• Welcome to AppraisersForum.com, the premier online  community for the discussion of real estate appraisal. Register a free account to be able to post and unlock additional forums and features.

Regression Residual Techniques Will Dominate (most) Residential

Status
Not open for further replies.

IVCA

Senior Member
Joined
Jun 27, 2017
Professional Status
Certified General Appraiser
State
California
I'm becoming convince that Regression Residual techniques will come to dominate residential appraisal in most urban areas. I appraise in Pacifica (and neighboring areas) - which is a motley mix of different residential types with greatly varying degrees of upgrades and condition. It is fairly difficult to appraise.

Once I create a model for an area - and it might be half of Pacifica, I can seamlessly adjust over a dozen comparables to often within 1.5% - without actually assigning adjustments to the comps myself - they all get assigned through the model. Consider to, that these comps often vary widely in GLA +/- 1200sf, lot size, location, condition and other factors! - The only subjective input really, is the scoring of the subject property - because it usually doesn't have a COE sale price. I do review the quality/condition/view/appeal settings given by the model (yes, it will assign values for these things!), as factors can enter the adjustments that need explanation. For example, sometimes it is probate, death, a need to quickly relocate that causes that factors into an unusually low price; although that is just a matter of explanation, the value has already been accounted for by the model!

A good appraiser can generate these models for fairly large areas and they are good for months, with some updating for recent sales and market conditions.

But, so ----, I see that what will be needed in the future are appraisers that are very comfortable dealing with statistics, statistical software, graphics software, reporting software like MS Offfice and Adobe, ArcGIS, Geology, and even programming. The future appraiser will pretty much need an IQ of 130 in math at least. How far off? Maybe not that far.
 
Last edited:

Meandering

Elite Member
Joined
Feb 26, 2006
Professional Status
Real Estate Agent or Broker
State
Pennsylvania
I'm becoming convince that Regression Residual techniques will come to dominate residential appraisal in most urban areas. I appraise in Pacifica (and neighboring areas) - which is a motley mix of different residential types with greatly varying degrees of upgrades and condition. It is fairly difficult to appraise.

Once I create a model for an area - and it might be half of Pacifica, I can seamlessly adjust over a dozen comparables to often within 1.5% - without actually assigning adjustments to the comps myself - they all get assigned through the model. Consider to, that these comps often vary widely in GLA +/- 1200sf, lot size, location, condition and other factors! - The only subjective input really, is the scoring of the subject property - because it usually doesn't have a COE sale price. I do review the quality/condition/view/appeal settings given by the model (yes, it will assign values for these things!), as factors can enter the adjustments that need explanation. For example, sometimes it is probate, death, a need to quickly relocate that causes that factors into an unusually low price; although that is just a matter of explanation, the value has already been accounted for by the model!

A good appraiser can generate these models for fairly large areas and they are good for months, with some updating for recent sales and market conditions.

But, so ----, I see that what will be needed in the future are appraisers that are very comfortable dealing with statistics, statistical software, graphics software, reporting software like MS Offfice and Adobe, ArcGIS, Geology, and even programming. The future appraiser will pretty much need an IQ of 130 in math at least. How far off? Maybe not that far.

Bert,
not trying to be annoying or spying on you, but

May I ask how many of the "land" features your model takes into account as value impacts to residential properties, other than size?

Do your models adjust for usable gross/net acreage? we have many areas here were minimum acreage is 1-2 acres, so swamps, wetlands, slopes, mountain shade and the like can be issues that should not be glossed over and especially if the driveway is sloped up from the road, down from the road and if it is paved or graveled. It is important to know that a house with a chimney, the chimney will not draft well if an idiot put it on the south facing wall, which, might not seem like an issue in sunny CA, until you are in the woods in PA in an ice storm and the trees have taken out the electric for a week or more.

