I've been too busy to publicly express my perspective on the economy lately but several items are making me curious as to how much manipulation the markets can handle before they correct themselves. The future, as presented by the US government and its monetary master, the Federal Reserve, appears quite well, and "market analysts" seem to be predicting a gradual strengthening which supposedly began when the "recession" ended over a year ago.
I have often spoken out against the use of statistical correlation with regard to attempting proof of a concept. Correlation, however, allows visualization of a composite picture of events. Such correlation helps to avoid tunnel vision, where fixation on a particular goal blinds the observer to other inputs. The analysis of correlated events will then provide a view with more depth perspective, and add "color" to a "black and white" picture.
The Fed has embarked on a course of Quantitative Easing (QE). QE is another term for printing money without increasing its backing. It "eases" the ability of member banks to lend money at low interest rates, and consequently, devalues the currency in which the money is issued.
I have a question for the wizards : if interest rates are at a historic low, and yet the lenders are having difficulty interesting businesses in borrowing more (because the businesses appear wary of overextending themselves), what are the banks going to do with the additional currency? Will they, themselves, take the opportunity to purchase additional assets at low interest rates?
A related area is the trend in consumer prices. If the Consumer Price Index has risen for the past 3 months at only fractions of a percent, this is perceived as a sign that inflation of the currency is under control. However, when the index rate for July and August combined was only 0.3 percent, yet the increase in August alone for fruits and vegetables (at the peak of harvest when prices should fall) was 0.4 percent, does this indicate that the composite index is understating the inflation of the core items which affect the largest numbers of consumers?
Another area being watched is the trend in retail sales. One prediction which I saw, credited to JP Morgan Chase, was that retail sales would be up 0.6 percent in September, but if autos are excluded, would be up only 0.4 percent. Question : if retail sales are measured by dollar expenditures, and they roughly parallel the rise in consumer prices, does this indicate that they are in reality flat or even slightly negative?
A third area being considered is the report on business inventories. The prediction (again attributed to JP Morgan Chase) is that the government will announce business inventories up 0.6 percent in August. This has been interpreted to mean that confidence is increasing. In view of the rise in consumer prices, coupled with the increase in retail sales roughly paralleling the price increases, could it also mean that perhaps people have stopped or slowed their buying?
I don't know. I confess to being an amateur in matters of complex macrofinances. Nevertheless, there is something smelly that the newspaper has been wrapped around. Looking at only one part of the data could make me hopeful. Correlating the data makes me skeptical.
Showing posts with label statistics. Show all posts
Showing posts with label statistics. Show all posts
Friday, October 15, 2010
The Rosy Crystal Ball
Saturday, April 10, 2010
Calling All Lollipops?
My famous pessimism (OK, infamous, if you will) with regard to things economic when they just don't make sense got a jolt today with a FOX News article The Dow's up but trades are scarce, worrying bulls which pointed out that the DJIA has been rising while the trading volume has been declining. So I jumped into Yahoo! and pulled a graph:
Yup. The prices are up about 70% from last year, while the number of trades is down about 25%. You would think that if the market is recovering and prices are rising, there would be an increase in the number of players trying to expand their holdings.
Do you smell something burning?
A few commenters on the article want to pin the scam (yes, I do think there is an attempt being made to create a sucker rally) on the Administration, but I hesitate to go that far. Invoking Occam, I would tend to say that some of the major banks and fund managers are behind this phenomenon, hoping to draw broad enough support in the market to allow them to dump their more toxic assets on the unwary.
The NYSE volume today was 4,511,569,000, of which 995,307,699 shares (22%) were Citibank (662,164,923), Ambac Financial (195,367,197), and Bank of America (134,825,884) -- the top 3 issues traded. This makes me deeply suspicious when almost 1/4 of the trades involved stocks which have negative P/E ratios and which are anticipated to pay no dividends. Remember that someone has to sell in order for someone else to buy, and it is the selling that first makes the buying possible; who is dumping these stocks? (I won't bother myself with who might be stupid enough to be buying them.)
Thursday, February 28, 2008
WARNING!! Correlation Does Not Prove Causation!!
Just thought I'd throw that out, since there could be some misconceptions about the data I am presenting and the correlations that appear.
This warning is brought to you courtesy of the on-going Global Warming controversy. Like most controversies involving science, that controversy exists because people forget that correlation does not prove cause. It has now occasioned a lawsuit by an Eskimo village in Alaska, which is suing Exxon for causing the sea level to rise and wash away part of the village.
The danger here is similar to that posed by the Dow Corning silicone implant lawsuit. Lawyers realize that in civil suits, all they need to do is convince the jury that the defendant needs to be punished. The typical jurist has no idea that correlation and causation are two completely different things. American jurisprudence thrives on circumstantial evidence, which is another way of saying that it is based in superstition.
Correlation is created by relationships in statistical data giving a probability that something could or could not occur, and is useless without a statement setting out the level at which the correlation would be considered significant. That statement is based on a purely subjective decision.
Causation is shown when witnesses can prove that a particular action resulted in a particular effect. It is a statement of fact, and has no subjective component.
Unfortunately, the American legal system is intoxicated with the concept of correlation implying causation. From a purely scientific standpoint, circumstantial evidence NEVER proves innocence or guilt beyond the level of confidence predetermined by the trier. Defense in civil suits is primarily a matter of proving one's innocence, with the same kind of intelligence and reasoning as pervaded the Salem Witchcraft Trials.
The Dow case resulted in a judgment against Dow Corning that was so large the company filed for bankruptcy. The lawyers got their fees, the class action members got their piddly awards, and lots of people lost their jobs, because the legal system allows pseudoscience in the courtroom. The fact that it was later shown that the Dow product was not the cause of the problems was simply too bad for common sense and justice.
My data does not imply causation of any kind.
Tuesday, February 26, 2008
Project Results
The hypothesis : there is a correlation between either the Auditor's appraised value (which is not market value) and the foreclosure appraisal value (which is supposed to be at market value), or between the outstanding loan balance and the foreclosure appraisal value.
The procedure : both the Kenmore (North) and Summit Lake neighborhoods were searched via a polygon map search for REO sales which took place in 2007. The sales were cross-checked with the Summit County Common Pleas Court Records and the Summit Count Sheriff's Sales data. The data was entered in an Excel spreadsheet, and included the names of the appraisers used by the Sheriff, and also included pre- and post REO sales data where it was deemed pertinent. The ratios of appraisal:auditor's appraisal for the prior tax year, the appraisal:judgment amount, and the appraisal:REO sales price were calculated, along with the range of ratios, the mean, median, and mode values for the ratios, and one standard deviation from the mean of the ratios.
Unfortunately, before the project began, no confidence level was chosen for rejecting the null hypotheses, namely, that no correlation would be found. The following items must be considered :With these things in mind, the percentage data was tabulated as follows :
- standard practice allows for a variation of +/- 5% between the value opinions provided by two or more individual appraisers before serious questions about the methodolgy or data are raised
- no standard ratio exists for determining the amount of a loan to be made relative to the appraised value (underwriting guidelines which allow 100% financing have made such determinations impossible)
- the REO sales price does not meet the definition of market value as set forth in 12 CFR Part 34, but because all of the REO sales had open market exposure in the MLS, there is some reason to treat those sales as though they approximate the market value for REO sales as a class
Conclusion : The range of the ratios was quite wide, and only the median ratio of the appraisal:judgment amount was below 100%. There does, however, seem to be a very close correlation in the median ratio of the appraisal to the judgment amount (within 2% either direction). It would thus be interesting to know if the appraisers had knowledge of the judgment amount prior to providing their analysis.
Appraisal: Auditor Appraisal: Judgment Appraisal: REO SP Minimum Ratio 56.26 31.36 114.62 Maximum Ratio 165.12 247.81 1500.00 Mean Ratio 107.56 101.96 366.34 Median Ratio 104.12 98.63 276.00 Standard Deviation 15.92 28.29 238.74
The fact that the mean and median ratios for the appraisal to the eventual REO sales price are 366% and 276% respectively is an area of concern. It can be accepted that the appraised value prior to the foreclosure would be higher than the REO sales price, since there would be both effects present regarding the condition of the foreclosed home, and the fact that the REO sale would take place under conditions of duress.
Because it was possible to separate REO sales from non-REO sales in the analysis, there was some other very interesting stuff that came out of this exercise. More Later.
Wednesday, November 21, 2007
Mining for Data - Part 9 : The Finale
We began with the recognition that understanding trends in a real estate market involves data collected not only over a historic time period, but that also the most current data may be several months out of date. We also stated that knowledge of current listings, their marketing histories, and the typical exposure time in the market helps us interpret current market pressures. Such information is important to investors in order for them to make sound financial decisions.
