prepared by bob mckenna licensed property tax consultant · 2020-04-16 · bedroom count 3.0 3.0 $0...
TRANSCRIPT
2510 ROSEWOOD AVE, AUSTIN 1
Appraisal Contestation Report for 2510 Rosewood Ave, Austin
Prepared by Bob McKenna
Licensed Property Tax Consultant
2510 ROSEWOOD AVE, AUSTIN 2
Appraisal Contestation Report for 2510 Rosewood Ave, Austin
Introduction
This report presents detailed evidence to contest the appraisal of the subject property, 2510
Rosewood Ave. The appraisal is contested because: 1. The Appraised Value is unequal compared with
other properties; and 2. The Appraised Value is not calculated at the property’s market value.
To be appraised in an equal and uniform manner, the appraisal district should set the appraised value
of the subject property to $643,969. To match the market value of the subject property as of January 1,
2020, the appraisal district should set the appraised value of the property to $608,800. Texas courts have
held that “If a conflict exists between taxation at market value and equal and uniform taxation, equal and
uniform taxation must prevail.” Consequently, the appraisal district should appraise the subject property at
$643,969 unless the equal and uniform contestation is rejected. If it is rejected, then the appraisal district
should instead appraise the subject property at $608,800.
Data utilized in this report is sourced from a combination of Travis County public data, publicly
available databases collected and distributed by Zillow and its affiliates, and personal attestation.
2510 ROSEWOOD AVE, AUSTIN 3
About the Subject Property
2510 Rosewood Ave is a single-family home located in the 78702 ZIP Code of Austin. It has 3
bedrooms and 2.5 bathrooms. It is a 2,441 square foot home on a lot of 8,187 square feet. The home was
built in 2012.
The maps on the following page show the location of 2510 Rosewood Ave with a red pin. The first
map situates the property within Travis County. The second map zooms in on its immediate neighborhood
in Austin.
2510 ROSEWOOD AVE, AUSTIN 4
Figure 1. 2510 Rosewood Ave within Austin and Travis County.
Figure 2. The immediate neighborhood of 2510 Rosewood Ave.
2510 ROSEWOOD AVE, AUSTIN 5
2510 Rosewood Ave is Not Appraised in an Equal and Uniform Manner
Summary of Results
The subject property, 2510 Rosewood Ave, is not appraised in an equal and uniform manner because
the appraised value of the property ($669,865) is greater than the median appraised value of a set of 5
comparable properties when appropriately adjusted ($643,969). Consequently, the appraisal district must
adjust their appraised value of the subject property to $643,969.
Comparable Properties and Adjustments
Following the guidelines found in The Appraisal of Real Estate, we assess property similarity
between our subject and its comparables on the basis of absolute adjustment in appraisal price, rather than
simply the net change in price from the adjustments.
Comparable Number Comparable Address Adjusted Appraisal
1 1412 Harvey St $670,130
2 2101 S L Davis Ave $683,054
3 2601 Sol Wilson Ave Unit#b $638,812
4 2601 Sol Wilson Ave Unit#a $638,873
5 1813 Maple Ave $643,969
Median Adjusted Appraisal $643,969
2510 ROSEWOOD AVE, AUSTIN 6
Comparable Property 1
Adjustment Grid
All adjustments are performed on a per-square-foot (PSF) basis. The Building SF adjustment reflects
that smaller properties are typically more valuable on a PSF basis than larger properties.
Attribute Subject Property Comparable 1 Adjustment (PSF)
Address 2510 Rosewood Ave 1412 Harvey St
ZIP Code 78702 78702 $0
Property Type A1 A1 $0
Bedroom Count 3.0 4.0 $0
Bathroom Count 2.5 3.5 $0
Year Built 2012 2013 $0
Building SF 2,441 2,452 $0
Lot SF 8,187 12,002 $-6
Neighborhood Code E3009 E3009 $0
Quality Code 5 5 $0
Appraisals and Adjustments
Comparable 1 Appraised Value PSF: $280
Comparable 1 Net Adjustments PSF: $-5
Comparable 1 Adjusted Value PSF: $275
By using the Subject Property’s value of 2,441 Square Feet, we arrive at the Comparable 1
Adjusted Appraised Value: $670,130
Property Similarity
Distance from Comparable 1 to Subject Property: 0.47 miles
Absolute Adjustments as Percent of Appraisal: 2.2%
2510 ROSEWOOD AVE, AUSTIN 7
Comparable Property 2
Adjustment Grid
All adjustments are performed on a per-square-foot (PSF) basis. The Building SF adjustment reflects
that smaller properties are typically more valuable on a PSF basis than larger properties.