I'm just wondering if different geographic regions have different environmental factors that impact value, and should not be glossed over for quick math? Naw, I'm not wondering about it, I'm flat telling you that the acceptability of the math only comes with a GSE saying that's what they will accept. Doesn't really matter if the math is "good" or not. Only matters that a GSE says it's "good enough" for them. So, work on getting GSE approvals first, then appraisers will buy it from you.

.
 

ICT RE

Junior Member
Joined
Oct 7, 2011
Professional Status
Certified Residential Appraiser
State
Kansas
Good thing we just lowered the base requirements for a cert residential to 30 credit hours of gen eds, a pulse, and basal body temperature of 98.7 degrees. I’m sure these new entrants via the revised qual ed and qual exp are lining up in que with their spreadsheet and financial calculator competencies in wait.

Juuuuust kidding! Where are those adjustment lists again? Can someone please print off a few thousand new copies?

Whew... there for a second (or ~3.5 years) there was almost a modicum of respect forming in the general public for the (resi app) career field. Glad we ducked that bullet, eh?
 
Last edited:

IVCA

Senior Member
Joined
Jun 27, 2017
Professional Status
Certified General Appraiser
State
California
Yea. Just wait. Same everywhere. Technology advances, the world gets more complicated. More standard work gets replaced by computers. You have to adapt if you can. Eventually we all sit around playing politics or whatever while robots do our work.

But during the transition, expect to work your butt off … until things actually get automated. I could use 4 or 5 people to help get things done, like cleaning data, updating software, maintaining my computers and databases, writing programs and creating Excel spreadsheets, handling digital images, setting up ArcGIS overlays, ……………………………. but, yes, the result improves dramatically - eventually - in leaps and bounds.

I like to compare that 1 Terabyte microSD card you can hold on a fingertip to my old 20 pound, 6"x15"x 12" (approx.) Corvus 6 Mb hard drive (http://www.computerhistory.org/collections/catalog/102626179) . That is 166,666,667 times more storage in probably that much less space. Took 36 or so years. … And now with Quantum Computers?

https://www.nextplatform.com/2018/01/10/quantum-computing-enters-2018-like-1968/

Note: Although you can find 1T microSD cards, they are not that reliable. The best now is probably the SanDIsk 400G card: https://www.sandisk.com/home/memory-cards/microsd-cards/extremepro-microsd-a2.
 
Last edited:

Amy Perkins

Senior Member
Joined
Jul 20, 2003
Professional Status
Certified Residential Appraiser
State
Tennessee
Everyone knows statistics can be manipulated, but nothing will replace an ethical appraisers mind that has the experience, knowledge, and education necessary to produce credible results.
 

IVCA

Senior Member
Joined
Jun 27, 2017
Professional Status
Certified General Appraiser
State
California
Bert,
not trying to be annoying or spying on you, but

May I ask how many of the "land" features your model takes into account as value impacts to residential properties, other than size?

Do your models adjust for usable gross/net acreage? we have many areas here were minimum acreage is 1-2 acres, so swamps, wetlands, slopes, mountain shade and the like can be issues that should not be glossed over and especially if the driveway is sloped up from the road, down from the road and if it is paved or graveled. It is important to know that a house with a chimney, the chimney will not draft well if an idiot put it on the south facing wall, which, might not seem like an issue in sunny CA, until you are in the woods in PA in an ice storm and the trees have taken out the electric for a week or more.

I'm just wondering if different geographic regions have different environmental factors that impact value, and should not be glossed over for quick math? Naw, I'm not wondering about it, I'm flat telling you that the acceptability of the math only comes with a GSE saying that's what they will accept. Doesn't really matter if the math is "good" or not. Only matters that a GSE says it's "good enough" for them. So, work on getting GSE approvals first, then appraisers will buy it from you.

.


With respect to your question. No problem.