In the second installment, we pointed out that an accurate neighborhood description is critical to maintaining the credibility of an appraisal report that has come under scrutiny. The third installment pointed out some of the general difficulties that would be encountered in trying to nail down an accurate neighborhood description. It also outlined some sources that can be used to help build a good neighborhood description; one that lends itself to gathering sales data on an objective basis and without ambiguity.
Session four dealt with the mechanics of searching the public records section of Realist.com in the CRIS MLS system for the recorded sales and exporting the data to a file for later analysis. Installment five walked us through the process of converting the Realist.com data exports into Excel spreadsheets and formatting the spreadsheet into a more useable tool.
Session six dealt with making certain the data in the spreadsheet fit some basic assumptions, and then presented the mechanics of finding the measures of central tendency for the data. The seventh installment tabulated and compared six years' worth of statistics from Realist.com and the CRIS MLS, for the same neighborhood, and session eight discussed the discrepancies in the two statistical sets.
The sad truth is that all of this has been tedious work. It has been an exercise that would be totally unappreciated (and possibly resented) by the loan officers who order the appraisals, and the findings would possibly be an irritant or inconvenience for the administrative reviewer or underwriter who had to deal with the appraisal report containing them. In most cases, the appraiser doing this kind of analysis would find that if he were searching out and using truly comparable sales for the appraisal of his subject, the neighborhood description would make no difference whatsoever in his opinion of value. Why, then, go to all this trouble?
Value is a subjective matter. The appraiser renders an opinion of value that is uniquely his own. The client then relies on that opinion to make a decision. It must be assumed that the decision is significant enough that the client has required an unbiased opinion of value. The credibility of the opinion rests on the trust that the client places in the appraiser, and in a modern marketplace, it is not usual that the client personally knows and trusts the appraiser. The appraiser's credibility is established in the truthfulness and internal consistency of his appraisal report.
The appraiser may say, "I know this neighborhood. Houses sell for $X,000 - $Z,000 and I believe my subject is worth $Y,000." He may have a well reasoned value opinion -- but what if the rest of his report was "Plucked From Air" to suit the comfort of his client?
We have reached a point in history where irresponsible lending practices, founded in the greed and carelessness of both investors and borrowers, are generating widespread economic tragedy. Homeowners are losing their homes to foreclosure, and investors are losing their hope of long-term gain. Litigation to fix the blame for this tragedy will drag many appraisers into the courtroom to testify in their own defense. The simple fact is that the data existed all along to establish the characteristics of a neighborhood with acceptable accuracy, and yet many appraisers bungled that part of their job.
Even if the appraiser in litigation can show that his comparable sales were valid, and can explain his reasoning for making adjustments to those sales, a clever attorney for the plaintiff can avail himself of the true neighborhood description for the subject property. It will come down to the heart-wrenching plight of a destitute borrower, who has become homeless, against the appraiser who was relied upon to be accurate in his reporting of the facts. Then, if the appraiser has no factual basis for the neighborhood description in his report, even though his value opinion may have been reasonable when viewed against the comparables, his credibility before a jury will have been diminished and he may be seen as a contributor to the plaintiff's woes.
Set against this, taking a bit of time to find the facts and report them, distasteful as they may be to the loan officer or real estate agent who simply wants a quick close to the sale or refinance at "whatever number it takes", is cheap insurance.
In the second installment, we pointed out that an accurate neighborhood description is critical to maintaining the credibility of an appraisal report that has come under scrutiny. The third installment pointed out some of the general difficulties that would be encountered in trying to nail down an accurate neighborhood description. It also outlined some sources that can be used to help build a good neighborhood description; one that lends itself to gathering sales data on an objective basis and without ambiguity.
Session four dealt with the mechanics of searching the public records section of Realist.com in the CRIS MLS system for the recorded sales and exporting the data to a file for later analysis. Installment five walked us through the process of converting the Realist.com data exports into Excel spreadsheets and formatting the spreadsheet into a more useable tool.
Session six dealt with making certain the data in the spreadsheet fit some basic assumptions, and then presented the mechanics of finding the measures of central tendency for the data. The seventh installment tabulated and compared six years' worth of statistics from Realist.com and the CRIS MLS, for the same neighborhood, and session eight discussed the discrepancies in the two statistical sets.
The sad truth is that all of this has been tedious work. It has been an exercise that would be totally unappreciated (and possibly resented) by the loan officers who order the appraisals, and the findings would possibly be an irritant or inconvenience for the administrative reviewer or underwriter who had to deal with the appraisal report containing them. In most cases, the appraiser doing this kind of analysis would find that if he were searching out and using truly comparable sales for the appraisal of his subject, the neighborhood description would make no difference whatsoever in his opinion of value. Why, then, go to all this trouble?
Value is a subjective matter. The appraiser renders an opinion of value that is uniquely his own. The client then relies on that opinion to make a decision. It must be assumed that the decision is significant enough that the client has required an unbiased opinion of value. The credibility of the opinion rests on the trust that the client places in the appraiser, and in a modern marketplace, it is not usual that the client personally knows and trusts the appraiser. The appraiser's credibility is established in the truthfulness and internal consistency of his appraisal report.
The appraiser may say, "I know this neighborhood. Houses sell for $X,000 - $Z,000 and I believe my subject is worth $Y,000." He may have a well reasoned value opinion -- but what if the rest of his report was "Plucked From Air" to suit the comfort of his client?
We have reached a point in history where irresponsible lending practices, founded in the greed and carelessness of both investors and borrowers, are generating widespread economic tragedy. Homeowners are losing their homes to foreclosure, and investors are losing their hope of long-term gain. Litigation to fix the blame for this tragedy will drag many appraisers into the courtroom to testify in their own defense. The simple fact is that the data existed all along to establish the characteristics of a neighborhood with acceptable accuracy, and yet many appraisers bungled that part of their job.
Even if the appraiser in litigation can show that his comparable sales were valid, and can explain his reasoning for making adjustments to those sales, a clever attorney for the plaintiff can avail himself of the true neighborhood description for the subject property. It will come down to the heart-wrenching plight of a destitute borrower, who has become homeless, against the appraiser who was relied upon to be accurate in his reporting of the facts. Then, if the appraiser has no factual basis for the neighborhood description in his report, even though his value opinion may have been reasonable when viewed against the comparables, his credibility before a jury will have been diminished and he may be seen as a contributor to the plaintiff's woes.
Set against this, taking a bit of time to find the facts and report them, distasteful as they may be to the loan officer or real estate agent who simply wants a quick close to the sale or refinance at "whatever number it takes", is cheap insurance.
Monday, November 12, 2007
Mining for Data - Part 8
Looking at the sales statistics from Realist.com, and the statistics from the same neighborhood, over the same time period, from the CRIS MLS, reveals some startling discrepancies. The mean sales price rose from $60,600 in 2001 to $73,000 (rounded) in 2006, according to the Realist.com figures -- a 20% increase. The MLS data, however, shows a decline from a mean sales price of $57,000 in 2001 to $39,300 in 2006, a 32% decrease. The Realist.com data shows an increase in the predominant (mode) sales price from $62,000 to $80,000; the MLS shows a decrease in the predominant price range. How could there be such a discrepancy?
Remember our assumptions from Part 6? We assumed that each transfer shown by Realist.com was the only transfer of the home that year. When we look at the MLS data and cross-check with the Auditor's sales histories, we find that a large number of the sales that are in the Realist data set are "flip" sales -- homes that sold earlier in the year as foreclosure and REO sales, and which were then resold at a much higher price by the investors who initially bought them.
The "flip" sales, for the most part, do not show up in the MLS because of a simple law of economics : real estate agents work on a commission basis and do not take on listings which they cannot sell on the open market. The majority of the "flip" sales are private sales made to buyers who do not understand the market, do not understand the terms of the financing that they are undertaking, and are not getting their money's worth in the deal. You can sometimes see these homes in the MLS as expired listings that failed to sell at their list price and went on to be sold for MORE THAN their open market list price.
If somebody offered to sell a home for $60,000 and nobody would buy it in a 60 or 90 day marketing period, would you pay $75,000 for that house? There seem to be people who are that stupid. There seem to be appraisers who are willing to value the real estate at the contract price regardless what the market as a whole is doing. There seem to be lenders who are willing to finance such deals. There seem to be securities investors who are willing to put such mortgages in the portfolios of 401(k) plans. Would you buy a stock or mutual fund that was based on such securities?
