New market analysis changes.

[Non Sequitur]

FB is facebook? Where is that?

Just search the appraiser groups. A guy who sells his own solution and classes copied and pasted a screen shot from a competing solution and is posting "...This is the disastrous result of the misapplication of a regression technique in the wrong circumstance...."

Maybe, maybe not. It's not a debate I'm interested in.

 

[Joe Flacco]

Curious what are some of the advantages of this. If home prices are higher in April than they were in December why does it matter if its seasonal or not if the opinion of market value is for one specific date. They either increased, decreased, or stayed stable, why does it matter if its due to seasonal and/or other factors?

Because with linear adjustments if the trend is 10% YoY, and you are appraising in the Spring using December comps, you would be adjusting those comps 3% when the adjustment to the December comps is 10% with non-linear adjustments.

With linear adjustments if the trend is 10% YoY, and you are appraising in the Winter using Spring comps, you would be adjusting the highs for the year Spring comps 5% when there would be no adjustments to Spring comps with non-linear adjustments.

 

[Trihard]

FHA has done the time adjustment work for you already.

 

[alebrewer]

That's really awful. Unfortunately, even some of the most well-respected MAIs don't understand.

I took George Dell's class and this was the example he used.
View attachment 96374

Jim Amorin posted the following example
View attachment 96375

This is a great example of the nothingburger using higher order/non-linear regressions in the majority of contexts. Ran one on a sample data set. 48 observations - random 'raw' prices. Ran a 3rd order regression as well as a standard linear regression. The following are the results (six comps):



Does choosing one (intricate) method over another (easier, more intuitive) method really result in better analysis?

 

[CGinMN]

Absolutely correct. They don't understand. They don't understand the market and they don't understand the nature of the data they are analyzing.

How are we going to expect GSE appraisers to do a credible high-level analysis when the MAI educators teaching it don't understand it? My interaction with employees at the GSEs tells me many of them don't understand it either. I know we've both said it, but I'll say it again... GSEs seem to be more interested in having appraisers validate their multi variable regression models than they are in encouraging appraisers to think and analyze markets critically. I have a feeling these changes will result in some lenders preferring incorrect but well-documented time adjustments from software (such as those shown in posts 84/85) as opposed to credible but less well-documented time adjustments based on old school methods of analyzing the underlying fundamentals along with sensitivity, pairs, resales, pendings, and agent interviews.

Curious what are some of the advantages of this. If home prices are higher in April than they were in December why does it matter if its seasonal or not if the opinion of market value is for one specific date. They either increased, decreased, or stayed stable, why does it matter if its due to seasonal and/or other factors?

Seasonal fluctuation in prices occurs in a stable market where you would not adjust a December sale to April, despite the trendline moving in that direction. This means using a trendline to determine the adjustment is often going to be incorrect, because the correct adjustment is $0. This is why it's important to understand whether it is seasonal (i.e. compositional) or whether prices are actually changing.

This is a great example of the nothingburger using higher order/non-linear regressions in the majority of contexts. Ran one on a sample data set. 48 observations - random 'raw' prices. Ran a 3rd order regression as well as a standard linear regression. The following are the results (six comps):

View attachment 96385

Does choosing one (intricate) method over another (easier, more intuitive) method really result in better analysis?

I see two different issues. First it whether linear vs. polynomial is better, the second is whether regression captures the true drivers of price change in the first place. Swapping a 3rd‐order polynomial for a linear slope isn't going to produce a much better result, because in both cases the data is unadjusted for property differences. If the data are not controlled for transaction and property differences, then the regression is not a better analysis. It's just a fancy‐looking chart to appease the UW. I'm not trashing regression at all because I use it all the time. But it's not the end all be all, it has weaknesses that if you are gonna use it you should understand.

 

[Fernando]

yes indeed. however, variance between the predominant neighborhood price and OV of the subject inevitably is easy to determine--subject is older or newer than the PV, bigger or smaller, with condition superior or inferior. Alternatively, "...the industry formula to determine the predominant value as based upon 12-month market activity during a year of continually-increasing values inevitably skews the results downwards towards the median value. Variance between the OV and a pending contract price should be relatively easy to define, either by describing factors that affect the former or the latter. by interviewing listing and selling agents if the pricing strategy isn't obvious as based upon published MLS data. Variance between the OV and the previous selling price usually is almost as straightforward, despite the temptation to say what one really feels about the need to compare a tangible value [OV] and the prior selling price--often affected by that appraiser's tendancy to support the contract price even if doing so is a sham [and also an opportunity to market one's services by offering to provide a retrospective appraisal if the client really wants an explanation]. LOL but really!!!

My subject is larger size and significantly over predominant value. I discussed subject as an overimprovement.
Still, client wanted more comments on why the large adjustments given the lack of similar large size comps.
The adjustments were higher than 25%. Is there an unwritten rule that clients don't want loans that are in the high end of the market?

 

[alebrewer]

If the data are not controlled for transaction and property differences, then the regression is not a better analysis.

For Excel/AI/r/MiniTab, I only model sales that are comparable to the subject. IOW - I'm controlling (to some extent) the variability of the data. For instance, the one I posted on was homes built between 2020 and 2024, 1200-1600' GLA, 0.5 acres and less, and geographically constrained. A couple of weeks back, I did an experiment where I adjusted for elements of comparison, then re-ran the regressions. It really didn't change that much that time either.

But to your point - I fully agree. If you're not constraining the market area (e.g. you're modeling all sales in a zip code) and/or you're not controlling for variability in the independent variables, it most definitely could reduce the viability of the results.

 

[CGinMN]

Even if you are constraining the market using quantitative factors (GLA, age, etc), it's the qualitative factors (condition, seller motivation) that produce much of the seasonal variation.

 

[shrubberyvaluation]

FHA has done the time adjustment work for you already.

View attachment 96384

If only it were that simple., well if you want to do a good job anyhow.

 

[shrubberyvaluation]

How are we going to expect GSE appraisers to do a credible high-level analysis when the MAI educators teaching it don't understand it? My interaction with employees at the GSEs tells me many of them don't understand it either. I know we've both said it, but I'll say it again... GSEs seem to be more interested in having appraisers validate their multi variable regression models than they are in encouraging appraisers to think and analyze markets critically. I have a feeling these changes will result in some lenders preferring incorrect but well-documented time adjustments from software (such as those shown in posts 84/85) as opposed to credible but less well-documented time adjustments based on old school methods of analyzing the underlying fundamentals along with sensitivity, pairs, resales, pendings, and agent interviews.

Seasonal fluctuation in prices occurs in a stable market where you would not adjust a December sale to April, despite the trendline moving in that direction. This means using a trendline to determine the adjustment is often going to be incorrect, because the correct adjustment is $0. This is why it's important to understand whether it is seasonal (i.e. compositional) or whether prices are actually changing.

I see two different issues. First it whether linear vs. polynomial is better, the second is whether regression captures the true drivers of price change in the first place. Swapping a 3rd‐order polynomial for a linear slope isn't going to produce a much better result, because in both cases the data is unadjusted for property differences. If the data are not controlled for transaction and property differences, then the regression is not a better analysis. It's just a fancy‐looking chart to appease the UW. I'm not trashing regression at all because I use it all the time. But it's not the end all be all, it has weaknesses that if you are gonna use it you should understand.

Seasonal changes of prices are changes. We are appraising for a specific date. If it happens to be in December it may be lower, if its in April it may be higher.