1. I enter MLS hard data into my regression and it produces a model that predicts sale price based on just those parameters. You can enter ANY hard data. I only enter lot size for land data, - so far this past year. But frontage would be something I would add in certain circumstances.
2. Run the model on your data to get a predicted (or estimated) sale price. Subtract that from the actual sale price to get a residual.
3. Sort the data by residual value from smallest to largest.
4. Create a new column next to residual values and call it something like CQA (Condition/Quality/Appeal/Other).
5. Insert this macro into the second cell under CQA: =INT((Row()-1)/(Observations/20))/2.
6. Replace "Observations" with the number of rows -1 (i.e the number of sales). Copy it down all the empty cells. It will replace itself with scores from 0.0 to 10.0 based on percentage of rows with smaller Residual values.
7. Copy the CQA column and replace it with itself "by Value" - to get rid of the macros, as you will be resorting the data and the macros would otherwise change their value.
8. Now run your regression again with CQA as the Target variable, and Residual Value as the predictor, to get a regression model for the CQA vs Residual value.
9. Now all of your comps are in the Excel spreadsheet and they all have CQA values. Those CQA Values assign a residual value via the function you just created via regression in step 8. This, in other words, tells you exactly what value contribution to assign to all of the features that you did not enter in step 1. It tells you the combined value contribution … and as it turns out that is good enough because, for other reasons, you can split it up among features any way you like for all comps and the subject and take differences to the subject for adjustments and add them up to get adjusted sales prices. Weight and average the sales prices to get a value for the subject. It doesn't matter how you split the values, the splitting has no effect on the adjusted values. Simple math. But splitting is necessary to explain WHY a comparable is better or worse than the subject.
10. Oh but forgot, yes, since you don't have a sale price for the subject to derive a CQA for it, you must ESTIMATE THE CQA for the subject yourself. You must rank it among the comps based on the features you did not enter in Step 1. Now, that is something the human mind is actually better at than computers, in more ways than one ( for one thing it can't walk through the subject and see 100% of everything [that is observable]). If regression has removed 80% of the price variance (reflected by an R2 value of 0.80), then a 10% error in your estimate of the subject's CQA would result in only a 2% (0.10*0.20) error. However, you can actually fine tune your estimate of the subject CQA through various means - not that it is really necessary.

But essential to getting a tight convergence of the adjusted values is getting a good model. I can only recommend Salford Systems MARS, with Automate/Bootstrapping … (combined with my own SECRET/PROPRIETARY run parameters - that I keep in my workfile. Everyone can work out their own methods in this regard. And I am confident you can get results with other multivariate statistical regression programs - I just don't have the time to investigate. And I am pretty sure if you try to use R for MARS, it will take a while to piece all the necessary packages together, with the "earth" package of primary interest. [I've heard of statisticans who do this … not at all sure how much work it is.]

See the next post for a sample model (as it is too long to append here.)

Oh .. and understand my model is based on sales going back to 2001. That's why you see those high SaleAge numbers (SaleAge is the number of days of COE before my date of value … and this is a retrospective appraisal).
 
Last edited:

IVCA

Senior Member
Joined
Jun 27, 2017
Professional Status
Certified General Appraiser
State
California
This is a sample model as a C# program. It reads maybe 12+ properties and spits out all adjustments:

using System;
using System.Linq;
using AutoMapper;
using Model.Properties;

namespace Model
{
public class SalesComparison03
{
public int[] area1 = {654, 657, 653, 655};
public int[] area2 = {652, 651, 654, 655, 656};
public int[] area3 = {652, 657, 653, 655};
public Property Subj { get; set; } // Subject Property


private static double max(double a, double b)
{
return Math.Max(a, b);
}


public double GetAdjustedSalePriceAndSalePriceEstimate(Property prop, out double salePriceEstimate)
{
if (prop.IsSubject) Subj = Mapper.Map<Property>(prop);