The situation is ready-made for a bookie. It is possible to examine the sales histories of many of these homes, drive by them to examine their physical condition, and then make a bet as to how long it will be before they are in foreclosure again. In fact, as I sorted through the sales data (and you probably saw it as well, if you did this exercise), some of these homes have been foreclosed two to three times in the past six years!! Some of the investors have bought a home from the bank, "flipped" it to a gullible buyer, and within the six year period, bought it back from another bank and "flipped" it again!
Maybe it is time for the Kingston Trio to redo "Tijuana Jail" with new lyrics?
Government statistics for real estate sales generally follow the path of the stats derived from the Realist.com data. To be sure, Realist.com DOES include all the sales, including the foreclosure/REO transactions, in the full report sheet for each property. It is the cumulative sales search that only reports the final price for the year in the search results. That is only logical; there is no practical way to report the multiple sales in a search which has fixed start and end time parameters. It is the negligence of the statistician, however, whether he be a bureaucrat or an appraiser, that is responsible for the misreporting and/or misinterpretation of the data.
Next -- finally -- a recap and conclusion.
Remember our assumptions from Part 6? We assumed that each transfer shown by Realist.com was the only transfer of the home that year. When we look at the MLS data and cross-check with the Auditor's sales histories, we find that a large number of the sales that are in the Realist data set are "flip" sales -- homes that sold earlier in the year as foreclosure and REO sales, and which were then resold at a much higher price by the investors who initially bought them.
The "flip" sales, for the most part, do not show up in the MLS because of a simple law of economics : real estate agents work on a commission basis and do not take on listings which they cannot sell on the open market. The majority of the "flip" sales are private sales made to buyers who do not understand the market, do not understand the terms of the financing that they are undertaking, and are not getting their money's worth in the deal. You can sometimes see these homes in the MLS as expired listings that failed to sell at their list price and went on to be sold for MORE THAN their open market list price.
If somebody offered to sell a home for $60,000 and nobody would buy it in a 60 or 90 day marketing period, would you pay $75,000 for that house? There seem to be people who are that stupid. There seem to be appraisers who are willing to value the real estate at the contract price regardless what the market as a whole is doing. There seem to be lenders who are willing to finance such deals. There seem to be securities investors who are willing to put such mortgages in the portfolios of 401(k) plans. Would you buy a stock or mutual fund that was based on such securities?
The situation is ready-made for a bookie. It is possible to examine the sales histories of many of these homes, drive by them to examine their physical condition, and then make a bet as to how long it will be before they are in foreclosure again. In fact, as I sorted through the sales data (and you probably saw it as well, if you did this exercise), some of these homes have been foreclosed two to three times in the past six years!! Some of the investors have bought a home from the bank, "flipped" it to a gullible buyer, and within the six year period, bought it back from another bank and "flipped" it again!
Maybe it is time for the Kingston Trio to redo "Tijuana Jail" with new lyrics?
Government statistics for real estate sales generally follow the path of the stats derived from the Realist.com data. To be sure, Realist.com DOES include all the sales, including the foreclosure/REO transactions, in the full report sheet for each property. It is the cumulative sales search that only reports the final price for the year in the search results. That is only logical; there is no practical way to report the multiple sales in a search which has fixed start and end time parameters. It is the negligence of the statistician, however, whether he be a bureaucrat or an appraiser, that is responsible for the misreporting and/or misinterpretation of the data.
Next -- finally -- a recap and conclusion.
Wednesday, November 7, 2007
Mining for Data - Part 7
We have gone to a lot of trouble to this point to gather sales data and massage it to provide some statistical picture of how the neighborhood market has been performing. The data from Realist.com can be summarized like this [apology for the massive white space]:
So what does this seem to tell us? It appears that the mean and median sales prices in this neighborhood have increased by roughly 25% over a six year period. This fits the general trend that has been trumpeted through the land with respect to rising property values. The data from the MLS needs to be added to provide some information about average market times :
WAIT A MINUTE!! THAT CAN'T BE RIGHT!!!!
Tune in next time for another exciting adventure with Top-Notch Appraiser, and we will try to make some sense of this. In the meantime, go back and think about those assumptions that had to be made with respect to the data from Realist.com.
| Summary of Data from Realist.com | |||||||
| Year | # Sold | Min SP | Max SP | Mean SP | Median SP | Mode | Std. Dev. |
| 2001 | 39 | $30,000 | $129,000 | $60,619.23 | $59,000 | $62,000 | $17,967.33 |
| 2002 | 42 | $23,000 | $118,500 | $62,476.19 | $59,550 | $35,000 | $20,502.13 |
| 2003 | 56 | $12,003 | $157,500 | $67,249.98 | $67,100 | $55,000 | $23,168.68 |
| 2004 | 84 | $15,000 | $175,000 | $72,363.56 | $75,000 | $80,000 | $30,403.45 |
| 2005 | 71 | $8,900 | $147,700 | $74,739.11 | $75,500 | $75,000 | $27,609.10 |
| 2006 | 89 | $21,525 | $116,500 | $72,970.78 | $75,000 | $80,000 | $19,934.08 |
So what does this seem to tell us? It appears that the mean and median sales prices in this neighborhood have increased by roughly 25% over a six year period. This fits the general trend that has been trumpeted through the land with respect to rising property values. The data from the MLS needs to be added to provide some information about average market times :
| Summary of Data from CRIS MLS | |||||||
| Year | # Sold | Avg DOM | Min SP | Max SP | Mean SP | Median SP | Pred. Price Range |
| 2001 | 42 | 62 | $19,000 | $129,000 | $56,957 | $50,200 | $30,000-$39,999 (8) |
| 2002 | 42 | 86 | $20,000 | $131,000 | $53,146 | $39,500 | $20,000-$29,999 (11) |
| 2003 | 60 | 75 | $7,000 | $157,500 | $42,365 | $31,550 | $30,000-$39.999 (18) |
| 2004 | 66 | 77 | $8,000 | $130,880 | $49,339 | $35,125 | $20,000-$29,999 (19) |
| 2005 | 69 | 83 | $11,000 | $139,900 | $40,506 | $32,000 | $20,000-$39,999 (18) |
| 2006 | 80 | 85 | $9,500 | $109,900 | $39,343 | $30,500 | $20,000-$29,999 (24) |
WAIT A MINUTE!! THAT CAN'T BE RIGHT!!!!
Tune in next time for another exciting adventure with Top-Notch Appraiser, and we will try to make some sense of this. In the meantime, go back and think about those assumptions that had to be made with respect to the data from Realist.com.
Tuesday, November 6, 2007
Mining for Data - Part 6
Spreadsheets were designed with number junkies in mind. They can do all kinds of wizardry with their built-in mathematical functions. Once you learn one of them, all the other spreadsheets are easy to adapt to, since they seem to have all been designed after the pattern of their grand-daddy, VisiCalc. We begin by opening the file "2001 sales data.xls" in the spreadsheet; I am using OpenOffice's Calc.
We have already set up columns A, B, and C to hold the information about Tax ID, Recording Date, and Sales Price, and the rows were ordered by ascending street number and street name. There are some assumptions we have to make at this point if we are not going to be going back and checking the Auditor's web site for every property on the list.
The first assumption is that the recording date shown is the only transfer for that property in that year. In the event the home transferred more than once (e.g., a foreclosure followed by REO sale followed by investor flip), only the final transfer will be shown. The second assumption is that the transfer is an actual arm's-length sale; this cannot be guaranteed since Realist sometimes reports as sales the sale of partial interests, transfers into revocable trusts, land contract agreements, and multi-parcel transfers by investors. It is worth the effort to scan down the list to see any "sales" that do not have a sales price and check the Auditor's web site; if those transfers need to be removed, simply click on the row header to the left of column A and delete the row. A third assumption is that Realist will not always have the entire data set; they sometimes miss a transaction. While the data is good, it will not be perfect.
One other consideration : beginning with the 2003 data, it seems that Realist is reporting a transfer having occurred in that year, but any subsequent transfer shows up with its more recent recording date and sales price. In the event such transfers show up, the Auditor's site must be visited to obtain and extract the correct data for the year under consideration. Example : 931 Bloomfield shows up in the list of 2003 transfers with a 10/11/2007 recording date and sale price of $50,000. That transfer is actually the second transfer in 2007 (a foreclosure sale) following an investor sale 6/21/2007 for $54,350. The 2003 sale occurred on 3/3/2003 for $69,900 (which, incidentally, was a "flip" after a sale 8/19/2002 for $22,500; a sale that did not show up in the 2002 Realist sales search).