Attribute Subject Property Comparable 2 Adjustment (PSF)
Address 2510 Rosewood Ave 2101 S L Davis Ave
ZIP Code 78702 78702 $0
Property Type A1 A1 $0
Bedroom Count 3.0 3.0 $0
Bathroom Count 2.5 3.0 $0
Year Built 2012 2014 $0
Building SF 2,441 2,434 $0
Lot SF 8,187 7,962 $0
Neighborhood Code E3009 E2010 $-3
Quality Code 5 5 $0
Appraisals and Adjustments
Comparable 2 Appraised Value PSF: $283
Comparable 2 Net Adjustments PSF: $-3
Comparable 2 Adjusted Value PSF: $280
By using the Subject Property’s value of 2,441 Square Feet, we arrive at the Comparable 2
Adjusted Appraised Value: $683,054
Property Similarity
Distance from Comparable 2 to Subject Property: 0.58 miles
Absolute Adjustments as Percent of Appraisal: 1.2%
2510 ROSEWOOD AVE, AUSTIN 8
Comparable Property 3
Adjustment Grid
All adjustments are performed on a per-square-foot (PSF) basis. The Building SF adjustment reflects
that smaller properties are typically more valuable on a PSF basis than larger properties.
Attribute Subject Property Comparable 3 Adjustment (PSF)
Address 2510 Rosewood Ave 2601 Sol Wilson Ave Unit#b
ZIP Code 78702 78702 $0
Property Type A1 A4 $0
Bedroom Count 3.0 3.0 $0
Bathroom Count 2.5 2.5 $0
Year Built 2012 2014 $0
Building SF 2,441 2,301 $-4
Lot SF 8,187 4,068 $6
Neighborhood Code E3009 E311CP $0
Quality Code 5 5 $0
Appraisals and Adjustments
Comparable 3 Appraised Value PSF: $260
Comparable 3 Net Adjustments PSF: $2
Comparable 3 Adjusted Value PSF: $262
By using the Subject Property’s value of 2,441 Square Feet, we arrive at the Comparable 3
Adjusted Appraised Value: $638,812
Property Similarity
Distance from Comparable 3 to Subject Property: 0.22 miles
Absolute Adjustments as Percent of Appraisal: 4.1%
2510 ROSEWOOD AVE, AUSTIN 9
Comparable Property 4
Adjustment Grid
All adjustments are performed on a per-square-foot (PSF) basis. The Building SF adjustment reflects
that smaller properties are typically more valuable on a PSF basis than larger properties.
Attribute Subject Property Comparable 4 Adjustment (PSF)
Address 2510 Rosewood Ave 2601 Sol Wilson Ave Unit#a
ZIP Code 78702 78702 $0
Property Type A1 A4 $0
Bedroom Count 3.0 3.0 $0
Bathroom Count 2.5 3.0 $0
Year Built 2012 2014 $0
Building SF 2,441 2,301 $-4
Lot SF 8,187 4,068 $6
Neighborhood Code E3009 E311CP $0
Quality Code 5 5 $0
Appraisals and Adjustments
Comparable 4 Appraised Value PSF: $260
Comparable 4 Net Adjustments PSF: $2
Comparable 4 Adjusted Value PSF: $262
By using the Subject Property’s value of 2,441 Square Feet, we arrive at the Comparable 4
Adjusted Appraised Value: $638,873
Property Similarity
Distance from Comparable 4 to Subject Property: 0.22 miles
Absolute Adjustments as Percent of Appraisal: 4.1%
2510 ROSEWOOD AVE, AUSTIN 10
Comparable Property 5
Adjustment Grid
All adjustments are performed on a per-square-foot (PSF) basis. The Building SF adjustment reflects
that smaller properties are typically more valuable on a PSF basis than larger properties.
