Independent Analysis of the Greenburgh Reassessment

Tyler Technologies was given the lucrative contract to reassess all of the properties in the Town of Greenburgh, and the result was egregiously wrong and fundamentally flawed, riddled with simple errors, and obvious examples of overvaluations and undervaluations.

Paul Feiner and Edye McCarthy admitted that while the goal of the reassessment was for “fairness”, they would not even consider the possibility that the reassessment was fundamentally flawed, no matter how many examples of errors were identified.

Mr. Feiner and Ms. McCarthy, in dismissing the Homestead option, revealed and admitted that an intention of the reassessment was to shift the burden from commercial property owners to residential property owners.

Haberman Associates, the “monitor” of the process was clearly unsuccessful in auditing the process, so I am attempting to run an independent analysis to see the exact process by which Tyler Technologies presumably determined the reassessed valuations. However, Tyler and the Town have consistently refused to provide any actual detail on the modeling used and even full details of the parameters used for the model of each property. The publicly available information on the model is intentionally obfuscating and incomplete, and the information per parcel made available on the MMRC website is incomplete and occasionally simply incorrect.

However, using just the data that was made available, some analyses were possible, comparing the new valuations to the final 2015 tax rolls. The results of these analyses will be posted here.

A general note on the analysis:

There are 28859 parcels in the 2015 roll. This data is available on the town’s website.

Of those 28859, Tyler had 28749 preliminary assessments in their publicly available information on the MMRC website. The data from those assessments were extracted and saved, but the data on this website did not include several data points used in their model (but available on the property card), including neighborhood designation, property notes, “influence factors” on the land, condition and grade information, any manual overrides made to the assessment, Tyler’s view of the validity of past sales, and entrance information.

Of those 28749, 1288 had a change in the “type” from 2015 to tyler’s assessment; these were excluded. Lots that were valued at $0 were also excluded.

“Fully exempt” properties (such as those owned by the town or a village, like school parcels) were also excluded.

This excel file extracted some of the data collected from the publicly available files.

GoFundMe Campaign

We recently started a gofundme campaign to get data from the Town and Tyler Technologies that they have been refusing to provide given a reasonable and contractual payment for the time used.

https://www.gofundme.com/greenburgh-reassessment-data

I have submitted several FOIL requests from the Town, and while many were simply ignored, they did say that they would deliver data and records related to the reassessment and revaluations; Since Greenburgh contracts the safekeeping of these records to Tyler Technologies, I was told that I would have to pay to get the data that is meant to be available for the public. Tyler said that this would take 1 hour of time, and based on the contracted daily rate of $1,200 per hour, I agreed to pay $150 to retrieve the data. However, they refuse to deliver anything unless I pay them $600!
So I am starting this campaign so that we can finally get the data public that belongs to the public! I will make everything freely available so that anyone will have the ability to analyze the data and come to informed opinions about the efficacy and accuracy of Tyler’s reassessment.

 

After the data is delivered (assuming that Tyler actually delivers it as promised), I will provide it to the public in the format provided. I will further analyze the data to see if the data that was publicly posted earlier matches their data, and if there is anything else strange about their analysis.

 

The following is the actual FOIL request sent and accepted (after an appeal) by the Town: FOIL FORM – Tyler database; the following shows the appeal request that was accepted: foil-appeal.

 

A timeline:

Sept 26: submitted the FOIL request
Sept 26: Town Attorney replies affimatively with “okay” via email.
Oct 4: after no response, asked again for a status
Oct 5: Town Clerk said the request was forwarded to the Town Attorney and was waiting for his response.
Oct 11: emailed the Town Clerk and the Town Attorney again asking for the status after I was informed that all the documents would be delivered.
Oct 16: spoke again with Mr. Lewis on the telephone who said he would find and deliver the missing records.
Oct 22: emailed the town clerk and the town attorney asking when the complete response would be delivered.
Oct 28: emailed again asking them for a status
Oct 28: spoke on the phone with Mr. Lewis about another FOIL request, but forgot to ask about this.
Oct 31: emailed the Town Clerk and the Town Attorney again asking for the status
Nov 6: emailed the Town Clerk and Town Attorney again asking for the status.
Nov 9: confirmed choosing option 1 (1 hour of work) and sent a check for $150
Nov 21: the town returned the check refusing to deliver the records
Dec 12: The Town clerk emailed the Town Assessor to confirm the cost of one hour of work. The assessor never replied.

