Ratio curve reporting view

dbt/models/reporting/docs.md

@@ -8,6 +8,20 @@ group, township, assessment stage, and year. Feeds public reporting assets.
 **Primary Key**: `year`, `stage_name`, `geo_id`
 {% enddocs %}
 
+# ratio_curve
+
+{% docs table_ratio_curve %}
+View for reporting model and desk review ratios by PIN and year. Feeds the desk
+review ratio curve dashboard.
+
+### Assumptions
+
+- Pre-mailed values match what Res Val used to provide us with as desk review
+values.
+
+**Primary Key**: `year`, `pin`
+{% enddocs %}
+
 # ratio_stats
 
 {% docs table_ratio_stats %}

dbt/models/reporting/reporting.ratio_curve.sql

@@ -0,0 +1,72 @@
+-- Gathers model values, pre-mailed (desk review) values, and sale prices for
+-- PINs that have both model and desk review values in the current year. Lets us
+-- compare sales ratios for model values against desk review values to ensure
+-- desk review is improving output.
+{{
+    config(
+        materialized='table'
+    )
+}}
+
+-- In order to compare desk review values against model values, we need final
+-- model run IDs by township and year
+WITH final_models AS (
+    SELECT
+        final_model.run_id,
+        final_model.type,
+        towns.township_code
+    FROM {{ ref('model.final_model') }} AS final_model
+    CROSS JOIN
+        UNNEST(final_model.township_code_coverage) AS towns (township_code)
+),
+
+-- Use the final model run IDs to grab the model values we need to compare Res
+-- Val's desk review values against
+model_vals AS (
+    SELECT
+        assessment_pin.meta_pin AS pin,
+        assessment_pin.pred_pin_final_fmv_round AS model_value,
+        NULLIF(assessment_pin.sale_ratio_study_price, 0) AS sale_price,
+        assessment_pin.sale_ratio_study_date AS sale_date,
+        assessment_pin.sale_ratio_study_document_num AS sale_document_number,
+        assessment_pin.year,
+        assessment_pin.run_id,
+        final_models.type
+    FROM {{ source('model', 'assessment_pin') }} AS assessment_pin
+    INNER JOIN final_models
+        ON assessment_pin.run_id = final_models.run_id
+        AND assessment_pin.township_code = final_models.township_code
+)
+
+-- We are not using the most_recent_pin CTE from the recurring data reqeuests
+-- analysis since we can't really explain why res val provides PINs that only
+-- appear in the prior year
+SELECT
+    vpu.pin,
+    vpu.triad_name AS triad,
+    vpu.township_name AS township,
+    vpu.nbhd_code AS neighborhood,
+    vpu.year,
+    COALESCE(model_vals.sale_price IS NOT NULL, FALSE)
+        AS has_sale,
+    CAST(model_vals.sale_price AS BIGINT) AS sale_price,
+    model_vals.sale_date,
+    model_vals.sale_document_number,
+    -- These values are slightly different than the desk review values Res Val
+    -- provides due to rounding in iasWorld, but the differences are negligible
+    CAST(vpv.pre_mailed_tot * 10 AS BIGINT) AS desk_review_value,
+    vpv.pre_mailed_tot * 10 / model_vals.sale_price AS desk_review_ratio,
+    CAST(model_vals.model_value AS BIGINT) AS model_value,
+    model_vals.model_value / model_vals.sale_price AS model_ratio,
+    model_vals.run_id AS model_run_id,
+    model_vals.type AS model_type
+FROM {{ ref('default.vw_pin_universe') }} AS vpu
+-- Inner joins to only pull PINs that have both model and desk review values
+-- This should also ensure that we are only pulling regression class parcels
+INNER JOIN model_vals
+    ON vpu.pin = model_vals.pin
+    AND vpu.year = model_vals.year
+INNER JOIN {{ ref('default.vw_pin_value') }} AS vpv
+    ON vpu.pin = vpv.pin
+    AND vpu.year = vpv.year
+    AND vpv.pre_mailed_tot IS NOT NULL