Attribute Subject Property Comparable 5 Adjustment (PSF)
Address 2510 Rosewood Ave 1813 Maple Ave
ZIP Code 78702 78702 $0
Property Type A1 A1 $0
Bedroom Count 3.0 3.0 $0
Bathroom Count 2.5 3.5 $0
Year Built 2012 2007 $0
Building SF 2,441 2,497 $2
Lot SF 8,187 5,938 $3
Neighborhood Code E3009 E2010 $-3
Quality Code 5 5 $0
Appraisals and Adjustments
Comparable 5 Appraised Value PSF: $262
Comparable 5 Net Adjustments PSF: $2
Comparable 5 Adjusted Value PSF: $264
By using the Subject Property’s value of 2,441 Square Feet, we arrive at the Comparable 5
Adjusted Appraised Value: $643,969
Property Similarity
Distance from Comparable 5 to Subject Property: 0.71 miles
Absolute Adjustments as Percent of Appraisal: 3.1%
2510 ROSEWOOD AVE, AUSTIN 11
Equal and Uniform Calculation Methodology
Legal Background. Texas Property Tax Code Section 42.26 mandates that property owners are
entitled to relief if they can demonstrate that the appraised value of their property “exceeds the median
appraised value of a reasonable number of comparable properties appropriately adjusted”.
The State Comptroller’s publication “Property Tax Basics” adds additional administrative
interpretation to this statute. That publication suggests a three-pronged test for deciding if a property
owner’s protest on an equal and uniform basis should succeed:
1. The number of comparable properties selected is reasonable. (The document suggests that three to
ten properties is considered reasonable).
2. That each of the properties is comparable to the subject property, and
3. That the value of those comparable properties has been appropriately adjusted, following generally
accepted standards.
Adjustment Model. Our adjustment model is specifically tuned to the subject property and its
comparables, is based on accurate and relevant data, and follows generally accepted practices in its
development and use.
Table 18.1 from the 14th (most recent) edition of The Appraisal of Real Estate, one of the books that
Texas law prescribes as generally accepted, suggests that appraisal adjustments for single-unit residential
properties are typically done either on a price or on a price-per-square-foot basis. Following these
guidelines we implemented an adjustment model on a price-per-square-foot basis. We provide additional
detail on our modeling process to demonstrate that our adjustment methodology follows generally accepted
guidelines.
As suggested by The Appraisal of Real Estate, we created a multiple linear regression adjustment
model that takes into account the ZIP code, property type, number of bedrooms and bathrooms, the square
footage of the building and lot, and the neighborhood code and a numerical encoding for the quality code
that has been assigned by the appraisal district. We also have accurate bedroom data for a large number of
homes by using data from recent home sales.
As input to train our adjustment model, we considered all appraised properties located within 5
miles of the subject property. We then filtered down the set of appraised properties to eliminate outliers that
2510 ROSEWOOD AVE, AUSTIN 12
could otherwise skew our results. The appraised properties we considered must satisfy all of the following
conditions:
• At least 100 square feet.
• Not owned by a municipality, the county, or a government entity like a housing authority.
• Have valid quality and neighborhood codes for appraisal.
• Have at least one bathroom.
• Have no more than three bathrooms more than the subject property.
• Have no more than three bedrooms more than the subject property, for properties with a known
number of bedrooms.
• Be built no more than 30 years before or after the subject property.
• Have no more than two times the square footage of the subject property.
• Have no more than two times the lot size of the subject property.
• Have a neighborhood code that is common to at least 10 properties within the filtered set.
We end up with a set of 20,124 appraised properties that meet these conditions. Since many of the
qualities are collinear (for instance, number of bedrooms, bathrooms, and square footage are all correlated)
we solve for a model that has a regularization parameter to prevent overfitting and is restricted to positive
coefficients in the dimensions we expect to be positive. We use five-fold cross-validation to validate our
regression choices and modeling; the model we produce is the most accurate averaged over the holdout
sets. Five-fold cross-validation means there is a 20% holdout set on each fold, precisely matching the
guidance given in The Appraisal of Real Estate.
2510 ROSEWOOD AVE, AUSTIN 13
2510 Rosewood Ave is Not Appraised at its Market Value
Summary of Results
The subject property, 2510 Rosewood Ave, is not appraised at its market value because the appraised
value of the property ($669,865) is inconsistent with the property’s market value on January 1, 2020 when
calculated using two generally accepted appraisal methods and techniques: Adjusted Comparable Sales
($616,739) and an Automated Valuation Model ($600,862).
These two values create a so-called “bracket" for the subject property’s market value on January 1,
2020. To match the property’s market value, the appraisal district should set the appraised value of the
property to the arithmetic mean of the two methods: $608,800.