An even better example is 438 Beechwood, which sold in foreclosure 1/15/03 for $25,000, as an REO on 1/31/2003 for $18,000, and then as a "flip" on 9/29/2003 for $69,900. To be as consistent with my data as possible, the latter is the information I will use for 2003. What will not show up is the fact that the home at 438 Beechwood was subsequently foreclosed again on 9/22/2005 for $36,000, sold as an REO 3/8/2006 for $24,000, and again as a "flip" on 5/15/2006 for $80,000. Any appraisal done on those sales was required by Federal law (USPAP) to have all prior sales within 36 months disclosed. Was the law broken? Did an underwriter ignore the facts presented? Did a loan officer somewhere alter the appraisal without the appraiser's knowledge? Only a jury can determine if fraud was committed.
Scrolling down to the bottom of my spreadsheet, I find that the final transfer is in row 40. Since row 1 contains the column headings, this means that Realist.com picked up 39 recorded sales in the neighborhood in 2001. In column A, I move to row 42 and type "Min", in row 43 "Max", in row 44 "Mean", in row 45 "Median", in row 46 "Mode", and in row 47 "SD". We will use the spreadsheet's math functions to find the range and measures of central tendency, as well as one standard deviation from the mean.
For those who just got lost, I recommend reading Conceptual Statistics for Beginners by Newman and Newman (1994). Dr. Isadore Newman is the man who opened my eyes to how statistics can be used and misused, and introduced me to the concept of the Null Hypothesis (the Type 1 Error), forever changing my view of how scientific research ought to be conducted and reported. Just a plug for a great teacher.
Moving to column C, row 42, type "=min(c2:c40)". This will find the lowest sales price for the data set in column C from rows 2 through 40. When you press "ENTER", the number $30,000.00 appears. You are now in row 43 of column C. Type "=max(c2:c4)"; the high sales price of $129,000.00 appears when you press "ENTER". In row 44 of column C, type "=average(c2:c40)" and press "ENTER"; the mean sales price for 2001 was $60,619.23. In row 45 type "=median(c2:c40)" to get the median sales price of $59,000.00, and in row 46 type "=mode(c2:c40)" to get what we can take to be the predominant sales price of about $62,000.00. In column C and row 47 type "=stdev(c2:c40)" and the spreadsheet will calculate the amount of one standard deviation from the mean, which comes out as $17,967.33.
This information tells us that the measures of central tendency -- the mean, median, and mode -- are all very close to each other; the average sales price was about $60,600, there were the same number of sales above $59,000 as there were below $59,000, and the most commonly occurring sales price was $62,000. If we were to graph the data we might expect a normal bell curve, and the standard deviation of about $18,000 tells us that 68% of all the sales in this neighborhood in 2001 occurred within the range of roughly $43,000 to $79,000.
An underwriter presented with this kind of information might ask a few more questions before granting a loan for more than the mean neighborhood sales price on a home in this neighborhood. As I mentioned in Part 2 of this series, however, he is not likely to get this kind of information because (1) appraisers have learned that in order to keep appraisal orders coming in, they should not report information that causes the value opinion to vary from the stated neighborhood predominant price by more than 10%, and, (2) nobody is paying them enough to take the trouble to do this kind of analysis; in fact, there is constant pressure to reduce the price for an appraisal and consequently to reduce the quality of the work. Had this kind of information been sought out and presented, we might not now be in the midst of a housing market melt-down.
Next : Summarizing some findings.
We have already set up columns A, B, and C to hold the information about Tax ID, Recording Date, and Sales Price, and the rows were ordered by ascending street number and street name. There are some assumptions we have to make at this point if we are not going to be going back and checking the Auditor's web site for every property on the list.
The first assumption is that the recording date shown is the only transfer for that property in that year. In the event the home transferred more than once (e.g., a foreclosure followed by REO sale followed by investor flip), only the final transfer will be shown. The second assumption is that the transfer is an actual arm's-length sale; this cannot be guaranteed since Realist sometimes reports as sales the sale of partial interests, transfers into revocable trusts, land contract agreements, and multi-parcel transfers by investors. It is worth the effort to scan down the list to see any "sales" that do not have a sales price and check the Auditor's web site; if those transfers need to be removed, simply click on the row header to the left of column A and delete the row. A third assumption is that Realist will not always have the entire data set; they sometimes miss a transaction. While the data is good, it will not be perfect.
One other consideration : beginning with the 2003 data, it seems that Realist is reporting a transfer having occurred in that year, but any subsequent transfer shows up with its more recent recording date and sales price. In the event such transfers show up, the Auditor's site must be visited to obtain and extract the correct data for the year under consideration. Example : 931 Bloomfield shows up in the list of 2003 transfers with a 10/11/2007 recording date and sale price of $50,000. That transfer is actually the second transfer in 2007 (a foreclosure sale) following an investor sale 6/21/2007 for $54,350. The 2003 sale occurred on 3/3/2003 for $69,900 (which, incidentally, was a "flip" after a sale 8/19/2002 for $22,500; a sale that did not show up in the 2002 Realist sales search).
An even better example is 438 Beechwood, which sold in foreclosure 1/15/03 for $25,000, as an REO on 1/31/2003 for $18,000, and then as a "flip" on 9/29/2003 for $69,900. To be as consistent with my data as possible, the latter is the information I will use for 2003. What will not show up is the fact that the home at 438 Beechwood was subsequently foreclosed again on 9/22/2005 for $36,000, sold as an REO 3/8/2006 for $24,000, and again as a "flip" on 5/15/2006 for $80,000. Any appraisal done on those sales was required by Federal law (USPAP) to have all prior sales within 36 months disclosed. Was the law broken? Did an underwriter ignore the facts presented? Did a loan officer somewhere alter the appraisal without the appraiser's knowledge? Only a jury can determine if fraud was committed.
Scrolling down to the bottom of my spreadsheet, I find that the final transfer is in row 40. Since row 1 contains the column headings, this means that Realist.com picked up 39 recorded sales in the neighborhood in 2001. In column A, I move to row 42 and type "Min", in row 43 "Max", in row 44 "Mean", in row 45 "Median", in row 46 "Mode", and in row 47 "SD". We will use the spreadsheet's math functions to find the range and measures of central tendency, as well as one standard deviation from the mean.
For those who just got lost, I recommend reading Conceptual Statistics for Beginners by Newman and Newman (1994). Dr. Isadore Newman is the man who opened my eyes to how statistics can be used and misused, and introduced me to the concept of the Null Hypothesis (the Type 1 Error), forever changing my view of how scientific research ought to be conducted and reported. Just a plug for a great teacher.
Moving to column C, row 42, type "=min(c2:c40)". This will find the lowest sales price for the data set in column C from rows 2 through 40. When you press "ENTER", the number $30,000.00 appears. You are now in row 43 of column C. Type "=max(c2:c4)"; the high sales price of $129,000.00 appears when you press "ENTER". In row 44 of column C, type "=average(c2:c40)" and press "ENTER"; the mean sales price for 2001 was $60,619.23. In row 45 type "=median(c2:c40)" to get the median sales price of $59,000.00, and in row 46 type "=mode(c2:c40)" to get what we can take to be the predominant sales price of about $62,000.00. In column C and row 47 type "=stdev(c2:c40)" and the spreadsheet will calculate the amount of one standard deviation from the mean, which comes out as $17,967.33.
This information tells us that the measures of central tendency -- the mean, median, and mode -- are all very close to each other; the average sales price was about $60,600, there were the same number of sales above $59,000 as there were below $59,000, and the most commonly occurring sales price was $62,000. If we were to graph the data we might expect a normal bell curve, and the standard deviation of about $18,000 tells us that 68% of all the sales in this neighborhood in 2001 occurred within the range of roughly $43,000 to $79,000.
An underwriter presented with this kind of information might ask a few more questions before granting a loan for more than the mean neighborhood sales price on a home in this neighborhood. As I mentioned in Part 2 of this series, however, he is not likely to get this kind of information because (1) appraisers have learned that in order to keep appraisal orders coming in, they should not report information that causes the value opinion to vary from the stated neighborhood predominant price by more than 10%, and, (2) nobody is paying them enough to take the trouble to do this kind of analysis; in fact, there is constant pressure to reduce the price for an appraisal and consequently to reduce the quality of the work. Had this kind of information been sought out and presented, we might not now be in the midst of a housing market melt-down.
Next : Summarizing some findings.