// Stage I
// Basis Functions
var BF1 = max(0, prop.SaleAge - 1824);
var BF2 = max(0, 1824 - prop.SaleAge);
var BF3 = max(0, prop.SaleAge - 4485);
var BF6 = max(0, prop.SaleAge - 2167);
var BF8 = max(0, prop.SaleAge - 5485);
var BF25 = max(0, prop.SaleAge - 3948);
var BF27 = max(0, prop.SaleAge - 3294);
var BF29 = max(0, prop.SaleAge - 4549);

var BF5 = max(0, prop.LivingArea - 1100);

var BF10 = max(0, prop.Garage - 3);
var BF11 = max(0, 3 - prop.Garage);

var BF14 = max(0, prop.LotSize - 8850);
var BF15 = max(0, 8850 - prop.LotSize);

var BF18 = max(0, prop.Baths - 1);

var BF21 = max(0, prop.Bedrooms - 3);
var BF22 = max(0, 3 - prop.Bedrooms);

var BF23 = max(0, prop.Stories - 2);
var BF24 = max(0, 2 - prop.Stories);

var BF12 = area1.Contains(Convert.ToInt32(Math.Round(prop.AreaID))) ? 1 : 0;
var BF16 = area2.Contains(Convert.ToInt32(Math.Round(prop.AreaID))) ? 1 : 0;
var BF19 = area3.Contains(Convert.ToInt32(Math.Round(prop.AreaID))) ? 1 : 0;



// Basis Functions grouped into meaningful component values
prop.Contributions.Prop.BaseValue = 552820;
prop.Contributions.Prop.Baths = +27019.6 * BF18;
prop.Contributions.Prop.AreaID = +56049 * BF12 + 21330 * BF16 - 18333.9 * BF19;
prop.Contributions.Prop.SaleAge = -352.26 * BF1 + 237.454 * BF2 + 593.034 * BF3 + 453.172 * BF6 +
249.207 * BF8 - 271.249 * BF25 + 138.478 * BF27 - 855.869 * BF29;
prop.Contributions.Prop.Stories = +71168.7 * BF23 + 4761.81 * BF24;

prop.Contributions.Prop.LivingArea = +130.953 * BF5;
prop.Contributions.Prop.LotSize = +2.93385 * BF14 - 8.6696 * BF15;

prop.Contributions.Prop.Garage = +171836 * BF10 - 21699.6 * BF11;
prop.Contributions.Prop.Bedrooms = -7657.26 * BF21 - 36040.3 * BF22;


// Stage II
// Basis Functions for the CQA-Residual regression
var CBF1 = max(0, prop.CQA - 9);
var CBF2 = max(0, 9 - prop.CQA);
var CBF3 = max(0, prop.CQA - 1);
var CBF5 = max(0, prop.CQA - 9.5);
var CBF7 = max(0, prop.CQA - 8);
var CBF9 = max(0, prop.CQA - 2);
var CBF11 = max(0, prop.CQA - 0.5);
var CBF13 = max(0, prop.CQA - 7);
var CBF15 = max(0, prop.CQA - 8.5);
var CBF19 = max(0, prop.CQA - 1.5);
var CBF21 = max(0, prop.CQA - 2.5);


// CQA-Residual Model
prop.Contributions.Prop.CQA = 931175 + 225801 * CBF1 - 122123 * CBF2 - 12179.9 * CBF3
+ 1.25007e+006 * CBF5 + 7978.88 * CBF7 - 6093.84 * CBF9
- 71256 * CBF11 + 5889.51 * CBF13 +
16269.2 * CBF15
- 10866.3 * CBF19 - 5559.8 * CBF21;


