What do u think of this methodology?

Question

Trying to create a pricing model for homes in the tristate area. I haven't calculated the coefficients for the regression- but i was going to start with structure

price of house = a(house sq feet) + b(land acreage) - c(avg train commute to nyc) - d(pv of property taxes)

are there any other factors u think will have significant explanatory power? # of bathrooms, # of bedrooms, age of house, some sorta rating or score for school district quality (think this might be at least partly explained by property tax), etc. any suggestions?

the point of this is two fold- first i figure if i can create a high r2 regression- i can find which neighborhoods are over/undervalued. secondly i can determine how well a house is priced within that market.

i ain't exactly a mathmatician- so all comments are welcome.

was planning to calculate the coefficients by bringing in a bunch of data from mls and using mathlab.

I agree that a single neighborhood model would be easier to model at first- i'll def do it that way.
my thesis is that pricing per sq foot of house/sq mile of land is esentially equal on a tax adjust basis for all neighborhoods- and that premiums/discounts arise b/c of people's preference to have a shorter commute. once i calculate this regression i want to figure out how much people pay for 1 minute of time savings on their commute.

Answer 1

Since the computer is going to do the work, and the linear system will be heavily overdetermined, one thing I'd definitely try is tossing in every factor for which I had data and seeing what comes out. If a factor has no influence on price its coefficient should reflect that (no guarantee that it will, though).

Ultimately you want the simplest model possible, so start with each factor on its own, and then add/swap/remove factors one at a time and see what happens. As you do this try to keep the complexity under control (increasing it one step at a time) so you learn something specific from each run.

First thought I had was that factors such as views won't be accounted for, so I'd expect a neighbourhood-scale model to perform much better. I'd try this both on an individual sale and a neighbourhood basis.

Answer 2

I would think that the data that's listed in the MLS is what most people are looking for. I'd start out using those inputs.

By the way, I just finished a linear models class using the program R. How were you planning on getting those coefficients?

Answer 3

Yahoo answers are not suitable here considering your problem. The matter is not correctly solving an equation, the matter is how to put problem correctly. A professional only can do it. So hire one. Let him put you many a ‘stupid’ question and he will cope with it. An amateur approach may result in big losses.