Friday, November 2, 2007
Mining for Data - Part 5
There are a number of ways to manipulate data. Since I am an old hacker from way back -- my first spreadsheet was DynaCalc, running under Microware OS-9 back in 1986 -- I tend to prefer spreadsheets over databases. They may not be as pretty sometimes, but they get the job done. Excel seems to be the spreadsheet of choice, but many people balk at the price of Microsoft Office, which does not normally come bundled with their computer, so an excellent alternative is OpenOffice's Calc application ("scalc"). Calc looks like Excel, behaves like Excel, and it is free.
We have downloaded our data reports in .csv format, so they are comma delimited text files. They also have funny file names, courtesy of Realist, such as "1856407_39153.csv", which can be a pain to keep track of. We need to organize our data. With the right mouse button, click on the .csv file name, and "Open With..." scalc (or Excel, or whatever your spreadsheet is called). Once the file is open, you can see from the data just what kind of fish you have caught. In this case, we see 2001 sales data for the 44302 ZIP Code. I saved that as "2001 44302 raw.xls" in a folder called "Realist Raw Data".
The full report starts out with tax mailing information, which I don't need for anything, so I simply highlight all those columns by holding down the Control ("CTRL") key and clicking on each column header, and then delete. That leaves the Tax ID field in column A. By clicking on the column B header, all of column B is highlighted, and a new column can be inserted. This is done twice, to make two new empty columns. I then go across the top of the spreadsheet, reformatting the column widths to whatever I find comfortable. Because both columns B and C are now empty -- the whole sheet has been shifted right by two columns -- I have room to move the "Recording Date" and "Sales Price" fields.
Click the header cell directly above "Recording Date". The entire column will highlight. Position the cursor over the "Recording Date" cell, press and hold down the mouse button, and drag the entire column to the left, into column B. Do the same thing with the "Sales Price" column, dragging it to column C. You now have the Tax ID number in Column A, the Recording Date in column B, and the Sales Price in column C. This will be handy for doing the stats later. Don't forget to save your spreadsheet early and often.
The 44302 ZIP Code has addresses from several Census Tracts. Click on cell A1 to park the cursor in a handy spot, then from the Data menu, choose Sort. A Sort Criteria window will open. At this point we will sort by Census Tract; the second sort criterion can be set to Carrier Route ( a little trick that will make it easier to fine tune the Census Tract Block Numbers, which are not a provided sort criterion). When the sort is done, we find that we can highlight all the data for Census Tract 5064 and 5065, copy it, and paste it into a new spreadsheet. I named the new spreadsheet "2001 sales data.xls" and saved it in a folder called "Realist Stats".
The file "2001 44302 raw.xls" can now be closed. I repeated the process of creating .xls files from the .csv files for each of the downloaded reports; there were 18 altogether. The files that I had named "2001 44313 raw.xls" and "2001 44320 raw.xls" were manipulated in the same way as I had manipulated "2001 44302 raw.xls", and the data for Census Tracts 5064 Carrier Route C053 and Census Tract 5065 copied from each and pasted into "2001 sales data.xls". At that point, I had a spreadsheet with all the recorded sales for 2001 within the neighborhood bounded north by Amelia, east by West Exchange, south by Copley, and west by Storer.
The whole process was then repeated for each of the years 2002, 2003, 2004, 2005, and 2006. As each "200x sales data.xls" sheet was completed, it was resorted using as the criteria (1) Census Tract", (2) Street Name, and (3) House Number. The data was scanned to make sure that properties located on the northeast side of Exchange Street were not accidentally included, and the files were then ready for a year-by-year analysis of the sales data.
Next : Crunch Time
We have downloaded our data reports in .csv format, so they are comma delimited text files. They also have funny file names, courtesy of Realist, such as "1856407_39153.csv", which can be a pain to keep track of. We need to organize our data. With the right mouse button, click on the .csv file name, and "Open With..." scalc (or Excel, or whatever your spreadsheet is called). Once the file is open, you can see from the data just what kind of fish you have caught. In this case, we see 2001 sales data for the 44302 ZIP Code. I saved that as "2001 44302 raw.xls" in a folder called "Realist Raw Data".
The full report starts out with tax mailing information, which I don't need for anything, so I simply highlight all those columns by holding down the Control ("CTRL") key and clicking on each column header, and then delete. That leaves the Tax ID field in column A. By clicking on the column B header, all of column B is highlighted, and a new column can be inserted. This is done twice, to make two new empty columns. I then go across the top of the spreadsheet, reformatting the column widths to whatever I find comfortable. Because both columns B and C are now empty -- the whole sheet has been shifted right by two columns -- I have room to move the "Recording Date" and "Sales Price" fields.
Click the header cell directly above "Recording Date". The entire column will highlight. Position the cursor over the "Recording Date" cell, press and hold down the mouse button, and drag the entire column to the left, into column B. Do the same thing with the "Sales Price" column, dragging it to column C. You now have the Tax ID number in Column A, the Recording Date in column B, and the Sales Price in column C. This will be handy for doing the stats later. Don't forget to save your spreadsheet early and often.
The 44302 ZIP Code has addresses from several Census Tracts. Click on cell A1 to park the cursor in a handy spot, then from the Data menu, choose Sort. A Sort Criteria window will open. At this point we will sort by Census Tract; the second sort criterion can be set to Carrier Route ( a little trick that will make it easier to fine tune the Census Tract Block Numbers, which are not a provided sort criterion). When the sort is done, we find that we can highlight all the data for Census Tract 5064 and 5065, copy it, and paste it into a new spreadsheet. I named the new spreadsheet "2001 sales data.xls" and saved it in a folder called "Realist Stats".
The file "2001 44302 raw.xls" can now be closed. I repeated the process of creating .xls files from the .csv files for each of the downloaded reports; there were 18 altogether. The files that I had named "2001 44313 raw.xls" and "2001 44320 raw.xls" were manipulated in the same way as I had manipulated "2001 44302 raw.xls", and the data for Census Tracts 5064 Carrier Route C053 and Census Tract 5065 copied from each and pasted into "2001 sales data.xls". At that point, I had a spreadsheet with all the recorded sales for 2001 within the neighborhood bounded north by Amelia, east by West Exchange, south by Copley, and west by Storer.
The whole process was then repeated for each of the years 2002, 2003, 2004, 2005, and 2006. As each "200x sales data.xls" sheet was completed, it was resorted using as the criteria (1) Census Tract", (2) Street Name, and (3) House Number. The data was scanned to make sure that properties located on the northeast side of Exchange Street were not accidentally included, and the files were then ready for a year-by-year analysis of the sales data.
Next : Crunch Time
Wednesday, October 31, 2007
Mining for Data - Part 4
Having decided on a neighborhood -- in this case, the portion of CRIS MLS Market Area 21 that lies east of Storer and south of Amelia, and which just happens to take in Census Tracts 5065 and Tract 5064, Blocks 4 and 5, it is time to collect all the transactions in a given time period. A quick peek at a ZIP Code map of the area reveals that three separate ZIPs are found there : 44302, 44313, and 44320.
We will use Realist.com for the data source, since it is one of the tools supplied to Realtors in the CRIS system. This tool is accessed by clicking the "TAX" button in the main CRIS menu bar. We will do a General Search of the tax data.
The General Search window for Realist.com is used for setting our search parameters. Searches can be done in Realist by ZIP Code and by Municipality. If there are multiple Census Tracts in a given ZIP Code (as there will be with this search), the data can be sorted by Census Tract later as needed.
First, the general search criteria. We are looking for single family market data. The "State Code" portion of the Land Use menu should be set to "Sort Numerically"; this allows you to easily pick LUC 510. Then we choose "Akron City" and ZIP "44302". The recording date fields are now set; we are going to gather all the sales data since January 1, 2001, but in order to do this in a manageable way, we will take the data one year at a time. The starting and ending dates for this search are set at "1/1/2001" and "12/31/2001", and the "SEARCH" button is clicked. In short order the "Property List" records are presented. For my purposes, I prefer the "Single Line View".
Now to extract the information so that we can save it and not worry about having to do this search again. Click the "EXPORT" button. You will be taken to the "Export Builder". Be sure you check "Select All" for the fields to be printed in "Full Records". That will give you more information than you will need (or want), but again, the data can be fine-tuned later.
Set the "Full Records" export choice to "Excel", and click the "EXPORT" button. The Export Status window will soon appear. From there you can go to the Export Manager to download the report, back to the Property List, or back to the General Query for a new search. The exported data will wait in the Export Manager until you are ready to save the report to your hard drive. It will be in the form of a comma-delimited text file, or .csv format.
