adrian-carro/spatial-housing-model

Spatial agent-based model of the UK housing market.

A Spatial Agent-Based Model of the UK Housing Market

This is a spatial agent-based model of the UK housing market written by the Institute of New Economic Thinking (INET) at the Oxford Martin School, University of Oxford. It is based on a previous non-spatial model developed by INET in collaboration with the Bank of England. A detailed description of this initial non-spatial model can be found as a Staff Working Paper of the Bank of England .

The spatial model, while keeping the same agents as in the non-spatial version (owner-occupiers, renters, buy-to-let investors, a private bank, a central bank, a construction sector, and a government), incorporates a number of spatial features with their corresponding agent behaviours. Namely, households have to decide where to bid for housing given a certain spatial distribution of prices and commuting costs. These decisions, in turn, feed back into the market mechanism, possibly modifying the spatial distribution of prices.

spatial-housing-model's Issues

Review all the setActive, isActive stuff within the collectors

Check if it is doing anything useful or if it can simply be removed.

Remove serialization from all classes

Average Days or Months on Market

These quantities are currently computed as "flow" variables, i.e., as averages over properties sold this month. Given that the number of properties sold a month can be quite small, wouldn't it make more sense to use instead "stock" variables, i.e., to average over all properties currently on offer in the market?

Re-organise commuting algorithm methods between Household and HouseholdBehaviour classes

Most methods should be moved to HouseholdBehaviour, probably only bidForAHome to be left at Household.

recordTransaction and recordSale methods at the marketStats classes

Rethink if they can be joined in a single method!

Debug new commuting algorithm

General check, plotting main results (HPI, types of households), to make sure things are working as expected.

Revise DecideToSellHome method within HouseholdBehaviour

This method includes 2 unidentified fudge parameters, DECISION_TO_SELL_HPC and DECISION_TO_SELL_INTEREST, which are explicitly explained otherwise in the manuscript. The method explained in the manuscript should be implemented and tested.

BTL investors should be able to invest in different regions

For this one would need to change, at least, decisionToBuyBuyToLet at HouseholdBehaviour and step method at Household.

Change licence to Apache

Remove decideRentOrPurchaseOld

Subtract commuting cost from bank balance

Each month, current commuting cost (cost of commuting from job region to home region) should be subtracted from the household's bank balance.

Move geography to its own class

It should probably include housingStock and totalPopulation, with their corresponding getters and setters.

Replace Distance class and all corresponding files by commuting times

Remember to replace also the data files and config properties to read them. Also make it such that data is automatically read at model initialisation.

Port decideToSellHome bug solution from non-spatial code

The definition of buyer in the class of HouseBuyerRecord is not correct.

The incorrect code:
public Household buyer; // Who wants to buy the house

This is not exact, as the so-called buyer could be either purchaser or renter
This class has been used in both sale and rental markets.

Rethink number of rounds at the clearing mechanism within the HousingMarket

Currently, a cheap fix has been implemented to avoid over-accumulation of low quality houses by the construction sector. In the future, something more dynamic or with better thought rules should be implemented.

Move handleInputArguments to utilities

A new class InputArgumentsHandler should be created within utilities to store the current handleInputArguments method. Then outputFolder and configFileName can be obtained within the Model class through getters at the InputArgumentsHandler class.

Need different reference price for rental

Currently RentalMarketStats, RegionalRentalMarketStats call through their constructor the constructor at HousingMarketStats, RegionalHousingMarketStats, which initiates reference prices from data/HouseSaleMarket.

Remove Authority subclass from housingMarket

Not really needed!

No statistics are kept for rental prices

We should probably be keeping something like a Rental Price Index (RPI) to plot and compare with HPI.

Remove parameter RENT_GROSS_YIELD

Replace this parameter by a different way of initialising the average yield variables, such as real average rental prices (reference prices for rental market) (times 12 months) divided by the sale market reference prices.

Revise inheritance at Demographics

Currently, inheritance is always done at random and within the same region (which is the jobRegion of the households, not even the region where they live!). We should think of how to improve on this.

Bring parameters updates and other bug solutions from non-spatial code

In particular, the following commits should be applied from the non-spatial code:

• f01e9dc3ef02470c52f39a5ad33591d0e8f8f982
• c247a7b81b0b7d5d5656877e792c329409cb6464

Add total or average commuting cost as output

Probably total would be more helpful, as then it can be divided by total number of households or by the number of those with a home...

Cap factor at getDesiredPurchasePrice (HouseholdBehaviour)

The capping should be removed and a different equation should be used in order to avoid having negative or too large desired prices!

Improve households HPA expectations with exponential average of annual HPA

Replace current longTermHousePriceAppreciation field (and getLongTermHPA method) within both regionalHousingMarketStats and housingMarketStats in order to implement an exponential moving average. Thus, the current parameter HPA_YEARS_TO_CHECK would also need to be replaced with the corresponding smoothing parameter of the exponential average.

Rethink Demographics (remove fudge multiplier!)

Need to remove the fudge multiplier introduced at Demographics so as to ensure that the model would reach the target number of households. This fudge parameter should be removed or its value re-defined so as to make it adaptable to different system sizes.

Remove all references to Model from "regional" classes

Any needed field should be passed as a parameter in the constructor of the regional class