Hamburg land values, flood exposure, and identification

Flood Risk,
Waterfront Premium,
and Hamburg Land Values

The project now makes one strong empirical claim and one unresolved one. Heavy-rain risk appears to be negatively capitalized into Hamburg land values. Coastal risk is harder, because the city’s most exposed waterfront land is also some of its most desirable land.

Loading empirical Hamburg map...

Section 1

Start with specification discipline, not with a headline coefficient

The first question is what kind of comparison the model is making. Once district fixed effects and neighborhood controls enter, out-of-sample fit improves sharply. That is why the page starts with the specification ladder rather than with a single flood estimate.

The coastal coefficient staying positive in pooled hedonic models is not comforting. It is the sign that the citywide design is still bundling amenity value and hazard.

Loading specification ladder...

Section 2

Heavy rain is where the project can make a strong claim

The pluvial side is the cleanest result in the project. One-scenario regressions, nested comparisons, the relative SRI index, and cross-fitted DML all move in the same direction: more heavy-rain exposure is associated with lower land values.

The visual question is therefore not just whether the sign is negative, but how the discount steepens as exposed area grows and where the response surface starts to flatten.

Loading pluvial explorer...

Section 3

The coastal question is not solved by more coefficients

A positive citywide coastal coefficient does not mean the market rewards flood risk. It means the data bundle together two forces moving in opposite directions:

observed coastal coefficient=waterfront premium-hazard penalty+residual sorting

The right exploratory move is to shrink the comparison set: stay within waterfront land, stay inside the Elbe corridor, and then test the flood-boundary designs. That is where the local negative sign finally starts to appear.

Loading coastal identification explorer...

Section 4

Causal ML belongs in the project as a methods lab, not as a magic fix

Flexible learners help in two ways here: they show whether the pluvial result survives nuisance adjustment, and they reveal how much the coastal estimate depends on the observed covariates doing the work.

What they do not do is erase the selection-on-observables assumption or solve the waterfront premium problem by themselves. The section below is therefore built as a workbench rather than as a victory lap.

Loading causal-ML workbench...