We will go back to the General Search pane and follow this procedure for ZIP Codes 44313 and 44320 as well. When that information is gathered for 2001, we repeat the process for 2002, 2003, 2004, 2005, and 2006. If we were updating our information for a current appraisal assignment in 2007, we would also draw the information from 1/1/2007 through the current date, but until that information is needed, it is best to pull data sets for complete years. Realist allows CRIS users to download only 5000 records per month under their CRIS account, and this kind of information is too valuable to waste. Of course, you can always buy the data separately from First American, but part of the reason for this tutorial is to teach utilization of existing resources.
Finally, in the Export Manager, we will download each of the reports to the hard drive. We are ready to sort and crunch some numbers.
We will use Realist.com for the data source, since it is one of the tools supplied to Realtors in the CRIS system. This tool is accessed by clicking the "TAX" button in the main CRIS menu bar. We will do a General Search of the tax data.
The General Search window for Realist.com is used for setting our search parameters. Searches can be done in Realist by ZIP Code and by Municipality. If there are multiple Census Tracts in a given ZIP Code (as there will be with this search), the data can be sorted by Census Tract later as needed.
First, the general search criteria. We are looking for single family market data. The "State Code" portion of the Land Use menu should be set to "Sort Numerically"; this allows you to easily pick LUC 510. Then we choose "Akron City" and ZIP "44302". The recording date fields are now set; we are going to gather all the sales data since January 1, 2001, but in order to do this in a manageable way, we will take the data one year at a time. The starting and ending dates for this search are set at "1/1/2001" and "12/31/2001", and the "SEARCH" button is clicked. In short order the "Property List" records are presented. For my purposes, I prefer the "Single Line View".
Now to extract the information so that we can save it and not worry about having to do this search again. Click the "EXPORT" button. You will be taken to the "Export Builder". Be sure you check "Select All" for the fields to be printed in "Full Records". That will give you more information than you will need (or want), but again, the data can be fine-tuned later.
Set the "Full Records" export choice to "Excel", and click the "EXPORT" button. The Export Status window will soon appear. From there you can go to the Export Manager to download the report, back to the Property List, or back to the General Query for a new search. The exported data will wait in the Export Manager until you are ready to save the report to your hard drive. It will be in the form of a comma-delimited text file, or .csv format.
We will go back to the General Search pane and follow this procedure for ZIP Codes 44313 and 44320 as well. When that information is gathered for 2001, we repeat the process for 2002, 2003, 2004, 2005, and 2006. If we were updating our information for a current appraisal assignment in 2007, we would also draw the information from 1/1/2007 through the current date, but until that information is needed, it is best to pull data sets for complete years. Realist allows CRIS users to download only 5000 records per month under their CRIS account, and this kind of information is too valuable to waste. Of course, you can always buy the data separately from First American, but part of the reason for this tutorial is to teach utilization of existing resources.
Finally, in the Export Manager, we will download each of the reports to the hard drive. We are ready to sort and crunch some numbers.
Monday, October 29, 2007
Mining for Data - Part 3
Reporting the market characteristics of a neighborhood is not always an easy matter. There are some states which are non-disclosure states; the actual sales price of a property does not become a public record. Ohio requires disclosure of the final sales price, partly because the counties levy transfer taxes based on the amount of the transaction. To fail to record the actual sales price of a property is a matter of tax fraud, and thus the sales price for any non-exempt transaction is readily available in the public records. There are many exempt transactions, such as some foreclosure sales, which may not show a recorded sales price. It is because of those cases that multiple sources of data are handy.
For the Realtor (and every appraiser should belong to his local Board of Realtors simply to gain access to the MLS), the MLS is the primary source of sales data. It is probably the only reliable source of listing data as well. The MLS has come a long way from the days of paper books and stats charts compiled once a month; today, the appraiser can specifically define a search for a type of property, refine the search, and create meaningful statistical interpretations of the market over varying time periods.
Within the CRIS MLS system used in the Akron area, neighborhood searches can be performed by either specifying the subject's MLS market area, by specifying a radius around the subject, or by a "rubber band" map search using up to ten way-points. I prefer the latter, because the agents, who are responsible for the data being input into the system, sometimes deliberately put the property into an adjacent market area which has slightly better appeal. It is not uncommon, for example, to find homes located in the East Akron (East) area listed as being in Ellet, because they are in the area served by Ellet High School. The "rubber band" search uses GIS information from the public records, and if the neighborhood boundaries have been properly defined to include homes with the most similar age, size, design, and location characteristics, those properties will be included regardless which MLS area the agent indicated.
As an example, let us say that our subject is located on Noble Avenue, north of Copley Road, in what has been designated as the West Akron (South) MLS Area (Area 21). Area 21 is bounded generally on the north by Frank Boulevard, east by West Exchange Street, south by Copley Road, and west by Copley Township. The subject is in a submarket that can be described as primarily average quality two story homes built between WWI and WWII, in the region east of Storer Avenue and south of Amelia Avenue. To be sure, there are some newer Capes and ranches, and some older (and higher construction quality) two story homes in that area, but the predominant housing stock was built in the Roaring 20's (my grandfather built such a home on Noble Avenue).
It just turns out that the submarket described above includes all of Census Tract 5065 and most of Blocks 4 and 5 of Census Tract 5064 (the homes on Amelia and northward are of generally better quality and more like the homes east of Exchange Street in Blocks 1-3 of CT 5064, part of the Northwest Akron MLS area, Area 23). It is a neighborhood marked by a relatively high number of investor-owned, tenant occupied dwellings, and also has had a relatively high percentage of foreclosure and REO sales in the past several years. On the positive side, the city has invested heavily in upgrading the infrastructure in an attempt to increase the overall appeal of the area.
When a "rubber band" search of this region is done, all of the sales in a given time period will fall into the net, and using the different MLS reporting options, the sales data can be sorted so as to include all sales, or to exclude sales which obviously have not occurred at arm's length because they were REO. From there, a statistical report can be generated which will provide the number of sales, the average market time for the final listing period, the low, high, mean, and median sales prices, and a means of determining the predominant sales price range. The sales of homes which were obviously owner-occupied can be identified and studied. Armed with such data, the appraiser now has one way of supporting his statements about the market characteristics of the subject neighborhood.
You can bet your little red booties that the average appraiser has not gone to this trouble, because it is time-intensive, because the typical client doesn't care unless there happens to be a problem with default down the road, and because the results most likely would not make an underwriter comfortable when dealing with a refinance transaction where the opinion of value was high enough to meet the ratios but also high enough to warrant a detailed explanation as to why the appraiser thought the subject was an obvious overimprovement for its neighborhood. In the end the borrowers suffer, the lenders suffer, the neighborhood suffers, and the credibility of the appraisal profession suffers.
For the Realtor (and every appraiser should belong to his local Board of Realtors simply to gain access to the MLS), the MLS is the primary source of sales data. It is probably the only reliable source of listing data as well. The MLS has come a long way from the days of paper books and stats charts compiled once a month; today, the appraiser can specifically define a search for a type of property, refine the search, and create meaningful statistical interpretations of the market over varying time periods.
Within the CRIS MLS system used in the Akron area, neighborhood searches can be performed by either specifying the subject's MLS market area, by specifying a radius around the subject, or by a "rubber band" map search using up to ten way-points. I prefer the latter, because the agents, who are responsible for the data being input into the system, sometimes deliberately put the property into an adjacent market area which has slightly better appeal. It is not uncommon, for example, to find homes located in the East Akron (East) area listed as being in Ellet, because they are in the area served by Ellet High School. The "rubber band" search uses GIS information from the public records, and if the neighborhood boundaries have been properly defined to include homes with the most similar age, size, design, and location characteristics, those properties will be included regardless which MLS area the agent indicated.
As an example, let us say that our subject is located on Noble Avenue, north of Copley Road, in what has been designated as the West Akron (South) MLS Area (Area 21). Area 21 is bounded generally on the north by Frank Boulevard, east by West Exchange Street, south by Copley Road, and west by Copley Township. The subject is in a submarket that can be described as primarily average quality two story homes built between WWI and WWII, in the region east of Storer Avenue and south of Amelia Avenue. To be sure, there are some newer Capes and ranches, and some older (and higher construction quality) two story homes in that area, but the predominant housing stock was built in the Roaring 20's (my grandfather built such a home on Noble Avenue).
