Methodology

1. Data Ingestion and Validation

The tool accepts CSV and XLSX files containing property-level data. Column headers are auto-matched to expected fields using pattern recognition. Required fields are validated, and rows with missing critical values (sale price, RCN, age) are excluded with warnings. Non-numeric values in numeric fields trigger row-level warnings.

2. Implied Depreciation Calculation

For sales-based calibration, implied depreciation is computed for each record as:

If land value is not provided, the tool assumes the full sale price represents improvement value, and a warning is displayed. The depreciation rate is clamped between 0 and 1.

3. Outlier Detection

Outliers are identified using the Interquartile Range (IQR) method, stratified by building type. Records with implied depreciation values below Q1 - (k * IQR) or above Q3 + (k * IQR) are flagged and excluded, where k is the configurable IQR multiplier (default 1.5). Every excluded record is documented with its specific reason.

4. Age Bucketing

Records are grouped into age buckets of configurable width (1, 2, or 5 years). For each bucket, the median depreciation rate is computed. Median is preferred over mean for its robustness to remaining outliers. Buckets with fewer observations than the configured minimum threshold are flagged as "thin bins".

5. Monotonic Smoothing (PAVA)

The Pool Adjacent Violators Algorithm (PAVA) is applied to enforce isotonic (monotonically non-decreasing) depreciation with age. When a younger age bucket shows higher depreciation than an older one, the two are pooled and replaced with their weighted average. This process repeats until the entire curve is monotonic. Weights are set to the sample size of each bucket.

6. Residual Value Floor

A configurable residual value floor (default 10%) prevents depreciation from exceeding 90%. This reflects the economic principle that improvements retain some minimum value regardless of age.

7. Holdout Validation

When enabled, the data is split into training (80%) and test (20%) sets using stratified sampling by building type. The depreciation table is built on the training set, then applied to the test set to predict improvement values. Error metrics (MAE, MAPE, RMSE, median absolute error) are computed to quantify predictive accuracy.

8. Standards Alignment

While this tool does not claim compliance with any specific standard, its methodology aligns with principles described in IAAO publications regarding market-derived depreciation tables, including empirical derivation from comparable sales, age-life analysis, and statistical validation through holdout testing.

9. Limitations

• Results depend on data quality. Inaccurate RCN or sale price data will produce unreliable tables.
• Without land value separation, implied depreciation may be biased.
• This is a diagnostic tool -- generated tables should be reviewed by qualified assessment professionals.
• Local market conditions and jurisdictional standards should be considered before adoption.