Mallooki

Working paper / environmental urban economics / August 2026

When Does a Flood Map Become a Treatment?

An outcome-blind economic admission audit in Kuwait

By Abdulmalik Altararwa

At a glance

Question
Can public satellite candidates support a confirmatory flood-capitalization design in Kuwait?
Main finding
The audited optical and radar candidates do not support a confirmatory design.
Evidence limit
The proxy-price results are exploratory, thinly supported, and sensitive to timing and geography.
Download PDF · 26 pages · 970 KBBrowse all research

Satellite data can measure environmental change where conventional records are thin, but a physical map becomes an economic treatment only when it aligns with event timing, market geography, and the mechanism being estimated. Applied to Kuwait's November 2018 storms, the audited optical and radar candidates do not support a confirmatory flood-capitalization design. A complete exploratory proxy family shows thin support and strong timing-and-geography sensitivity rather than a stable flood-price association.

Evidence tier
Design & measurement audit
Satellite audit
Sentinel-1 and Sentinel-2, November 2018
Exploratory family
36 average-post cells; 12 estimable
Transaction source
Kuwait Ministry of Justice; author's calculations

Question and mechanism

Flooding may change property values through damage, unreliable access, repair cost, or updated beliefs about future drainage risk. A satellite records a surface response at a specific time. It does not directly record damage, access, buyer information, or transactions. The paper asks what evidence is required before that surface becomes a treatment in an urban-capitalization design.

Economic admission audit

The audit separates reproducible image processing, physical-target validity, economic salience, historical geography, assignment-unit support, and the eventual outcome design. These decisions were frozen before housing outcomes could choose a satellite threshold.

The conservative optical rule reaches 10 of 541 polygons in the reviewed CSB-to-MOJ bridge, while the same pre-mask rule reaches 15 of 232 held-out polygons, including Sabah Al-Ahmad. Radar candidates reach 49, 26, and 11 approved polygons, but none passes the frozen radar-optical agreement rule. Exact-time late-November point truth is unavailable in the audited public sources.

The unit glossary is deliberate: CSB polygons are source geography; MOJ-linked blocks are the bridge; assignment units are treatment-link units; and exploratory analysis blocks are the estimable observation units. Official KUNA reporting records public mobilization, closures, shelters, drainage, and reopening during the November sequence, but those reports do not identify exact-time surface water, realized block-level compensation, or market treatment: mobilization, closures and diversions, shelters, and drainage and reopening.

What the paper finds

Approved optical support10 / 541Candidate-positive polygons inside the reviewed geography bridge
Held-out optical support15 / 232Candidate-positive source polygons that cannot be linked as treatment or untreated
Radar candidate support49 / 26 / 11Approved polygons under t1, t2, and t3; all rules fail promotion
Exploratory average-post cells12 / 36Estimable; the rest lose usable within-neighborhood variation
Coefficient plot of completed-home associations under 18 optical-radar proxy definitions across three November timing contrasts; many strict definitions are not estimable and all intervals are wide.
Exploratory proxy associations. NE means absorbed or insufficient within-neighborhood variation, not an estimate of zero or any other verified flood effect. These are unvalidated machine proxies, not causal flood effects. Source: Kuwait Ministry of Justice; author's calculations.

View full-size figure (opens in a new tab)

After singleton removal, the highest-support early-sequence completed-home fit uses 12 candidate blocks in 10 neighborhoods. Its association is -0.020 log points with a 95 percent interval of [-0.417, 0.378]. A whole-November definition gives +0.094 and a post-13-November definition gives +0.125. The latter two proxy-positive block sets do not overlap, so the pattern is joint timing-and-geography sensitivity, not a clean comparison of clocks. Pretrend equivalence is not established and support is leverage sensitive. The analysis establishes neither a zero nor a nonzero flood effect; the proxy associations are compatible with both directions.

What the paper does and does not claim

The paper shows that reproducible pixels do not automatically identify an economic effect. Geography linkage and market salience are part of treatment measurement. It does not claim that only ten places flooded, that the storms had no price effect, or that the exploratory coefficients measure verified flood damage.

A causal extension needs event-time positive and dry validation, damage or access evidence, repaired historical geography, dispersed block support, and a design for sale selection and spatial inference.

Citation and data source

Suggested citation: Abdulmalik Altararwa. 2026. "When Does a Flood Map Become a Treatment? An Outcome-Blind Economic Admission Audit in Kuwait." Working paper.

Transaction source: Kuwait Ministry of Justice, Live Registered Real-Estate Sale-Price Query. Exploratory price estimates are the author's calculations from registered transactions. The public PDF contains aggregate estimates and no transaction rows, deed identifiers, or exact transaction locations.