It just turns out that the submarket described above includes all of Census Tract 5065 and most of Blocks 4 and 5 of Census Tract 5064 (the homes on Amelia and northward are of generally better quality and more like the homes east of Exchange Street in Blocks 1-3 of CT 5064, part of the Northwest Akron MLS area, Area 23). It is a neighborhood marked by a relatively high number of investor-owned, tenant occupied dwellings, and also has had a relatively high percentage of foreclosure and REO sales in the past several years. On the positive side, the city has invested heavily in upgrading the infrastructure in an attempt to increase the overall appeal of the area.
When a "rubber band" search of this region is done, all of the sales in a given time period will fall into the net, and using the different MLS reporting options, the sales data can be sorted so as to include all sales, or to exclude sales which obviously have not occurred at arm's length because they were REO. From there, a statistical report can be generated which will provide the number of sales, the average market time for the final listing period, the low, high, mean, and median sales prices, and a means of determining the predominant sales price range. The sales of homes which were obviously owner-occupied can be identified and studied. Armed with such data, the appraiser now has one way of supporting his statements about the market characteristics of the subject neighborhood.
You can bet your little red booties that the average appraiser has not gone to this trouble, because it is time-intensive, because the typical client doesn't care unless there happens to be a problem with default down the road, and because the results most likely would not make an underwriter comfortable when dealing with a refinance transaction where the opinion of value was high enough to meet the ratios but also high enough to warrant a detailed explanation as to why the appraiser thought the subject was an obvious overimprovement for its neighborhood. In the end the borrowers suffer, the lenders suffer, the neighborhood suffers, and the credibility of the appraisal profession suffers.
Wednesday, October 24, 2007
Mining for Data - Part 2
A knowledge of the subject property's market area is essential to providing a credible opinion of value for the property. The second section of the 1996 URAR Form and the third section of the 2005 FNMA URAR Form deal with the characteristics of the neighborhood. It is sad but also a bit amusing, when reviewing appraisal reports, to see how frequently appraisers appear to use PFA (Pluck From Air) as their primary data source for neighborhood information. Much of this nonsense is lender-driven, because underwriters set guidelines for neighborhoods and housing that is acceptable for their lending portfolio, and the loan officers then pressure appraisers to make sure the report does not contain information that would cause a loan application to be rejected.
It is uproariously funny to read an appraisal report for a house in the City of Akron (a major metropolitan area), done by an appraiser from Cleveland, which states that a home in central Akron is in a "suburban" neighborhood. This helps allow the appraiser to explain why he used comparables located more than a mile away to support a value opinion that nearby sales of similar homes would simply not support. It is also good for a chuckle to see an appraiser from just about anywhere, (for example -- I have reviewed work from Texas and Montana where the population density was perhaps 10 people per square mile, and even from Wayne County, Ohio, the second largest dairy county in the US in terms of milk production), describe a rural neighborhood as "suburban" simply because there are lenders who shy away from lending in rural areas. Because some lenders require a specific explanation when the subject's value is more than 10% above or below the predominant neighborhood sales price, some appraisers will simply plug in their final value opinion for the subject as the predominant price. Whatever makes the client happy (and generates repeat business), or whatever is fastest (because the pressure is on to provide speedy turnaround) has been the prevailing business model for many appraisers.
There are several places to get neighborhood information, and census.gov is a good one, albeit in 2007 the data collected in 2000 is somewhat old. The appraiser simply needs to take care in disclosing where he got the information, especially if change is taking place in the neighborhood. Land use patterns can be developed by accessing regional planning commissions, or local municipal planners, or, if databases like Metroscan or Realist.com are available, through searches based on land use codes.
The process of developing accurate descriptions of neighborhood characteristics is often tedious, and will consume much time for which the appraiser will not be paid by any client. Lazy appraisers may simply copy from other appraisers' reports without verifying the information, or may simply guess at what they hope will pass scrutiny. The chance that the appraiser will be held accountable for the neighborhood description is very small, since most of the emphasis in review and litigation centers around the description of the subject and the sales comparables. There is, however, that old saying in real estate that the three most important factors are location, location, and location; inaccuracy in describing the neighborhood will result in a deceptive value opinion. Further, if the appraisal is challenged in court, the credibility of the entire report may suffer if the neighborhood data section can be shown to have insupportable information.
The neighborhood boundaries must be well defined. FNMA prefers that they be defined by physical characteristics, such as streets, bodies of water, land uses, or even types of dwellings. Every real state agent knows that buyers are strongly influenced by school system boundaries, and a good example is the Ellet market in Akron. Traditionally, the area served by Ellet High School has been highly regarded, and a portion of the East Akron (East) CRIS MLS market area is within the Ellet district. It can be shown, statistically, that homes in that portion of the market tend to sell for more than homes of similar age, size, and design in the portion of the market served by East High School. The boundaries for the high school districts are readily available from the Akron Public School system.
However, there is a strong possibility that use of a school system boundary can open the appraiser to a charge of violating civil rights laws, since schools are associated with children and children are associated with familial status, which is a protected class. It is fortunate that in the case of the Ellet district, the boundaries also very closely correspond to census tract boundaries, and if those are used as search criteria, there is much less chance of a civil rights violation being charged as long as the demographic data does not enter into the description. Because the Ellet area is so prominent, mention of the school system cannot be avoided, but the use of the school boundaries alone is probably not defensible.
With some data services, the number of search parameters is limited, but the data provided can be further sorted by using spreadsheets and then analyzed via the built in statistical procedures. In the end, however, the appraiser must have a grasp of what is statistically significant and what is not, and that generally means that he must be familiar with his target market area.
Again, the time consumed in building a neighborhood description means that it is best to build a good one and stick with it, regardless whether the subject property "fits" the client's guidelines. It is safer to provide an explanation for a property that is not "typical" than it is to redefine "typical" in terms of the subject simply to provide a quick turnaround time or make an underwriter happy. It may not be good for business -- clients tend to shoot the messenger when they get bad news -- but it is legal and ethical and if your bills can be paid somehow, you will sleep better at night.
It is uproariously funny to read an appraisal report for a house in the City of Akron (a major metropolitan area), done by an appraiser from Cleveland, which states that a home in central Akron is in a "suburban" neighborhood. This helps allow the appraiser to explain why he used comparables located more than a mile away to support a value opinion that nearby sales of similar homes would simply not support. It is also good for a chuckle to see an appraiser from just about anywhere, (for example -- I have reviewed work from Texas and Montana where the population density was perhaps 10 people per square mile, and even from Wayne County, Ohio, the second largest dairy county in the US in terms of milk production), describe a rural neighborhood as "suburban" simply because there are lenders who shy away from lending in rural areas. Because some lenders require a specific explanation when the subject's value is more than 10% above or below the predominant neighborhood sales price, some appraisers will simply plug in their final value opinion for the subject as the predominant price. Whatever makes the client happy (and generates repeat business), or whatever is fastest (because the pressure is on to provide speedy turnaround) has been the prevailing business model for many appraisers.
There are several places to get neighborhood information, and census.gov is a good one, albeit in 2007 the data collected in 2000 is somewhat old. The appraiser simply needs to take care in disclosing where he got the information, especially if change is taking place in the neighborhood. Land use patterns can be developed by accessing regional planning commissions, or local municipal planners, or, if databases like Metroscan or Realist.com are available, through searches based on land use codes.
The process of developing accurate descriptions of neighborhood characteristics is often tedious, and will consume much time for which the appraiser will not be paid by any client. Lazy appraisers may simply copy from other appraisers' reports without verifying the information, or may simply guess at what they hope will pass scrutiny. The chance that the appraiser will be held accountable for the neighborhood description is very small, since most of the emphasis in review and litigation centers around the description of the subject and the sales comparables. There is, however, that old saying in real estate that the three most important factors are location, location, and location; inaccuracy in describing the neighborhood will result in a deceptive value opinion. Further, if the appraisal is challenged in court, the credibility of the entire report may suffer if the neighborhood data section can be shown to have insupportable information.
The neighborhood boundaries must be well defined. FNMA prefers that they be defined by physical characteristics, such as streets, bodies of water, land uses, or even types of dwellings. Every real state agent knows that buyers are strongly influenced by school system boundaries, and a good example is the Ellet market in Akron. Traditionally, the area served by Ellet High School has been highly regarded, and a portion of the East Akron (East) CRIS MLS market area is within the Ellet district. It can be shown, statistically, that homes in that portion of the market tend to sell for more than homes of similar age, size, and design in the portion of the market served by East High School. The boundaries for the high school districts are readily available from the Akron Public School system.
