negatively (having received a good shock in the past, households make less precautionary saving, their consumption growth is flatter).
This seems to be the case in the data! When I drop the possibly endogenous instrument and implement an estimator based only on the robust instrument, I get a large average elasticity. It shifts from 0.05 to 0.60. An *elasticity* of 0.60 implies a lower bound on the *MPC* of 0.32.
Now, why did Kaplan and Violante (2010) found the BPP estimator to be robust? It is because they use, not the original BPP, but a version in which the transitory shocks are not persistent, in which case there is only one instrument and it coincides with my robust estimator.
Finally, I show with numerical simulations that a standard life-cycle model with incomplete markets can produce an average elasticity of 0.60. It is not the most novel thing: Kaplan and Violante (2014) had already generated high MPCs with a two-asset model (liquid and illiquid).
I do it with a shortcut, assuming away the illiquid assets, and broadly matching the distribution of wealth in the simulations to the distribution of liquid assets in the population.
And here we are. All my thanks to the AEA and editor! I'm happy to answer any questions!
