Towards 2.0

Status Quo

HARK 1.0 Status Quo

HARK currently scaffolds the modeling and solving of two kinds of problems:

• Bellman equations of continuous MDP models
  • In the implementation of the solutions to these problems, HARK generally uses policy iteration, a dynamic programming algorithm.
• Bellman equations for rational expectation equilibrium MMDP models

Currently, there is no class in HARK that represents the Bellman equation form of a model generically, and the solution code for this form. Rather, each subclass of AgentType implements the policy iteration algorithm anew.

Dolo Status Quo

Dolo provides methods for:

• Reading a .yaml file specified according to dolang, and compiling it into a dolo.compiler.model.Model class.
• A variety of algorithms, in dolo.algos, which can operate on the models and find solutions/generate results. These include:
  • Value iteration
  • Stochastic simulations

2.0 Integration Plan

dolark

dolark is envisioned as a toolkit that uses the machinery of dolo, but expanded to use rational expectation equilibrium MMDP's.

HARK 2.0

In the future, HARK can import dolo. The HARK interface can be an alternative way of writing dolo.compiler.model.Model instances (and subclasses of them).

This way, HARK can take advantage of more generic model and solution code while being backwards compatible for REMARKs and pedagogical materials.