Appraisers' Free Forum

Increasingly, I see appraisers incorporating charts into their reports. I have slim data bases in my area of the world, but when I have been able to use them i think the charts add to the impact of the report. I understand the overriding problem with statistics is having sufficient data. However, with regression it seems that trends and even averages are fairly apparent and often support what I find in the market otherwise.

Are any of you using charts? If so, what are your experiences?

I use very simple charts in every report. I have plenty of data.

5 years of median sale prices, avg DOM and total sales. I use similar sales filtered to my search criteria. It is an 'overall' image of the market of similar homes over a period of time.

I was going into more detail but I think the simple version works. At times the people reading the report may not be able to understand complicated charts. I put avg DOM and total sales in one chart and the median sale prices in a seperate chart. Very simple to copy and paste from Excel into the report.

Of course, I am at the kindergarten level when compared to Jim P. and many others. I use the KISS approach. I have it down to about 10 - 15 minutes on each report. It certainly helps me ... not sure it helps anyone else - but they can see I did my research on the market.

Have you had any positive feed back from clients, reviewers or other appraisers.

Edd, my ex-business partner is a lover of statistics. He has audited a couple of PHD level stat classes at Akron University...are you ready...for the fun of it. He has been to Denver 3 times to take stat & regression analysis courses that are real estate oriented. Apparently this is a series of 4 courses. He has one to go and plans to go back and take the fourth course, even though he has not appraised a single house in 3 years. He swears by the classes. He attended his first class when he was still appraising, the other two since he quit appraising (but he still retains his license).

If you want details on the classes (contacts, etc.) I will get it from him for you.

I've had positive feed back from other appraisers - most of them want to learn how. I am under the impression that they can sit down and learn it like I did. I'm simply downloading a text file and putting market research into a graph that a 6th grader can do. I'm not doing in depth regressions and statistical analysis. (I'll look to Jim for that one and I'd be totally lost)

I bought Anthony Young's CD and it is really great when you're first learning. I taught myself the basics and expanded with his CD. Mine are very simple Edd - not earth shaking. I figure the people reading the reports have an 8th grade level education and they need to 'see' why I marked the declining box.

http://www.graphthetrend.net/For_Appraisers.php

As far as client feedback - I've had none. But then I have very little human contact with any of my clients. FNMA is one of the biggest now and I've never even talked to a human. Everything is done email and AP.

I think you would have positive feedback if you have local clients and people you can actually talk to. I don't have that luxury.

I've had a few comments from other appraisers at meetings. But I do not tell a lot of people because 'it ain't none of their business'. I tell my close appraiser friends and most of them are so much more advanced than me they probably just think it is a cute idea. I try to associate with people who are a LOT smarter than me hoping it will rub off. Not sure if that is working - but I have people locally I can call and that helps tremendously when I have a question.

Some Realtor friends are completely impressed with the little graphs. Especially those who deal with REO properties. (a BIG business here)

It would be a great marketing tool for someone to analyze the market areas (for several years back) and present it in booklet form to potential clients. Tawfik Ahdab taught me a lot about market analysis and I was doing it the hard way for a long time. Then I started putting it into graph form. It is much easier.

It adds about 10 - 15 mintues to a report for me. Time well spent.

Tel,
Let me know where the courses are.

Rhonda, thanks for the feed back. I have been graphing in the narrative reports I do and I too believe it makes a better report from a readability point, but more importantly I find myself relying more and more on it for my own understanding of the market. The concern I have is that most the scanty data we have here is public (especially if you join the MLS) and I wonder if I am doing stuff that any AVM can do. I have met with a lot of resistance from appraisers around here, but then they ask me for market info from time to time. I have not yet had an assignment that required (or would permit the time and cost) primary market analysis.

Thanks for the info, I too am self taught and always need instruction. I am told by statistical critics that too little data discredits the result, but I figure if its all the data you have using statistical analysis is better than shooting from the hip.

I depend on it also to show me the overall health of the defined market area.

But I deal with 100% REO work and 99.5% of the markets I analyze are in decline. Occasionally I run across one that is increasing in this environment. And that is usually because of one 'weird' sale where the Realtor was pulling some funny stuff. I can spot them a mile away. Everything selling for 10-15k and suddenly there is a sale for 80k. WTH?

In very difficult areas (like you're in) it can help to. You've trimmed down your search criteria to match the subject as much as you can. It gives an overall picture of the market and it will tell you if similar homes are in decline.

It is much more difficult in rural areas. I've done analysis on some areas where I used to live and it was totally screwed up. There would only be a few similar sales when I searched and those ranges were whacky.

I've had negative feed back from a few appraisers - but since I don't tell many people what I do, they probably just think I'm some dumb old blonde struggling to get by. I'll let them keep thinking that as long as they want.

I would love to take some statistics classes. But I don't have time!

