Thomas Chaney's discussion (TSE) of Xavier Gabaix's presentation "Behavioral Macroeconomics Via Sparse Dynamic Programming" (NYU Stern), at the Banque de France and Sciences Po joint workshop on Granularity of Macroeconomics Fluctuations, 24 June 2016. Slides of presentations & discussions are available online: https://www.banque-france.fr/en/economics-statistics/research/seminars-and-symposiums/research-workshop-on-the-granularity-of-macroeconomic-fluctuations-where-do-we-stand.html
Banque de France's Workshop on Granularity: Thomas Chaney's discussion, June 2016
1. Xavier Gabaix:
Behavioral Macro Via
Sparse Dynamic Programming
Discussion by Thomas Chaney
Toulouse School of Economics
Banque de France, June, 2016
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 1 / 11
2. Some perspective
Lucas JET 1972:
1 confronted large unexplained puzzles (Philips curve)
2 confronted lack of micro-foundations for dynamic macro
3 offered novel concept (rational expectation equilibrium)
4 offered tractable tools (dynamic programming)
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 2 / 11
3. Some perspective
Lucas JET 1972:
1 confronted large unexplained puzzles (Philips curve)
2 confronted lack of micro-foundations for dynamic macro
3 offered novel concept (rational expectation equilibrium)
4 offered tractable tools (dynamic programming)
Gabaix 2014/16:
1 confronts a series of (smaller?) puzzles
2 confronts unease with macro models (full rationality)
3 offers novel concept (sparse bounded rationality)
4 offers tractable tools (sparse dynamic programming)
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 2 / 11
4. Roadmap
1 Representative agent/heterogeneous agents.
2 Utility accounting.
3 Setting the “default” model.
4 Small comments.
5 A network application.
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 3 / 11
5. 1- Representative agent?
Unlikely a “representative agent” will be sparse BR.
What about the aggregation of sparse BR agents?
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 4 / 11
6. 1- Representative agent?
Unlikely a “representative agent” will be sparse BR.
What about the aggregation of sparse BR agents?
You can make progress:
1 Your model: which agent drops which state variable.
2 ,! extensive margin of attention.
3 Composition effect from this extensive margin.
4 Tractability should allow to deal with heterogeneity?
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 4 / 11
7. 1- Representative agent? (cont’d)
De-coupling of dimensions?
1 Narrow framing or not? (footnote 51)
2 If I buy a house, do I become aware of the interest rate for other
decisions (e.g. consumption-saving)?
3 Ultimately, an empirical question.
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 5 / 11
8. 1- Representative agent? (cont’d)
Lucas critique:
1 Lucas: what looks like money illusion (Philips curve).
2 Gabaix: sort of the same.
3 Smooth version of Lucas: big/small shocks, aware/unaware agents.
4 Gabaix is empirically relevant version of Lucas.
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 6 / 11
9. 2- Utility accounting
Importance of mental cost for welfare/policy recommendations.
Cost, kg (m), does not appear in preferences (only through A).
Does accounting for mental cost affect time [in]consistency?
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 7 / 11
10. 3- Setting the default model
Very upfront about the arbitrariness of choosing a default model.
But you can say more: how changing the default model affects
behavior!
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 8 / 11
11. 3- Setting the default model
Very upfront about the arbitrariness of choosing a default model.
But you can say more: how changing the default model affects
behavior!
Example:
1 Tax code tutorials make it the default model (Chetty et al.)
2 Changes in the tax code (with tutorials) change the default model.
3 Gabaix tells us how other shocks interact with the tax code.
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 8 / 11
12. Smaller comment I: cross partials
Agent weighs utility gain against cognitive cost one state variable at a
time.
What about two variables at once? any combination of them?
Question: Could it be the agent misses out on cross-partial terms?
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 9 / 11
13. Smaller comment II: outsourcing complexity
Well defined utility cost of contemplating one state variable (kg (mi )).
Well defined benefit as well (could be expressed in income-equivalent).
Question: Could complexity be outsourced?
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 10 / 11
14. Smaller comment II: outsourcing complexity
Well defined utility cost of contemplating one state variable (kg (mi )).
Well defined benefit as well (could be expressed in income-equivalent).
Question: Could complexity be outsourced?
Example:
1 e.g. g (m) = ma with a = 0 (fixed cost).
2 Economies of scale for a service provider (e.g. if similar clients).
3 Model makes predictions re: when outsourcing is more likely.
4 Refinement: An intermediary can easily cheat a sparse BR agent.
5 Should it be regulated?
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 10 / 11
15. Application: Oligopoly in an input-output network
Thomas Chaney (Toulouse) Sparse Dynamics Programming BdF, June 2016 11 / 11