// Stage I + II Combined to model Estimated Sale Price, with R2 of 0.98+
prop.EstimatedSalePrice = prop.Contributions.Prop.BaseValue // 0 for subject
+ prop.Contributions.Prop.LotSize
+ prop.Contributions.Prop.AreaID
+ prop.Contributions.Prop.LivingArea
+ prop.Contributions.Prop.Baths
+ prop.Contributions.Prop.CQA
+ prop.Contributions.Prop.Garage
+ prop.Contributions.Prop.Bedrooms
+ prop.Contributions.Prop.Stories
+ prop.Contributions.Prop.SaleAge;

salePriceEstimate = prop.EstimatedSalePrice;

if (prop.IsSubject)
{
prop.Display("Property");
Subj.Contributions = Mapper.Map<Contributions>(prop.Contributions);

Subj.Contributions.Display(Subj);
Console.WriteLine("Adjusted SalePrice:" + string.Format("{0:0.00}",
prop.EstimatedSalePrice));

return 0.0;
}

// Calculate the adjustments for the comp

//prop.Adjustments.Prop.Latitude = Subj.Contributions.Prop.Latitude - prop.Contributions.Prop.Latitude;
//prop.Adjustments.Prop.Longitude = Subj.Contributions.Prop.Longitude - prop.Contributions.Prop.Longitude;
prop.Adjustments.Prop.Baths = Subj.Contributions.Prop.Baths - prop.Contributions.Prop.Baths;
prop.Adjustments.Prop.Bedrooms = Subj.Contributions.Prop.Bedrooms - prop.Contributions.Prop.Bedrooms;
prop.Adjustments.Prop.Stories = Subj.Contributions.Prop.Stories - prop.Contributions.Prop.Stories;
prop.Adjustments.Prop.AreaID = Subj.Contributions.Prop.AreaID - prop.Contributions.Prop.AreaID;
prop.Adjustments.Prop.LivingArea = Subj.Contributions.Prop.LivingArea - prop.Contributions.Prop.LivingArea;
prop.Adjustments.Prop.CQA = Subj.Contributions.Prop.CQA - prop.Contributions.Prop.CQA;
prop.Adjustments.Prop.Garage = Subj.Contributions.Prop.Garage - prop.Contributions.Prop.Garage;
prop.Adjustments.Prop.LotSize = Subj.Contributions.Prop.LotSize - prop.Contributions.Prop.LotSize;
prop.Adjustments.Prop.SaleAge = Subj.Contributions.Prop.SaleAge - prop.Contributions.Prop.SaleAge;
// prop.Adjustments.Prop.Age = Subj.Contributions.Prop.Age - prop.Contributions.Prop.Age;
//prop.Adjustments.Prop.Stories = Subj.Contributions.Prop.Stories - prop.Contributions.Prop.Stories;

prop.Adjustments.Display(prop);

// A comparable sale price is now adjusted for differences to the subject
var adjustedSalePrice = prop.SalePrice
// + prop.Adjustments.Prop.Longitude
// + prop.Adjustments.Prop.Latitude
+ prop.Adjustments.Prop.AreaID
+ prop.Adjustments.Prop.LotSize
+ prop.Adjustments.Prop.LivingArea
+ prop.Adjustments.Prop.Bedrooms
+ prop.Adjustments.Prop.Stories
+ prop.Adjustments.Prop.SaleAge
+ prop.Adjustments.Prop.Garage
+ prop.Adjustments.Prop.Baths
// + prop.Adjustments.Prop.Age
+ prop.Adjustments.Prop.CQA;
// + prop.Adjustments.Bedrooms;

prop.Display("Property");
Console.WriteLine("Adjusted Sale Price:" + string.Format("{0:0.00}",
adjustedSalePrice));
Console.WriteLine("");
return adjustedSalePrice;
}
}
}
 

IVCA

Senior Member
Joined
Jun 27, 2017
Professional Status
Certified General Appraiser
State
California
Bert,
not trying to be annoying or spying on you, but

May I ask how many of the "land" features your model takes into account as value impacts to residential properties, other than size?