Don't worry about AVMs and what they can do. If your clients wanted an AVM - they would order one! The more I study my markets, the more I learn. I can go to some areas and tell you when the year built changes by street, where the 'good' pockets are in a bad area, which side of the subdivision is better, etc. It helps me understand my market areas. And since it only takes a few minutes I do it for myself as well as for my clients.

When I first started graphing I spent hours learning and studying the tri-county area. I started with each county (like NAR does) and saved that information for the future. It was 'down' time and it took many hours but it was a great way to learn! My info has been printed in the local paper a couple of times but I learned quickly that they never print exactly what I say. So I stopped chasing the press. They call occasionally to ask a question but the local board frowns on anyone communicating with the press unless it is their 'chosen' one. And since I'm not from the correct 'bloodline', I'll let the high and mighties deal with the local paper. It is all crap anyway and I don't even take the paper anymore.

I would do it for myself even if I didn't put it in the report. The majority of my areas have lots of sales.

Have you been challenged about validity? I'm thinking ahead to cross examination.

Hey, Rhonda!
You need to blow fifty bucks on AI's new book: An Introduction to Statistics for Appraisers by Marvin L. Wolverton, PhD, MAI. (The rest of you could make good use of this book too.) This guy is both an appraiser and a professional educator. Good example: Why be suspicious of Excel's R-square? Well, if your data are linear, and the QQ plot shows that they're linear, using the Pearson's Product-Moment correlation is accurate. But if your data are non-linear, you should be using the Spearman's Rho calculation to determine the correlation. You square the correlation coefficient to get the R-square. (Which correlation is Excel calculating?) Pearson's will always underestimate correlation for non-linear data. After about three chapters of this book, you'll fully understand what I just wrote--and you'll know why it's true, and you'll be able to write down the formulae on a legal pad and calculate these values by hand.

Buy the book, work through it chapter by chapter, answer the quiz questions at the end of each chapter. Re-read until you've got it. Then move on to the next chapter. And note that Excel isn't totally useless. In fact, the author gives good examples of doing these calculations using Excel. When you've finished, you'll be using Excel functions you've never used before: STDEV, STDEVP, VAR, VARP, finding the coefficient of variation (STDEV/AVERAGE), and on and on and on.

Until I started reading this book, I would gather my data, make a new column for $/sqft GLA, and take the MIN, MAX, MED, and STDEV of these. I would then eliminate any sales whose sale price/sqft was more than one standard deviation from the mean in order to show a general trend. As the book shows, this is incorrect. To eliminate the fringes and just examine the central trend, one should use the interquartile range instead of mean-plus-or-minus-1stdev.

The reason is obvious once you've read the book (or applied a couple of brain cells to the problem). Interquartile range is calculated like a median. To find the median, you sort the prices and pick the middle one. Now take each half of the data and find the median of that. Now you've got four "quarters," called "quartiles" in statistics. Use the data points contained within quartiles 2 and 3, and you have a good representation of the central trend of the data; this is called the "interquartile range." It is defined from the first data point in Quartile 2 to the last data point in Quartile 3.

Here's why that works: Assume your sales are mostly of houses less than $100K. There are several that are over $100K and extending to $350K, but most of them are under. If we take the mean (a.k.a. Average) of these sales, it might fall, for example at $140K with a big STDEV. Subtracting and adding the STDEV to/from the mean is going to give you properties which mostly favor the high side of the range, because of where the mean falls. But if you use the interquartile range, you'll be working with the middle 50% of your data, which is usually what you want when you're trying to discern general trends.

Jim,

Have you cited the book as authority for the use of metrics in any of your reports or testimony? If so how has it been received?

And thanks for the caution about elimination of outliers in the data. I have been doing the same thing or just relying on the excel regression line that generates for scatter graphs. Guess I'll get the book too, I've been using other texts that aren't specifically written for appraisers. I hope he has something in there about what to do with six sales in the last two years. That happens a lot here.

Jim Plante wrote:Hey, Rhonda!
You need to blow fifty bucks on AI's new book: An Introduction to Statistics for Appraisers by Marvin L. Wolverton, PhD, MAI. (The rest of you could make good use of this book too.) This guy is both an appraiser and a professional educator. Good example: Why be suspicious of Excel's R-square? Well, if your data are linear, and the QQ plot shows that they're linear, using the Pearson's Product-Moment correlation is accurate. But if your data are non-linear, you should be using the Spearman's Rho calculation to determine the correlation. You square the correlation coefficient to get the R-square. (Which correlation is Excel calculating?) Pearson's will always underestimate correlation for non-linear data. After about three chapters of this book, you'll fully understand what I just wrote--and you'll know why it's true, and you'll be able to write down the formulae on a legal pad and calculate these values by hand.