Sales Ratio Study

As part of its modeling, Tyler needs to prove its algorithm works. So it handpicked 1294 sales (and counted two of those twice), and produced a “Sales Ratio Study“.

Based on those results, it looks great! And well within the limits for the coefficient of dispersion given by New York State: https://www.tax.ny.gov/pubs_and_bulls/orpts/ownershandbook.htm.

But that’s because they only cherry-picked 53% of the sales during the period they were looking at — they chose to ignore 1,135 sales! So what if we used ALL of the sales between 7/1/2013 and 7/1/2015? and what if we used the sales AFTER 7/1/2015?

sales ratio study analysis

Instead of a good R-squared metric of 0.97, you get under 0.9, with a coefficient of dispersion of 159% and 289%, showing how much worse the actual model is than implied by the 7.5% coefficient calculated when cherry-picking 1300 sales out of the available 2400.

This Excel file (sales ratio study check) shows the details of arriving at the above numbers.

Another look at the land model and the 2I7 neighborhood

Tyler printed and then scanned “details” of their land model, without any context or explanation.  As discussed in an earlier post, fewer than a third of land values actually follow this model, so let’s take a closer look at a particular neighborhood, “2I7” in Irvington; a commenter had asked about this in an earlier post.

Neighborhood “2I7” is the “village” area of Irvington and includes Main Street and its side streets, as well as Barney Park, Station Road and its side streets. There are about 401 parcels, of which 285 are “residential” parcels.

The land model published on their website is adjusted by a fudge factor in over 95% of the residences in this neighborhood, applying a 5% to 20% penalty, and some adjusting the land value 50% more than their published model. It seems that Tyler is admitting that their algorithm doesn’t work and is using these fudge  factors to chase the sales for the appearance of accuracy.

The land model assumes that the minimum value of any residential lot (0.3 acres) is $483,000, yet there were several sales in the neighborhood for larger lots and livable houses at values far less than this minimum value; some examples: one at $365,000 and another for $230,000. Clearly, this land model is inappropriate.

As we can see, based on the model, and the smallest parcel in the neighborhood, the first 0.3 acres is valued at $16million per acre, and after that, the remainder valued at $787,500 per acre, unless some of your lot is deemed to be “residual” or is divided legally into a separate parcel as “vacant land”, where that land would be valued at only $62,500 per acre.

So if you have a 0.2 acre parcel, Tyler would value the land at $617k, but if that same 0.2 acre parcel was legally two 0.1 parcels, the land would be valued at $544k; clearly this is a mistake.

The two charts below are just two demonstrations how ludicrous the model is:

valueperacre

The following assumes a change to “residual” value after 0.4 acres; that bend is arbitrary and may be applied per the whim of the Tyler adjuster:

landvalue

 

Some other notes on the neighborhood:

This neighborhood was one of the hardest hit, with Tyler’s initial preliminary assessment increasing the market value on 99% of the residences, more than half seeing their assessments increasing by 75%, and more than half seeing an increase of over $300,000

Between the preliminary assessment and the 2016 tentative rolls, Tyler made “corrections” on many of the residences, adjusting 41% downward, seemingly admitting that the whole algorithm is wrong.

This is another clear example of how Tyler Technologies got the reassessment completely wrong, yet the politicians and overseers of the project have 100% faith in the process, never once questioning it.

A closer look at Tyler’s “intra-class shift” analysis

On page 6 of Tyler’s “Reassessment Tabular Statistics” pdf document, they provide a disingenuous analysis of the intra-class shift. In this analysis, they include all ‘fully exempt’ properties, and classify them as ‘commercial’ properties, fudging the numbers to hide the full impact of the erroneous reassessment process.

