0.17.2

This is a moderately sized release with several exciting new features, as well as many small improvements and fixes. Most of the breaking changes (see below) are very small adjustments to parameter names or formats; two functions also had their name change. The only significant breaking change is a reworking of the interaction between AgentType instances and their associated Market with respect to aggregate-level parameters.

The new features are headlined by the addition of two models with consumption habits in the new ConsHabitModel module. Additionally, HARK's automatic HA-SSJ construction method has been extended to life-cycle models, rather than only infinite horizon models.

There are some breaking changes:

• AgentType subclasses that had a get_economy_data method now use the general AgentType.get_market_params method, which exactly replicates their prior operation. See #1719
• As a consequence of the above, random seeds on the distributions of some AgentType subclasses will change because the order in which they are created during instantiation has changed.
• Parameter PortfolioBool has been deprecated. To allow portfolio choice for RiskyAssetConsumerType, just set RiskyShareFixed=None. #1740
• The parameter BeqCRRA has been deprecated; agents with a warm glow bequest motive must use the same CRRA as their ordinary utility function. #1758
• "Terminal bequest parameters" have been deprecated; agents have the same bequest motive in period T as they do in all other periods. #1758
• calc_expectation has been renamed to expected_with_loop; use expected and pass vectorized=False for this functionality. #1763
• The argument dist in expected has been renamed to dstn. #1763
• The function make_exponential_grid has been renamed to make_polynomial_grid to reduce confusion with make_grid_exp_mult. #1762

Major Changes

• The new way to set up AgentType instances with an associated Market is to create them (with the agents in the Market's agents attribute), then invoke the Market's new give_agent_params() method. #1719
• The above method calls each agent's get_market_params() method, which references the market_vars class attribute for the names of objects to take from the associated Market.
• All interpolator classes now have default derivative methods using finite differences. These are fallback methods, and are already overridden by most subclasses. #1723
• New consumption-saving model with habit formation has been added; extends IndShockConsumerType model. #1739
• Added habit-formation model with portfolio allocation, along with example notebooks. #1748
• Simulator class has new method simulate_shock_by_grids to perturb the steady state distribution and then simulate by matrix transition methods. #1754
• Simplify parameters in ConsBequestModel.py to eliminate "terminal" bequest parameters and different CRRA for bequests than consumption. #1758
• The make_basic_SSJ method can now handle life-cycle models (cycles=1) as well as standard infinite horizon models. #1718

Minor Changes

• The special constructor get_it_from can now interpret the referenced attribute being a single value (any numeric or string) and will simply copy it to the new name. #1719
• A Market's calc_dynamics function/method can now use arguments other than those named in track_vars; HARK will look for those names as attributes of the Market. #1719
• The _derY method for LowerEnvelope2D and LowerEnvelope3D were previously bugged and returned nonsense, now fixed. #1723
• Updated syntax in a few places that tried to convert singleton array to a float, to ensure compatibility with NumPy 2.4+ #1725
• Add new income shock constructor that incorporates Velasquez-Giraldo's representation of medical expenses as negative transitory income shocks. #1724
• Add parameter dictionary with Fulford and Low's estimates for all expenses (not just medical) for use by MedShockConsumerType. #1724
• Refactoring of representative agent model solver and the "labeled" submodule. #1727
• Example notebooks for all models with portfolio choice have been significantly expanded and improved. #1740
• Example notebooks for models in ConsAggShockModel.py have been improved and expanded from their prior form. #1738
• The labels argument now works as intended with distribution.expected. #1742
• Example notebook Transition_Matrix_Example.ipynb has been cleaned up and expanded. #1744
• AggIndMarkovConsumerType added for models with both aggregate (shared) and idiosyncratic discrete states; KrusellSmithType refactored to extend it. #1747
• Light safety fixes to the new HabitConsumerType. #1753
• Example notebooks for models in ConsBequestModel.py have been improved and expanded from their prior form. #1754
• Example notebooks for KinkedRconsumerType, MarkovConsumerType, LaborIntMargConsumerType, and TractableBufferStockConsumerType have been improved and expanded. #1743
• Handling of income shocks for model "newborns" has been made consistent across models, with transitory shocks optional. #1760
• Tests added to handle a variety of unusual corner cases. #1761
• Computing expectations now always uses expected; if the function cannot accept vector arguments, pass vectorized=False. #1763
• Matrix transition methods (including HA-SSJ) now support multi-exponential grids, as well as fully custom grids. #1762
• HARK.interpolation refactored to reduce repetition and code clutter. #1765
• Small documentation notebook for life-cycle HA-SSJ construction has been added. #1718
• HARK.simulator and experimental Monte Carlo submodule refactored to reduce repetition. #1766
• HARK.distributions refactored to reduce repetition and improve structures. #1767
• Additional refactoring in Labeled, SSJutils, utilities, and metric to reduce code repetition. #1768