Buy the book, work through it chapter by chapter, answer the quiz questions at the end of each chapter. Re-read until you've got it. Then move on to the next chapter. And note that Excel isn't totally useless. In fact, the author gives good examples of doing these calculations using Excel. When you've finished, you'll be using Excel functions you've never used before: STDEV, STDEVP, VAR, VARP, finding the coefficient of variation (STDEV/AVERAGE), and on and on and on.

Until I started reading this book, I would gather my data, make a new column for $/sqft GLA, and take the MIN, MAX, MED, and STDEV of these. I would then eliminate any sales whose sale price/sqft was more than one standard deviation from the mean in order to show a general trend. As the book shows, this is incorrect. To eliminate the fringes and just examine the central trend, one should use the interquartile range instead of mean-plus-or-minus-1stdev.

The reason is obvious once you've read the book (or applied a couple of brain cells to the problem). Interquartile range is calculated like a median. To find the median, you sort the prices and pick the middle one. Now take each half of the data and find the median of that. Now you've got four "quarters," called "quartiles" in statistics. Use the data points contained within quartiles 2 and 3, and you have a good representation of the central trend of the data; this is called the "interquartile range." It is defined from the first data point in Quartile 2 to the last data point in Quartile 3.

Here's why that works: Assume your sales are mostly of houses less than $100K. There are several that are over $100K and extending to $350K, but most of them are under. If we take the mean (a.k.a. Average) of these sales, it might fall, for example at $140K with a big STDEV. Subtracting and adding the STDEV to/from the mean is going to give you properties which mostly favor the high side of the range, because of where the mean falls. But if you use the interquartile range, you'll be working with the middle 50% of your data, which is usually what you want when you're trying to discern general trends.

Say what?

I'll have to read the book to figure out what you just said. Thanks Jim, I'll order it.

Edd - I've never been questioned about validity. Our MLS system spits out similar information - low, high, median, etc. I just download the data to get to it. Mine always differ from theirs because I scrub the data. (when an agent puts 160k instead of 16k) Sometimes there are million dollar properties that are supposed to be 100k properties. Typos or dumb agents - who knows. But I do know that the MLS system does not know when an error has been made and I can spot them in the downloaded data.

Edd,
As far as citation to authority is concerned, the author has impeccable credentials. And the publisher, AI, is no slouch in the credential department either. The main thing you have to worry about with statistics is inadvertent misapplication of a recognized technique, e.g., the example I cited. Using the interquartile range technique will probably change the slope of the regression line (for linear data) slightly, and will result in a lower indication of central tendency for right-skewed data. What right-skewed data is, well, imagine the bell curve we're all familiar with. Now take the right tail of that bell curve and pull it to the right. Scrunch the peak over to the left a little. That's right-skewed data. You can measure the skewness of data, as well as its direction, with Excel's SKEW function. (Refer to the help file in excel.)

Now, understand that if you have a normally distributed data set--one that isn't skewed right or left--using the mean ± STDEV will work just fine. But most of our real estate data is pretty skewed, and usually to the right. You can run MAX, MIN, MEDIAN, AVERAGE, STDEV on a column, and check the skew by comparing the mean to the median. If the data are nearly normally distributed, the median and mean will be close to the same number. Right-skewed data will show a mean greater than the median; left-skewed is exactly the opposite.

it takes twenty times longer to explain something than it does to do it. I started what I thought was a short R tutorial, and the damned thing started turning into a course.

Thanks Jim,

One of my concerns is the number of sales or whatever that are included in any sample. We have limited data always and the preferred method among most here is to shoot from the hip,

Yo Jim,

Your turn.

Okay, okay! I was in the john.

You don't really have "limited" data. What you have is limited *recent* data. In my rural area (county population about 25,000), we have about 250 real residential sales per year in a normal year. That's across the whole spectrum of residential props, from MH's to McMansions. So to define trends in any one sector, you have to go back several years and chart the data by quarter. When I pull the last five years' sales for a given neighborhood, I see clear seasonal trends: Q4 and Q1 are sparse; you'd expect that in winter months. Q2 and Q3 are the most active. Usually Q3 is the most active, but not always. But there doesn't seem to be significant price variation based on seasonality.

On one recent commercial job (which you and I talked about on the phone) I had to go back 15 or more years to get a picture of vacant land prices. To appraise the tough ones, you need to go out in area and back in time in order to get enough data to analyze. Sometimes the data you have to work with are really old, and you have to watch out when you cross a milestone market event like 4Q 2007 (subprime went away), and mention that event in your report. Do you adjust for it? I dunno. It depends on the property and a lot of other things. But it's one more thing you can trip over if you just blindly apply regression or any other technique to the data without controlling for it.

I'm about halfway through that book I mentioned earlier, and it's pretty good at explaining the basics of statistics. In a later chapter, he'll discuss model building. I'm going through the book slowly and carefully, and it's really hard not to jump ahead to see what's coming.