To Analyze a Sales Population, Consider the Expanded Percentile Curve

(Click on the image to enlarge)

As we know, not all segments of the market move in tandem. When the market starts to move up, it generally begins at the bottom of the value strata (start-up homes) and graduates up the value ladder. Therefore, while analyzing a large sales population, it is prudent to use the entire percentile curve (as shown in the above Miami graphic) rather than just the Median as it may musk the actual pictures on both ends of the curve, say below the 25th percentile and above the 75th percentile and more precisely below the 10th and above the 90th.

How to Analyze a Sales Population

1. Single Parameter – Instead of just one parameter (like the Median), it's better to consider the expanded percentile curve, preferably 1st percentile to 99th percentile, avoiding minimum and maximum as they may skew the picture as well.

2. Sample Selection – When confronted with all sales meaning both arms-length and non-arms length and virtually no time to validate the sales, the 5th to 95th percentile sample is more meaningful, without having to spend time on manual validations. Conversely, if the sample comprises only the arms-length sales, the 1st to 99th percentile range could be more meaningful.

3. Outlier Analysis – Therefore, while studying outliers of a sample lacking validation, it's better to consider only the cases below the 5th and above the 95th percentile. Likewise, below 1st and above the 99th could be a better starting point for a validated sample, gradually extending out to the outliers on both ends of the percentile curve (as time permits).

4. Sales Timeframe – When the timeframe is extended (9 to 24 months), sales must be time-adjusted, preferably at the monthly level (deriving monthly time factors). If the sample comprises 3-4 years' of sales, quarterly adjustments will make more statistical sense. When an extended series (e.g., 10+ years) is analyzed, annual factors would be appropriate. Most extended series analyses are performed to detect seasonality in the data.

5. Growth Factors – As we all know, the residential market is as local as it gets. Therefore, a good sales analysis must additionally be broken down to the sub-market level as long as those sub-markets are well-established and accepted. Since the growth rates vary by the market, time adjustment factors must be derived at the sub-market level (e.g., 12% in our example for the City of Miami). Applying national or even regional factors could result in flawed and indefensible results. Time adjustment in AVM is generally different (will be discussed later).

6. Use of Median – Due to time constraints, If one has to choose one parameter to ascertain time's impact, it must be the Median, as it is less prone to outliers (outliers heavily influence average, often distorting the analysis). In an even like that, the sale's Median must be compared with the normalized (by Bldg SF) Median, ensuring they are close to each other.

(Click on the image to enlarge)

7. Spatial Distribution – As part of the sales sampling, one must also ensure that the sales are spatially distributed in line with the population, so a meaningful spatial chart is in order alongside the data tables. In the above example, one must understand that the Median ASP and Median Bldg SF are mutually exclusive, but they may be connected to get a general idea of the ASP/SF, but not for any serious analysis. To analyze the normalized ASP/SF, one needs to create the organic variable (row-wise ASP/SF) and the run percentile stats.

8. Creating Sales Ratios – In the above example, in addition to the percentile analysis of sales, the distribution of sales ratios (ratio of County Market Values to Time-adjusted Sales or ASP) is shown, thus connecting the apples-to-apples dots. The spatial chart additionally depicts the stratified sales ratios. Caution: While creating the sales ratios, one must time adjust the sales to the valuation date (in this case, 01-01-2019) as the tax roll values are as of that date, or else it would be an apples-to-oranges.

9. Regression Values – Ideally, ASP should be modeled (using multiple regressions or any other industry-accepted methodology). The resulting sales ratios of the regression values (which are smoother and statistically more significant) used all analyses, Including defining and removal of the model outliers. Regression values could also be used to challenge the tax roll market values additionally. When there is a paucity of comps, such regression values could also be used to proxy actual comps in a comparables grid.

In a nutshell, to get a better picture of the overall market, an expanded percentile distribution analysis of sales is significantly more meaningful than a simplistic median-based sales analysis. Additionally, normalized values and spatial sales ratios could provide better insight into the building blocks.

-Sid Som, MBA, MIM

homequant@gmail.com

Link to the Book