Sensitivity Analysis/Bracketing

[George Hatch]

My theory is that the more complex the appraisal problem is, the fewer critics there are for the solution the appraiser chooses to use. Everybody and their dog has an opinion about how the appraiser appraised that subdivision dogbox. OTOH, you never hear a peep about it when your appraisal problem and the solution it took to appraise it ends up looking like a murder scene.

 

[alebrewer]

So, for me - the sensitivity analysis tool is more than just a means of supporting a set of adjustments. It also helps conceptualize the idea that the goal of an SCA analysis is to minimize the adjusted sales range. For example, say you start with an unadjusted range of ~ $90k. You do your analysis and end up with an adjusted range of ~ $95k. In that scenario, you've completely missed what the market is telling you about how it feels regarding different elements of comparison. Ideally, the adjusted sales range should be $0 - that's the goal. However, this is not practical or possible in most scenarios (with maybe the exception of a very homogenous condo complex or similar). As mentioned previously, the RE market is a very inefficient market, so we'll almost always be left with a range (or if you think in regression terms - a residual). That adjusted range, however, should be MUCH tighter than the unadjusted range.

 

[alebrewer]

This advice is intended for those who wish to remain in blissful ignorance. Accept it and pay the consequences soon. Just ask the California appraisers who now have nothing to protect themselves against the bogus racism claims.

Of course, another option is to continue pulling adjustments from your bodily orifices (whichever you may use). I've found, though, that this method generally ends up putting appraisers at odds with their respective state boards...

 

[Jerry Bone Jr]

using linear regression, y = 7231.89x + 490483 , r^2=.9657

 

[alebrewer]

using linear regression, y = 7231.89x + 490483 , r^2=.9657

Yep - as stated initially, this tool is basically a manual approach to a single variate regression.

 

[glenn walker]

Yep - as stated initially, this tool is basically a manual approach to a single variate regression.

I found regression only-works well in situations where you are using large data pools not on a 6 Comp Appraisal. and-after looking at Berts MARS Regression Flux Capacitors adjustments it made my head spin.

 

[alebrewer]

I found regression only-works well in situations where you are using large data pools not on a 6 Comp Appraisal. and-after looking at Berts MARS Regression Flux Capacitors adjustments it made my head spin.

Regression is just a tool that minimizes the sum of the squares of the differences between the DV and the IV(s). Which is exactly what you're doing with sensitivity analysis - well, you're not minimizing the sum of the squares, but you're minimizing the 'net' adjustments). If you're able to bracket the element of comparison that you're adjusting for, using sensitivity analysis always works - it may not work 'well' (insofar as there may still be a wide range after adjustment), but it will always reduce the adjusted range. It's when you can't bracket the element of comparison that it just doesn't work at all.

 

[George Hatch]

If your most direct comparables aren't responding to an adjustment factor then........