Tyler’s analysis:

tylershift

Using the changes from the 2015 final roll to the 2016 tentative roll, only counting matching parcels whose “class” did not change, and not counting any fully exempt properties that have zero tax liability.

actualshift

Unlike Tyler, who has hidden behind selective analyses and has repeatedly refused to provide a full dataset which could be used to provide a competing (and arguably more correct) model or even validating analyses, I am providing an excel sheet that was used to generate this analysis, for anyone to download and analyze the data as they see fit. The data in this spreadsheet came directly from the assessor’s website:

 

I have more data available; if you want access to the full database, please email me and we can give you the whole dataset.

 

The Bottom Line for the Residential Class:

Median shift across the town is +1%.

Median shift within Irvington and Edgemont is 5%, approximately $1,000 increase.

Tyler Tech’s Corrections

Since the 2016 tentative rolls were published (http://www.greenburghny.com/Documents/2016%20TENTATIVE%20ASSESSMENT%20ROLL.pdf), we can check on how many lots Tyler adjusted:

of the approximately 27,000 taxable lots, 680 saw increases and 2880 saw decreases in the assessed value from the preliminary to what was published; total net decrease of about $376million.

Increases on single family homes were as much as 4million, with percentage increases of up to 150%. Decreases on single family homes reduced the assessed value as much as 48%.

Vacant lots were much more egregious; these saw decreases of as much as 100% (fully discounted!) and increases of over 350%.

By the type of lot:

single-family homes: 11.4% had decreases, 2.5% had increases (1984, 433 of 17422)

1, 2, and 3 family homes: 11.1% decreases, 2.4% increases (2080, 455 of 18672)

residential vacant land: 25.5% decreases, 0.9% increases (464, 16 of 1813)

 

Clearly, Tyler was doing something wrong to get that many changes, and so many with significant variances. Also note that the above does not count anything they didn’t look at again, like most under-assessed lots, as well as any lots they opted to not change out of pride, spite, or indifference.

 

breaking down the corrections based on school district:

school % homes decrease % homes increase
tarrytown 9.4% 2.9%
irvington 19.3% 3.5%
dobbs ferry 11.5% 7.2%
hastings 14.9% 2.9%
ardsley 9.7% 1.6%
edgemont 20.3% 2.0%
g7 5.2% 1.4%
elmsford 4.6% 0.6%
pocantico 12.8% 2.6%
valhalla 12.0% 1.0%

Property Tax’s Effect on Market Value

 

The Town’s official stance is that property tax capitalization is a myth or a fiction. They claim that the tax burden on a property has no effect on its market valuation. Mr. Feiner and Ms. McCarthy would have you believe that two neighboring houses, all other things being equal, with house A having a granite countertop in the kitchen, and house B having a formica countertop but with $10,000 less in annual taxes, house A would command a higher price in the market! I don’t know about them, but I’d much rather save $10k every year!

Even Melissa Baer, a project manager from Tyler Technologies, admitted that for areas that were hit the hardest with tax increases, the market values would likely decrease as a result.

Taxes are clearly a major factor in any home purchasing decision. To say that a sale of a house in June 2015 when taxes were 10k would command the same market value as that sale price in July 2015 when taxes increase to 30k is simply ludicrous.

Although taxes are a factor in the sales prices and market valuations (and Tyler Tech agrees with this point), taxes were never considered a potential factor within the Tyler Tech algorithmic model. How could the model ever be considered accurate if a significant factor is not even used in the model? If they did not consider the acreage of the parcel, the model would clearly be rejected; likewise, this model should be rejected.

Reassessment Tabular Statistics

On page 10 of the tabular statistics report, the percentage shift in tax burden from commercial properties to residential properties is shown: http://www.greenburghny.com/documents/Reassessment%20Tabular%20Statistics.PDF

The percent  of intra-class shift is disingenuous because it doesn’t take into account exemptions, almost all of which apply mostly to residential properties.

If you just look at the Enhanced STAR and basic STAR exemptions, 14,143 1, 2, or 3 family lots have exemptions, about 75%. These exemptions reduce the value by $2 billion “real market value” dollars, so the actual effective tax increase on residences is closer to 7.5%, not 3.6%, which is a significant change.