The evidence and methodologies for our Adjusted Comparable Sales value and Automated
Valuation Model value are outlined in the following pages.
2510 ROSEWOOD AVE, AUSTIN 14
Adjusted Comparable Sales
To create a sales comparison model, we begin by training an adjustment model specifically tuned to
the subject property and its potential comparables based on actual sales data. We use that adjustment model
to adjust the sale price of each potential comparable sale to match the attributes of the subject property.
How much we need to adjust the sale price is generally the largest factor in determining how similar the
subject property is to a potential comparable sale.
We then selected 3 comparable sales based on their similarities to the subject property. We adjust
those comparable sales based on our model, and then produce an estimated sale price for the property on
January 1, 2020 by doing a similarity-weighted average of the adjusted comparable sales.
Comparable Number Similarity to Subject Property Adjusted Sale Price
1 0.76 $645,042
2 0.75 $593,477
3 0.75 $611,090
Similarity-Weighted Estimated Sale Price $616,739
NB: For legal reasons related to data sharing, we have obscured certain identifying
information for the comparable sales listed on the following pages.
2510 ROSEWOOD AVE, AUSTIN 15
Comparable Property 1
All adjustments are performed on a per-square-foot (PSF) basis. The Building SF adjustment reflects
that smaller properties are typically more valuable on a PSF basis than larger properties.
Adjustment Grid
Attribute Subject Property Comparable 1 Adjustment (PSF)
Address 2510 Rosewood Ave 10•• Muse Ln ••••••••
ZIP Code 78702 78702 $0
Property Type A1 A4 $-2
Bedroom Count 3.0 3.0 $0
Bathroom Count 2.5 2.5 $0
Year Built 2012 2016 $-4
Building SF 2,441 1,7•• None
Lot SF 8,187 2,7•• $17
Sale Date – October 29, 2019 $2
Sale Price Adjustments
Comparable 1 Sale Price PSF: $279
Comparable 1 Net Adjustments PSF: $-15
Comparable 1 Adjusted Sale Price PSF: $264
By using the Subject Property’s value of 2,441 Square Feet, we arrive at the Comparable 1
Adjusted Sale Price: $645,042
Property and Sale Similarity
Distance from Comparable 1 to Subject Property: 0.44 miles
Absolute Adjustments as Percent of Sale Price: 18.8%
Sale Date in Decimal Years Before January 1, 2020: 0.18
Sale Similarity Measure: 0.76
2510 ROSEWOOD AVE, AUSTIN 16
Comparable Property 2
All adjustments are performed on a per-square-foot (PSF) basis. The Building SF adjustment reflects
that smaller properties are typically more valuable on a PSF basis than larger properties.
Adjustment Grid
Attribute Subject Property Comparable 2 Adjustment (PSF)
Address 2510 Rosewood Ave 27•• Halcyon D•••••••••
ZIP Code 78702 78702 $0
Property Type A1 A4 $-2
Bedroom Count 3.0 3.0 $0
Bathroom Count 2.5 2.5 $0
Year Built 2012 2015 $-3
Building SF 2,441 1,8•• None
Lot SF 8,187 3,0•• $16
Sale Date – August 16, 2019 $5
Sale Price Adjustments
Comparable 2 Sale Price PSF: $249
Comparable 2 Net Adjustments PSF: $-6
Comparable 2 Adjusted Sale Price PSF: $243
By using the Subject Property’s value of 2,441 Square Feet, we arrive at the Comparable 2
Adjusted Sale Price: $593,477
Property and Sale Similarity
Distance from Comparable 2 to Subject Property: 0.49 miles
Absolute Adjustments as Percent of Sale Price: 18.9%
Sale Date in Decimal Years Before January 1, 2020: 0.38
Sale Similarity Measure: 0.75
2510 ROSEWOOD AVE, AUSTIN 17
Comparable Property 3
All adjustments are performed on a per-square-foot (PSF) basis. The Building SF adjustment reflects
that smaller properties are typically more valuable on a PSF basis than larger properties.
