Bottom Up

# Overview

Solution, simulation, and estimation (e.g., by indirect inference) of dynamic stochastic optimization models of household choice is increasingly a standard tool in literatures ranging from the modeling of consumption, portfolio, and other financial decisions to labor economics (e.g., for studying optimal unemployment insurance design), public finance (optimal tax rates), health economics (investment in "health capital" resembles the investment in human capital which has something in common with financial capital), and more.

An obstacle to progress in all these literatures is that new scholars may find that it takes years of work to become proficient in the solution, simulation, and estimation of these models.

We have constructed a set of tools that embody a state-of-the-art, transparent, and flexible solution to arguably the canonical problem in this literature: The life cycle consumption/saving problem with uncertain income and lifetimes. These tools will be made available with an open-source license (probably on GitHub) along with infrastructure (both technological and social) designed to make it as easy (and rewarding) as possible for others to build on that foundation.

The "bottom-up" morning session describes what we have accomplished so far, what we are intending to do by the time that we roll out the "public" version of the web resource, and how all of that relates to other resources that exist now, and will solicit participant input on those topics.

## Suggested Modules List

A key component for the bottom-up approach is the list of necessary modules. Many of these will exist in some well-established form; for those we simply want to create a simple "shell" which establishes the desired interface.

The modules suggested by the models in the list below include:

• Monte-Carlo Simulation
• Dynamic Stochastic Optimization
• General Purpose Algorithms for calculating expectations
  • Nice example of the expectation function
• Simulation of population behaving according to the defined rules.
• Computing ergodic distribution (in cases where that is useful)
• Tool for computing stats on the simulated population that can be compared to stats on the data
• Doing the indirect inference
• Events-study module
  • An input: "what should population look like at moment the event occurs."

This should be modified and updated as we proceed.

## Examples in Progress

• [Carroll 2012]: "Solving Microeconomic Dynamic Stochastic Optimization Problems: Lecture Notes." link

• [Carroll-Slacalek-Tokuoka-White 2014]: "The Distribution of Wealth and the Marginal Propensity to Consume" link

## Models to Produce

See the wishlist for the official list of models to replicate with this list. These are the examples which will drive the development of an API for the bottom-up project.

See the Zotero group linked in the sidebar for the formal literature list.

## Important Questions

• What should the API look like? What we've described is essentially an API for these modules. Drive this with an example. [In progress; see the API page.]

• How to incorporate contributions from people on the cutting edge, who don't program in Python / Julia? This is a key concern if we want to incorporate cutting edge researchers.

  • There are ways to generate "wrappers" for C/C++ and Fortran, for use in Python and Julia. These have varying degrees of difficulty of application. See here for Python and here for Julia
  • As with everything, the first important step will almost certainly be implementing an example and learning from that experience.