Furthermore, this analysis is only provided for the town-wide shift, and in those villages where the residential properties were significantly affected, the shift in taxes from commercial to residential is extremely significant.

Assessment change by school

Per Tyler’s analysis (http://www.greenburghny.com/documents/Assessment%20Change%20Percentage%20-%20School.pdf), the change in assessment per school district was only as high as 18% for Irvington, and as low as -11% for Greenburgh. But the numbers simply don’t work out, no matter how you define “residential”:

tyler’s analysis of “residential” 1-2-3 family condo condos &
1-2-3 family
tarrytown -0.47% 12.5% 0.7% 7.2%
irvington 18.02% 28.8% -0.1% 25.0%
dobbs ferry 5.94% 20.2% 0.3% 17.8%
hastings 15.68% 29.3% -1.1% 26.0%
ardsley 3.46% 11.4% -0.5% 10.5%
edgemont 9.70% 24.4% 5.5% 22.9%
g7 -10.75% -0.5% 0.2% -0.3%
elmsford -9.63% -3.4% 9.0% -1.2%
pocantico -4.07% 15.0% -1.3% 10.1%
valhalla -1.47% 11.6% 7.8% 11.2%

Interestingly, if you take a closer look at each school district, it’s clear that the Hastings  and Irvington school districts were hit particularly hard, with average increases of nearly 30%, and almost a quarter of homes increasing by 50% or more. In Irvington, 6.5% of homes saw their assessment more than double:

school district avg chg median chg % of lots that had an increase % of lots w >25% increase % of lots w >50% increase % of lots w >double increase
tarrytown 12.5% 12.5% 70.4% 30.7% 10.1% 1.5%
irvington 28.8% 23.9% 86.2% 48.9% 21.6% 6.5%
dobbs ferry 20.2% 17.9% 81.9% 37.3% 12.0% 1.2%
hastings 29.3% 28.0% 90.5% 54.5% 24.2% 4.5%
ardsley 11.4% 10.8% 78.1% 18.1% 2.9% 0.6%
edgemont 24.4% 22.8% 91.7% 45.3% 11.5% 0.8%
g7 -0.5% -0.9% 48.1% 8.8% 1.9% 0.5%
elmsford -3.4% -3.3% 43.1% 7.3% 1.4% 0.4%
pocantico 15.0% 9.3% 68.2% 24.5% 5.0% 1.4%
valhalla 11.6% 12.5% 75.8% 17.9% 2.4% 0.7%

Land Value Model

The land value model provided by Tyler (http://mmrc.tylertech.com/_dnn/Portals/0/20160422064601.pdf) is fundamentally flawed; they provide no data showing how they arrived at the data nor any data showing how accurate it is.

Less than 1/3 of residences utilize the land value model shown on the pdf provided on the tyler website. More than 2/3 of residences have “adjustments” or “influence factors” that is only kept on the property card and is not made available publicly on their website.

When asked for this information, Tyler claims that they are not available. When I asked why they were not available – did they delete them and throw them away, the Tyler representative said that yes, they did destroy the information and hung up.

Assessment Change based on Property Type

According to Tyler’s analysis (http://www.greenburghny.com/documents/Reassessment%20Tabular%20Statistics.PDF), the tax impact looks to be roughly evenly split, but again, the data does not bear this out:

Market value change – Residential (18203)
  Reduced Approx. the same Increased
Tyler (tax) 24% 42% 34%
Actual (market value) 15% 37% 48%

 

Market value change – Condos (5158)
  Reduced Approx. the same Increased
Tyler (tax) 14% 82% 3.4%
Actual (market value) 8% 84% 8%

 

Market value change – Vacant (1813)
  Reduced Approx. the same Increased
Tyler (tax) 3% 59% 38%
Actual (market value) 29% 34% 37%

 

Market value change – All others (3156)
  Reduced Approx. the same Increased
Tyler (tax, corporate) 40% 47% 12%
Actual (market value) 31% 25% 44%

Definitions:

  • Wholly exempt and $0 value lots were excluded
  • “Approx the same” is +/- 10%
  • Residential: 1, 2, and 3 family residences
  • Vacant: all vacant lots