Adjustment Grid
Attribute Subject Property Comparable 3 Adjustment (PSF)
Address 2510 Rosewood Ave 11•• Euphoria Bn••••••••••
ZIP Code 78702 78702 $0
Property Type A1 A4 $-2
Bedroom Count 3.0 3.0 $0
Bathroom Count 2.5 2.5 $0
Year Built 2012 2015 $-3
Building SF 2,441 1,6•• None
Lot SF 8,187 2,6•• $17
Sale Date – November 21, 2019 $1
Sale Price Adjustments
Comparable 3 Sale Price PSF: $266
Comparable 3 Net Adjustments PSF: $-15
Comparable 3 Adjusted Sale Price PSF: $250
By using the Subject Property’s value of 2,441 Square Feet, we arrive at the Comparable 3
Adjusted Sale Price: $611,090
Property and Sale Similarity
Distance from Comparable 3 to Subject Property: 0.49 miles
Absolute Adjustments as Percent of Sale Price: 19.8%
Sale Date in Decimal Years Before January 1, 2020: 0.11
Sale Similarity Measure: 0.75
2510 ROSEWOOD AVE, AUSTIN 18
Sales Comparison and Adjustment Methodology
Legal Background. Texas Property Tax Code Section 42.23 mandates that property in Texas, with
limited exceptions, should be appraised at its market value as of January 1, and that the market value
should be determined based on generally accepted appraisal methods and techniques.
The Travis County Appraisal District writes that “In the absence of a sale of the subject, sales prices
of comparable properties are usually considered the best evidence of market value.” Adjusted Comparable
Sales is a generally accepted appraisal method that incorporates sale prices.
Comparable Sales Adjustment Model. To ensure adherence to generally accepted standards, we
referenced the 14th (most recent) edition of The Appraisal of Real Estate in the development of our
adjustment model. Furthermore, our adjustment model is specifically tuned to the subject property and its
comparables. The data used in the model is accurate and relevant data, as described in the Introduction of
this report.
Table 18.1 from the 14th (most recent) edition of The Appraisal of Real Estate, one of the books that
Texas law prescribes as generally accepted, suggests that appraisal adjustments for single-unit residential
properties are typically done either on a price or on a price-per-square-foot basis. Following these
guidelines we implemented an adjustment model on a price-per-square-foot basis. We provide additional
detail on our modeling process to demonstrate that our adjustment methodology follows generally accepted
guidelines.
As suggested by The Appraisal of Real Estate and required by Texas law, we created a multiple
linear regression adjustment model that takes into account the ZIP code, property type, number of
bedrooms and bathrooms, the square footage of the building and lot, and the date of sale. As required by
Texas law, we only consider sales within the 36 months prior to January 1, 2020.
As input to train our adjustment model, we considered all sold properties located within 10 miles of
the subject property. We then filtered down the set of sold properties to eliminate outliers that could
otherwise skew our results. The sold properties we considered must satisfy all of the following conditions:
• At least 100 square feet.
• Have at least 1 bathroom.
• Have no more than three bathrooms more than the subject property.
2510 ROSEWOOD AVE, AUSTIN 19
• Have no more than three bedrooms more than the subject property.
• Be built no more than 30 years before or after the subject property.
• Have no more than two times the square footage of the subject property.
• Have no more than two times the lot size of the subject property.
We end up with a set of 9,629 property sales that meet these conditions.
Measuring and Weighting Similarity. Following the suggestions of The Appraisal of Real Estate
we assess property similarity between our subject property and its comparables using three inputs:
1. The absolute adjustment in sale price as a fraction of the sale price.
2. The date of the sale.
3. The distance from the subject property.
Consider a comparable sale i. Let the absolute adjustment in sale price as a fraction of the sale price
be ai ≥ 0, let the date of the sale, expressed as a decimal number of years before January 1, 2020 be
di ≥ 0, and let the distance between the comparable and the subject property be ∆i ≥ 0. Our similarity
function then sets the similarity of comparable i to the subject property, si, by:
si = 1 − ai − 0.1d2i − 0.1∆i
We believe this is a reasonable choice of function. It treats a (very small) 2.5% absolute sale
adjustment identically to six months of time passing or a quarter-mile of distance. It treats a (modest) 10%
absolute sale adjustment identically to one year of time passing or one mile of distance. Furthermore, the
similarity function has the mathematical characteristic that (only) the sale of the subject property on
January 1, 2020 would have a similarity of 1.0.
