Linking obligor-specific data to individual municipal bond securities

We can weight our inference of the percentage of the population with regular preventative care accordingly. Within the blue boundary, we would infer a total population of 90 residents (18 in tract 1 and 72 in tract 2). Of these 90 residents, the number inferred to have access to regular preventative care would be: (30% access x 18 residents of tract 1) + (70% access x 72 residents of tract 2). In other words, the added information about the distribution of homes allows us to refine our inference and conclude that 62% of the residents living within the blue obligor boundary likely have regular preventative care.

Simple average-based inference:50% of the population in the blue obligor boundary has access to regular medical care.

Weighted inference:62% of the population in the blue obligor boundary has access to regular medical care.

As the above display shows, without the additional high-resolution information about the distribution of residential neighborhoods in these tracts (the percentage of homes, locations of fields, etc.), our simple average-based inference would likely have been a significant underestimate.

ICE Climate uses exactly this kind of weighting approach for all municipal bond obligor boundaries, but on a much wider range of spatial scales. Data provided at the census tract and county levels is downscaled onto a high-resolution 100-meter by 100-meter grid across the entire United States-1.1 billion grid cells in total. This downscaling is done for all sorts of ICE Climate data sets, including the percentage of people with health insurance coverage, the percentage of residents with a commute that is 20-40 minutes, the percentage of people on Medicare and Medicaid, the ambient diesel concentrations in the air, life expectancy, and road deaths, to name only a few.

For most of these types of data, ICE uses a weighting scheme based on high-resolution (100 meter by 100 meter) inferences about the distribution of residential properties within the smallest scale of census-defined boundary. Census block and tract geometries are in fact built on top of an even more granular set of over 20 million census-defined polygons, called 'topological faces.' Constructing weights for all high resolution (100 meter x 100 meter) grid cells within each of these topological faces is obviously not a simple process. To do this, ICE built probabilistic classification models that incorporate parcel data alongside information about road proximity, building footprints, and building heights to infer the likely distribution of building types (residential, commercial, industrial, etc.) within the intersection of each high-resolution grid cell and its census-defined topological face. Ultimately, for the residential case, weights represent the estimated proportion of total residential building footprint area in the topological face that are located within that grid cell.

Depending on the specific case, ICE Climate also constructs and uses weights for these grid cells that are based on other types of high-resolution information, like the estimated distribution of non-residential buildings, commercial buildings, office buildings, protected land areas, unbuilt areas, or cultivated crop areas. For example, for EPA data on the locations of toxic releases, weights are based on the presence of "any structures" within a given high resolution grid cell.

Importantly, in this process, the data being downscaled always comes first. Thus, if the U.S. Census provides a median household income value of $70,000 for a given tract, but our residential weights suggest that the tract has no residential properties, we do not re-assign that tract a median household income of $0. Instead, we apply a series of other weighting schemes in order of relevance, starting with weights based on all structures and unbuilt land. In the most extreme cases-tracts that contain only water-if the census provides a value for the tract, we will distribute that value evenly across all grid cells in the tract. The idea is to use the weights to make smart inferences about U.S. Census Data, without ever altering the underlying values themselves.

Once appropriately downscaled, data can then be aggregated up to any geospatial boundary-whether it be a drinking water, irrigation, wastewater, energy utility service area, a school district, a special development district within a town, a turnpike authority, or the area within a 30-minute drive time from a major hospital.

Whether applied to data from the U.S. Census, the EPA, FEMA, or any other source, ICE's downscaling and re-aggregation methodology is a powerful tool. With it, ICE can go beyond just linking securities to obligor geospatial boundaries-we can connect those same securities or boundaries to accurate information about the communities that live within them.

Public link

Please use the above public link if you want to share this noodl on another website.