However, there is a strong possibility that use of a school system boundary can open the appraiser to a charge of violating civil rights laws, since schools are associated with children and children are associated with familial status, which is a protected class. It is fortunate that in the case of the Ellet district, the boundaries also very closely correspond to census tract boundaries, and if those are used as search criteria, there is much less chance of a civil rights violation being charged as long as the demographic data does not enter into the description. Because the Ellet area is so prominent, mention of the school system cannot be avoided, but the use of the school boundaries alone is probably not defensible.
With some data services, the number of search parameters is limited, but the data provided can be further sorted by using spreadsheets and then analyzed via the built in statistical procedures. In the end, however, the appraiser must have a grasp of what is statistically significant and what is not, and that generally means that he must be familiar with his target market area.
Again, the time consumed in building a neighborhood description means that it is best to build a good one and stick with it, regardless whether the subject property "fits" the client's guidelines. It is safer to provide an explanation for a property that is not "typical" than it is to redefine "typical" in terms of the subject simply to provide a quick turnaround time or make an underwriter happy. It may not be good for business -- clients tend to shoot the messenger when they get bad news -- but it is legal and ethical and if your bills can be paid somehow, you will sleep better at night.
Monday, October 22, 2007
Mining for Data - Part 1
The appraiser's opinion of the value of a property on a given date is always subject to certain assumptions. If I am an investor trying to maximize my gain through trading of stocks or securities, I want the most up-to-date information about the market, including both the current price quote and some information on trends not only with respect to the price itself, but also the number of shares being traded in a given trading session. If the price is rising with large numbers of shares being traded, it is a different market than one where the price is rising with few shares being traded. Supply and demand always has an impact on price and value.
Real estate sales reports are victims of much longer time lags than stock trading reports. It is for that reason that so many people were caught by surprise when the national real estate market began its decline in late 2005. While technology has allowed us to close the information gap somewhat by speeding up certain processes, the real estate sales cycle does not normally operate very quickly (at least not from the perspective of a stock trader). Let us examine the process, first.
When a real property owner decides to sell his property, he does not simply submit a "sell" order to his broker and wait for the notice to arrive in his e-mail that the sale has taken place. He needs to determine what asking price should be attached to the property, and that entails a valuation of some sort. He may do it himself by looking at advertisements for other properties along with public data about sales of other properties near his. Or, he may hire an appraiser to do the work for him.
Of course, I recommend the latter -- that is my business -- but it is going to cost the property owner about $350 in this area for a good residential appraisal (one that is more than a simple "CMA" done for free by a real estate agent hoping for a listing). In actuality, that price for an appraisal may be a very cheap investment tool, when the time value of money is considered. A $100,000 mortgage for 30 years at 6.5% has a $632/month payment; if the appraisal helps to sell the home only two weeks faster, it has paid for itself in saved house payments. Relocation companies, which buy a house from the employee as an incentive to the move and then hope to resell the home and break even, generally contract for at least two separate appraisals, which must have values within 5% of each other; if they are not, a third appraisal is sometimes ordered. Relocation appraisals are more complex, and rather than providing "market value" provide an anticipated sales price involving a forecast of the next 90-120 days in the market; they are also more expensive (about $500 each) and yet the relocation companies consider them to be important investment tools because they save money in the long run.
There are some limitations to the ordinary property owner doing his own research and valuation. The first is a lack of objectivity; the fact that anything a person owns is automatically going to be seen as being "better" or more valuable than most other properties competing with it. The second is a restricted data pool from which to draw information; a professional should have access to subscription data sources which are much more extensive than those an ordinary consumer can access. A third is general knowledge about market trends, both long and short term, since the most available data is always that which tells of the success of other participants in the market, and downplays their failures and mistakes.
I need to inject a caveat here which may sound like I am up on my soapbox again, but government statistics are notoriously untrustworthy. They tend to be compiled by agencies having a vested interest in the outcome, whether it be the increased ability to collect taxes, or the need to trumpet the success of a government program and ensure the employment of the bureaucrats involved. Further, the expansion of on-line information sources, many compiled by individuals motivated by "causes", tends to increase the possibility that any data obtained could be tainted by prejudicial effects. Care must be exercised in choosing data sources.
The data used in an ordinary appraisal for marketing purposes should be as recent as possible, and if older data is used, sales prices of the comparables may need to be adjusted for the effect of market change over time. Six months is normally considered to be a reasonable parameter for the date of sale, but it should be kept in mind that the closing date alone may not be the only time factor involved. If a person puts his property on the market today, he will do so at what he thinks is a proper asking price. If the property is priced correctly, it can be expected that a certain "exposure time" will elapse before an agreement with a buyer will occur. That information is hard to come by, unless the listing history of the comparable, including all price reductions, is available, because a "reasonable exposure time" may not be the same as the total market time of the property. Further, after a sales agreement is executed, there is usually a delay between the time of agreement and the actual recording of the sale (the closing date). Thus, if a typical marketing time (based on a reasonable exposure period) is 60-90 days, and a typical time to closing is 3-4 weeks, use of a six month old sale without the additional marketing information can give a very deceptive indication of value depending on whether the market is rising, stable, or falling.
I'm going to keep these posts as short as I can, so this novella will be in installments.
Real estate sales reports are victims of much longer time lags than stock trading reports. It is for that reason that so many people were caught by surprise when the national real estate market began its decline in late 2005. While technology has allowed us to close the information gap somewhat by speeding up certain processes, the real estate sales cycle does not normally operate very quickly (at least not from the perspective of a stock trader). Let us examine the process, first.
When a real property owner decides to sell his property, he does not simply submit a "sell" order to his broker and wait for the notice to arrive in his e-mail that the sale has taken place. He needs to determine what asking price should be attached to the property, and that entails a valuation of some sort. He may do it himself by looking at advertisements for other properties along with public data about sales of other properties near his. Or, he may hire an appraiser to do the work for him.
Of course, I recommend the latter -- that is my business -- but it is going to cost the property owner about $350 in this area for a good residential appraisal (one that is more than a simple "CMA" done for free by a real estate agent hoping for a listing). In actuality, that price for an appraisal may be a very cheap investment tool, when the time value of money is considered. A $100,000 mortgage for 30 years at 6.5% has a $632/month payment; if the appraisal helps to sell the home only two weeks faster, it has paid for itself in saved house payments. Relocation companies, which buy a house from the employee as an incentive to the move and then hope to resell the home and break even, generally contract for at least two separate appraisals, which must have values within 5% of each other; if they are not, a third appraisal is sometimes ordered. Relocation appraisals are more complex, and rather than providing "market value" provide an anticipated sales price involving a forecast of the next 90-120 days in the market; they are also more expensive (about $500 each) and yet the relocation companies consider them to be important investment tools because they save money in the long run.
There are some limitations to the ordinary property owner doing his own research and valuation. The first is a lack of objectivity; the fact that anything a person owns is automatically going to be seen as being "better" or more valuable than most other properties competing with it. The second is a restricted data pool from which to draw information; a professional should have access to subscription data sources which are much more extensive than those an ordinary consumer can access. A third is general knowledge about market trends, both long and short term, since the most available data is always that which tells of the success of other participants in the market, and downplays their failures and mistakes.
I need to inject a caveat here which may sound like I am up on my soapbox again, but government statistics are notoriously untrustworthy. They tend to be compiled by agencies having a vested interest in the outcome, whether it be the increased ability to collect taxes, or the need to trumpet the success of a government program and ensure the employment of the bureaucrats involved. Further, the expansion of on-line information sources, many compiled by individuals motivated by "causes", tends to increase the possibility that any data obtained could be tainted by prejudicial effects. Care must be exercised in choosing data sources.
The data used in an ordinary appraisal for marketing purposes should be as recent as possible, and if older data is used, sales prices of the comparables may need to be adjusted for the effect of market change over time. Six months is normally considered to be a reasonable parameter for the date of sale, but it should be kept in mind that the closing date alone may not be the only time factor involved. If a person puts his property on the market today, he will do so at what he thinks is a proper asking price. If the property is priced correctly, it can be expected that a certain "exposure time" will elapse before an agreement with a buyer will occur. That information is hard to come by, unless the listing history of the comparable, including all price reductions, is available, because a "reasonable exposure time" may not be the same as the total market time of the property. Further, after a sales agreement is executed, there is usually a delay between the time of agreement and the actual recording of the sale (the closing date). Thus, if a typical marketing time (based on a reasonable exposure period) is 60-90 days, and a typical time to closing is 3-4 weeks, use of a six month old sale without the additional marketing information can give a very deceptive indication of value depending on whether the market is rising, stable, or falling.
I'm going to keep these posts as short as I can, so this novella will be in installments.
Subscribe to:
Posts (Atom)