Finally, following the guidelines found in The Appraisal of Real Estate, we combine different
adjusted sale prices not by using a mean or median but rather by doing an adjustment that takes into
account the relative similarity of each comparable sale to our subject property. We produce the estimated
sale price e by weighting each adjusted comparable sale price pi by its relative similarity si:
e =∑
i pi × si∑i si
2510 ROSEWOOD AVE, AUSTIN 20
Estimated Sale Price over Time. Texas Property Tax Code Section 23.13(c) is explicit about the
need to adjust any comparable sale for the date of the sale. Because our adjustment model uses multiple
linear regression with the date of sale as a direct input, we can easily demonstrate how our estimated sale
price of 2510 Rosewood Ave has changed over time.
Figure 3. The estimated sale price of 2510 Rosewood Ave over the past three years.
Figure 3 shows our estimate of the sale price of 2510 Rosewood Ave over the past three years. This
time effect has been applied to all of the comparable sales used to develop our price estimate.
2510 ROSEWOOD AVE, AUSTIN 21
Appraisal by an Automated Valuation Model
Modern Automated Valuation Models (AVMs) use machine learning algorithms to estimate
property values. These models figure out the combined property value of various features associated with a
home, including standard ones like square footage and number of bedrooms and more interesting ones like
where the home is (i.e., neighborhood, school zoning, etc.). Using optimization techniques drawn from
artificial intelligence, the valuation model figures out how best to combine these features in order to create
the most accurate predictive model. The drawback of AVMs is that they are, generally speaking, very
difficult for humans to fully understand, although it is possible for humans to see the inputs and assess the
accuracy of an AVM’s outputs. This drawback can be outweighed by the primary benefit of AVMs: that
they can be significantly more accurate than human-interpretable models.
The Appraisal of Real Estate outlines professional standards around the use of an AVM. An
appraiser must be able to answer affirmatively to all five of the following criteria:
1. Does the appraiser have a basic understanding of how the AVM works?
2. Can the appraiser use the AVM properly?
3. Are the AVM and the data it uses appropriate given the intended use of assignment results?
4. Is the AVM output credible?
5. Is the AVM sufficiently reliable for use in the assignment?
As we will show below, we are able to answer affirmatively to all five of these criteria.
The Appraisal of Real Estate is ambivalent about the use of AVMs for appraisal. On the one hand,
the book suggests that “the output of an AVM is not, by itself, an appraisal” and suggests that the AVM
should be used as the basis of a professional opinion if it is credible. Our use of an AVM follows these
guidelines. On the other hand, The Appraisal of Real Estate also recognizes that our specific use case in
this report, “property tax assessment and equity studies”, is a valid application for AVMs.
The Appraisal of Real Estate also writes that: “Appraisers should not attempt to build a statistical
model that is beyond the limits of their education and experience. As with any appraisal specialty, the
ability to address complex statistical problems grows with experience. Experience is best gained by
2510 ROSEWOOD AVE, AUSTIN 22
collaboration with a more qualified statistical analyst”. Following this recommendation, the Property Tax
Consultant preparing this report was assisted in the construction of an AVM by PXN Residential LLC, a
real-estate software company whose principals are faculty members at the Wharton School (University of
Pennsylvania) and the University of California, Berkeley, and who have deep academic and practical
experience in the construction of machine learning models for predictive analytics.
Results with the AVM. One of the strengths of the AVM is that the price of the subject property
on January 1, 2020 can be precisely estimated by inputting the characteristics of 2510 Rosewood Ave into
the model and setting a (hypothetical) sale date of January 1, 2020. When we did that, the AVM produced a
guessed sale price of $600,862.
Professional Standards Test with our AVM. Our use of the AVM meets the five-prong test
established by the Appraisal Institute.
1. Does the appraiser have a basic understanding of how the AVM works?
Yes. The AVM creates a regression model to predict the sale price for all single-family properties
with between 2 and 4 bedrooms in the 78702 ZIP code with square footage within 25% of 2510
Rosewood Ave and lot square footage within 50% of 2510 Rosewood Ave within 2 miles of 2510
Rosewood Ave. The model only considers those homes sold in the 36 months prior to January 1,
2020.
2. Can the appraiser use the AVM properly?
Yes. This AVM is designed for the prediction of sale prices. It takes as input the attributes of the
subject property and outputs a sale price.
3. Are the AVM and the data it uses appropriate given the intended use of assignment results?
Yes. The data considered by the model includes the ZIP code, the number of bedrooms, the number
of bathrooms, the size of the property, the location of the property as well as the date of the sale.
