-
Notifications
You must be signed in to change notification settings - Fork 3
Spring 2016 Research Plan
Faculty: Paul Waddell, Alexei Pozdnukov Grad Students: Danqing Zhang, Max Gardner, Nat Decker, Sam Blanchard, Sam Maurer (very limited time this semester)
Transit-oriented development projects. Urban displacement project: low income households. Unexpected effect of affordable housing policies.
Complete improvements to UrbanSim. 1. Households. 2. Developer behavior.
HH relocation (move out) model. Task 1. relocation probability as Binary Logit Model, (move=1, stay=0), with Renters and owners considered separately. Calibrated from: ACS micro-sample, yes/no if moved in the last year, and what rent is paid now (ideally must use longitudinal data, panel household). Sam has some documentation with explanations and model structures.
1.2. Household location choice model.
(only households below a certain income are allowed to purchase or to rent them). These units do not allow higher income households to locate there.
Remove the bias on the price (or rent) parameter by coding the Control Function Method (adapt from previously implemented version in the OPUS code). Here are a couple of references:
Christian Guevara's dissertation, which contains the control function method in a chapter. https://www.researchgate.net/publication/260021234_Endogeneity_and_Sampling_of_Alternatives_in_Spatial_Choice_Models
A more recent critical assessment of several methods to address the same problem https://www.researchgate.net/publication/282909292_Critical_Assessment_of_Five_Methods_to_Correct_for_Endogeneity_in_Discrete-Choice_Models
An alternative to the Control Function method that Christian Guevara just published should also be considered: https://www.researchgate.net/publication/291790737_Correcting_For_Endogeneity_Due_to_Omitted_Attributes_in_Discrete-Choice_Models_The_Multiple_Indicator_Solution_MIS
Improve the way we represent budget constraints in the Household Location Choice Model (HLCM).
We can not oversimplify - lots of HH have 100% or even higher cost burden in the data. Interaction terms for income/prices have been used in previous HLCM applications, but are not sufficient to keep from over-predicting the number of households choosing locations that are excessively high cost for their income (more than we see in the observed data). We can work on improving the specification of how we interact household income and housing cost to improve this, and compare the predictions to observed distributions of cost burdens. If needed, we can add calibration constants for high cost burden categories and calibrate these so that they produce a very high disutility of choosing an excessively expensive housing unit.
HH is a price taker individually, but should in the aggregate impact prices within submarkets in the short term. There is a price equilibration algorithm implemented in UrbanSim in which we aggregate the location choice probabilities across households for each location (or submarket aggregation of locations) and change the price to reflect excess demand or excess supply at each location (submarket). The current approach could be improved and made less ad-hoc. We have done prior work on an Availability constraints that might also be relevant if we want to try improving the estimation of HLCM to account for these aggregate effects. There us no bidding process in the model now (households are either allocated on a first come first served approach, or are allocated only after the price equilibration process ensures that demand is less than or equal to supply in all submarkets.
(we had less time to discuss this, so it's shorter)
Developer model. ‘Messiest' of the models in UrbanSim. Uses a pro forma to analyze the costs and expected revenues for alternative development projects, subject to the parcel geometry constraints and zoning constraints. Goal this semester is to add two elements to it, and improve the code robustness in general.
IDs of policies that apply to the parcel, such as Inclusionary Zoning. Nat is building this with some help on assigning it to parcels from Sam. Nat will also provide overall guidance on how these models should work.
Max would use the policy lookup for each parcel to determine whether there are any inclusionary housing policies that apply to a parcel, and if so, make changes in the revenue calculation to account for the reduced revenue caused by a specific percentage of units in the project being reserved for a certain rent level.
Max would implement a different model for development of 100% affordable units. There is some code in UrbanSim now to track Funds availability in each jurisdiction, to be used for construction of affordable housing. Factor in building costs, etc. Nat will provide more specification detail here, drawing on earlier drafts developed by Pedro Peterson (now at SF Planning but also DCRP PhD) and Aksel Olsen (now at ABAG but also in DCRP PhD).
Urban Analytics Lab | 401A Wurster Hall | Berkeley, CA