Do your models adjust for usable gross/net acreage? we have many areas here were minimum acreage is 1-2 acres, so swamps, wetlands, slopes, mountain shade and the like can be issues that should not be glossed over and especially if the driveway is sloped up from the road, down from the road and if it is paved or graveled. It is important to know that a house with a chimney, the chimney will not draft well if an idiot put it on the south facing wall, which, might not seem like an issue in sunny CA, until you are in the woods in PA in an ice storm and the trees have taken out the electric for a week or more.

I'm just wondering if different geographic regions have different environmental factors that impact value, and should not be glossed over for quick math? Naw, I'm not wondering about it, I'm flat telling you that the acceptability of the math only comes with a GSE saying that's what they will accept. Doesn't really matter if the math is "good" or not. Only matters that a GSE says it's "good enough" for them. So, work on getting GSE approvals first, then appraisers will buy it from you.

.


Oh, by the way, I don't do appraisals for GSE!!! (At least, not since 2008.) They can go f--- themselves. As well as most if not all AMCs. And probably most lenders - although I am open to possibilities. No, I think I can avoid all that crap. I have other ways to make income, if it gets down to that. Too many regulations destroy the appraisal process, as well as too many retarded reviewers, underwriters and the like. But, oh heck, the retards and bureaucrats win in the end no matter what anyway. I just like to reduce the size of the problem.

To your suggestion, I don't see any need for GSE approvals. My appraisals are within the parameters of common practice. Other appraisers use regression and many asundry techniques. WIthin a given company such as CBRE, the appraisers use the techniques their "peers" at the company use. But when you jump from one commercial company to another, you see wildly different techniques employed. Yes, DCF is common. But when you look at the complex Excel spreadsheets - you see a lot of difference in techniques in handling all the stuff that goes around DCF. So, this USPAP requirement regarding peer practices does not carry a lot of weight. Are you going to survey appraisers on their exact techniques? Good luck on that. It won't happen. Confidentiality, proprietary technology and such. It won't happen. And finally, if everyone did the same, how would we make progress? My first MAI mentor said that was one of the most important things: To learn to appraise following the procedures that other appraisers use. I thought "How boring!." "How VERY boring." Anyway, CBRE did things a much different way as well as others I worked with. No such thing.

"Peer practice" is a weak argument in most professions and is superseded by many other requirements such as knowledge, good judgement. So, for example, there are a lot of bad doctors, and in between bad and good, and differences in opinion on treatment; "peer common practice" doesn't have much weight in reality. Like you would accept an argument from a doctor: "But that is exactly what Dr. X down the street does," after making you sterile or crippled or grossly disfigured.

I've been to appraisal seminars where appraisers brag about never justifying comps with something like "Doctors don't have to explain why they prescribe certain treatments.".

=======

NOTE: However, I would consider doing a GSE for an existing client after doing a real appraisal at my standard fee, as a service for a loan. I would move the adjustments across and reference the prior appraisal. I don't see any real conflict ... but I would have to check the latest regs.
 
Last edited:

Sid Holderly

Senior Member
Gold Supporting Member
Joined
Jun 16, 2005
Professional Status
Certified Residential Appraiser
State
Indiana
It will work in the cities. In rural areas there just is not enough sales within meaningful parameters to make viable assumptions from. Low sales count = low confidence factors.
Wildly diverse parameters = perverse adjustments. Build an analysis method (spreadsheet or spreadsheet set) that works for the divergent but only comps available. One value per unit for land size across the grid often leads you in the wrong direction. In this location the value per unit of a 2 acre rural parcel with home and a 10 acres woods with home need to be different.
 
Status
Not open for further replies.
Find a Real Estate Appraiser - Enter Zip Code

Copyright © 2000-, AppraisersForum.com, All Rights Reserved
AppraisersForum.com is proudly hosted by the folks at
AppraiserSites.com
Top

AdBlock Detected

We get it, advertisements are annoying!

Sure, ad-blocking software does a great job at blocking ads, but it also blocks useful features of our website. For the best site experience please disable your AdBlocker.

I've Disabled AdBlock
No Thanks