4. Is the AVM output credible?
Yes. The results of the AVM are shown graphically as Figure 4. Each dot represents one of the 64
relevant sold homes; the dot is placed by plotting the actual sale price against the AVM’s predicted
sale price. The blue line represents perfect predictions, where the guessed prices exactly match the
2510 ROSEWOOD AVE, AUSTIN 23
model. The dots cluster around the blue line, indicating an excellent model fit of the input data. The
large red star in the center of the plot represents the AVM’s valuation of 2510 Rosewood Ave.
Figure 4. A comparison of guessed sale prices to actual sale prices in our site-specific AVM. Each black
dot represents a relevant sale and is plotted based on the sale price from our model and its actual sale price.
The blue line represents perfect accuracy, wherein the model’s prices would exactly predict sale prices. The
model’s estimate of the price of 2510 Rosewood Ave on January 1, 2020 is shown by the red star.
An important quantitative characteristic for assessing the credibility of the AVM is error. The error
is the average amount that the model either overpredicts or underpredicts the real sale price. For
instance, if the model guesses a sale price of $350,000 for a home and the home actually sold for
$355,000, then there is an error of $5,000 associated with that prediction. The AVM has an average
error of $2,210 for the homes featured in Figure 4. This represents less than 1% of the model’s
valuation of 2510 Rosewood Ave. Given these error rates, the site-specific AVM we developed here
is both appropriate for use in estimating the price of 2510 Rosewood Ave and is significantly more
accurate than Zillow’s Zestimate AVM.
5. Is the AVM sufficiently reliable for use in the assignment?
2510 ROSEWOOD AVE, AUSTIN 24
Yes. A quantitive measure of the reliability of an AVM is known as bias. A biased model
consistently guesses too high or too low in its home price predictions. For the homes featured in
Figure 4, the AVM has a bias of $15, meaning that, on average, it guesses that home prices are $15
less than the actual sale price. Since the average sale price of homes in Figure 4 is much larger than
this value, we can safely say that the AVM is essentially unbiased, and therefore sufficiently reliable
for use in this assignment.
Regarding Zillow’s Zestimate. Zillow’s Zestimate is the best-known AVM. However, we do not
consider using Zestimate for 2510 Rosewood Ave because it suffers from two problems.
First, because of data fragmentation in the State of Texas, Zillow does not have enough data to
accurately value properties in Travis County. Zillow’s own corporate communications rate Zestimate’s
accuracy in Travis County as just “one star” out of four (where two stars are “Fair” quality, and three stars
are “Good” quality), and moreover, Zillow is unwilling to attest to the accuracy of Zestimates in Travis
County. Zillow generally reports Zestimate errors (how close actual sale prices are to Zestimates) for most
counties across the USA. But in Travis County, Zillow does not report an average error for Zestimates
versus actual sale prices. On a nationwide basis Zillow claims that Zestimate is typically 1.9% away from a
home’s true value, but that error is likely to be higher in Travis County. Zillow’s own communications
describe Zestimate accuracy as worse than “Fair” in Travis County, but in Tarrant County, where the
Zestimate accuracy is listed as “Fair”, Zestimate’s typical error is 7.4%. This suggests the error of typical
Zestimates in Travis County is very likely to be higher than 8%.1
The second reason we do not use Zestimate is that it is necessarily less specific to properties in
Travis County. A Zestimate has a very broad job: it is designed to produce price predictions for every
home in every county across the USA. Instead, for our purposes it is possible to build a more precise AVM
specific to 2510 Rosewood Ave.
1 All information on the accuracy of Zestimates in the state of Texas is publicly available at:
https://www.zillow.com/howto/DataCoverageZestimateAccuracyTX.htm
2510 ROSEWOOD AVE, AUSTIN 25
Disclaimer
While care has been taken to present accurate data, neither Bob McKenna, Austin Real Estate
Partners, nor PXN Residential LLC can guarantee the accuracy of the data used in this report.
Certain data in this report is Copyright 2020 Austin Central Texas Real Estate Information Services,
Inc (ACTRIS). This report is based on information from the Austin Board of REALTORS (R)
(alternatively, from ACTRIS) for the period January 1, 2015 to January 1, 2020. Neither the Board nor
ACTRIS guarantees or is any way responsible for its accuracy. All data is provided ‘AS IS’ and with all
faults. Data maintained by the Board or ACTRIS may not reflect all real estate activity in the market.