Predictive Analytics in Deal Finding?

Question

Hey everyone,

Is anyone else using predictive analytics in their deal-finding efforts? I'm currently attempting to construct a model to predict which off-market properties are likely to list in my county. I'd love to chat about methods and data sources with anyone interested.

Best,

Shane

Trevor Ewen · Answer

@Shane Counts

So here was my process behind the scraper:

- Find single family homes within 20 miles of a military base using non-standard sources (NOT Zillow, Redfin, Trulia, etc.)
- Find the median listing price for a home in that region.
- Filter down to any homes priced 25% or more below the median (grouped by criteria, ie bedrooms and bathrooms).

So, let's use a simple (and fake) example case here:

Search within 20 miles of Ft. Bragg, 3 Bedroom, 2 Bath:

Found $80,000; $85,000; $100,000; $120,000; $130,000; $200,000; $750,000

One of the reasons you shy away from averages in these cases is because of the extreme examples ($750,000). Average ends up being $209,285, but we know that's clearly very high for the area.

25% less than $120,000 (median) gives us a $90,000 threshold, so anything below that gets filtered into the analysis.

From there you get a nice subset you can run additional analysis on (Foreclosures, Short Sales, School Districts, Days on Market, etc.).

As an out-of-state investor, I have to get things at a severe discount before it's even close to profitable for me. Startup and operational costs just run higher in my position, so it's much better to do a broad analysis to see where there is really a strong opportunity in the numbers.

Jason L. · Answer

I don't know what tools you are using, but if you are an R user, there is a new package that was designed to scrape real estate data easily (I think he's scraping Zillow, but it's not written in the documentation). The package is called realtR, and you can check out the documentation here. I am not sure if this can be used on individual property data, but there is a host of macro-level data that may be of use.

Omar Khan · Answer

@Shane Counts I will echo @Sean T. comment. Most publicly available data is reworked or stale. Your key issue would be getting a fresh data feed or a way to extract information from the current available data that other folks aren't already doing.

As @Trevor Ewen has pointed out, in simple terms, what is your end goal - why type of deal? price range? demographics? neighborhood?

From the outside, this area is ripe for disruption but folks keep running into data issues, specifically quality and how real-time it is.

Sean T. · Answer

10+ years in predictive and ML here.

The primary issue with this model is the raw data, quality of data, and access to economic changes in markets that can be hard for a model to work with.

One thought would be to do a basic tribal approach to PA.  Take a data point and look at segmentation and correlation within a market.  I.e. economic data suggests listings are climbing.  What does a regression analysis show happens when listings are climbing and who is likely to list based on past listings.  You could then tie a decision tree in to basically say something like .... if X happens at Y % then Z should follow.

Happy to discuss further but really the key here is raw data from MLS(as example) and economic indicators which are often reworked months after release(causing variance in the model).

Sean

Shane C. · Answer

@Trevor Ewen That's exactly the kind of data-driven approach I was talking about. Thanks for sharing!

@Jason L. I am an R user, that looks like a terrific resource to use, I'll definitely check it out. Thank you!

Omar Khan · Answer

@Shane Counts I haven't tried doing anything similar because we focus on larger commercial properties. IMO, the residential market is driven by emotion, exhibits more volatility and provides less opportunities to scale.

To be honest, I could get similar results by partnering with an investor-friendly agent/wholesaler who knows the ins and outs of a market.

Not saying the project isn't worth looking into, just that I feel there are existing resources that I can leverage in my market better.

Shane C. · Answer

@Omar Khan @Sean T. @Trevor Ewen Thanks for posting!

The (current)  goal is to determine which properties are likely to be listed on the market in my county due to variables like days since last sale, neighborhood, use of property, age of property, etc. which is all available from the county's website. I have a target segment (2-4 family units, $50-$300k, grade C+ through B+ in certain "good neighborhoods"). I have data from 2016-2017 with 575k observations to train and was hoping to use the assessment data from 2018 to test. I'm not sure if there will be any significance to any of the findings, but even if it were 5-10% more accurate than current methods it might be useful. I was going to start with either a logistic regression or naive bayes since most of the variables are categorical and the dependent variable is binary but a decision tree might be the way to go.

I was also thinking of trying to tie in life events (birth, birth +17 years, marriage, divorce, credit card default, tax lien, etc) to the model but there are substantial hurdles: access to data, reliability, scraping and\or text mining, having a dynamic stream, accurately tying the life event to the right person\household, etc.

My actual end goal is to approach deal-finding in a data-driven way, my current thought process on how to achieve that goal may not be correct, and I'm more than willing to adjust, but I'm thinking there must be a more analytical approach than relying on word of mouth or MLS emails.

Trevor, what was your goal when scraping properties near military bases? How did it turn out?

Omar and Sean, have you attempted anything similar or do you have any advice on better methods?

Trevor Ewen · Answer

@Shane Counts

As a technologist, I have dabbled with various scrapers and data finders. My favorite one I did was focused on properties near military bases.

Definitely depends what you're trying to do. Things like that have a short shelf-life because of how various websites change. My issue was that I did not have a specific goal in mind at the time, so I wasn't able to tune it accordingly.

Predictive Analytics in Deal Finding?

Most Popular Reply