Investment Statistical Analysis
Involved In Real Estate · Houston, TX · Member since 2008 · 39 posts · 7 votes
Most advice on BP and all other investor groups focuses on individual experience, anecdotal knowledge, and common wisdom. These things are great, and I have the utmost gratitude for all those who share.
Nonetheless, I have a dream of reaching a higher level of knowledge about the investment decisions I make through applying sophisticated statistics to large data sets. Unfortunately, I'm not sophisticated, nor am I a statistician.
I know I'm surely not the first to think along these lines. Can anyone point me in the right direction for more info on the idea?
Any BP Stats PhD's out there that want to collaborate for fun and profit?
What you're proposing is exactly what zillow is trying to do. I do believe this is possible, at least theoretically. This is an application of "super crunching". See Super Crunchers, and The End of Theory. The Watson program that won on Jeopardy is another example. Also The Creative Destruction of Medicine. All these are making arguments for the application of statstical analysis of massive datasets in order to find relationships that a person can't easily spot. "The End of Theory" argues its no longer necessary to apply the scientific method by coming up with a hypothesis and then designing experiments to test the hypothesis. Instead you can test the hypothesis by mining existing data for evidence to support or refute the hypothesis. Since I'm pretty sure nobody here has much involvement with massive numbers of data tapes, I'll also throw out that my day job just released a tape analytics product I designed for performing this same sort of analysis on data tapes and drives to detect potential problems before they result in data loss.
The challenge for what you describe, Matt, is data. It should be relatively trivial to look for a relationship between rents and the type of applicances. If you had the data. Even getting rent data is not easy. I don't report actual rent data to anyone. I suspect many other small landlords don't either. You might get some data from from ads, but I've only used a yard sign for the last couple of years. And there's certainly not much information about the kind of appliances in most houses. At best, someone could look at pictures in ads or read text to identify the appliances. So, the challenge is going to be finding data to use for the analysis.
This might lend itself to some sort of crowdsourcing. That is, you have a way for prospective tenants to report information about what they're seeing when they're looking for a place. You would likely have to offer some small compensation. Similarly for ads. You could have some sort of browser that allows people to put in info about units based on what they see in ads. Even then you may have a data quality issue. Ads are inconsistent and now always even correct. And the people doing the data entry may make mistakes.
I've also been quite curious about this. I'm not a stat PhD, but got my undergrad in stats and am a practicing actuary and programmer, so this is right up my alley.
I'd be willing to collaborate and or provide guidance/assistance from a technical and/or statistical standpoint. The hard part is going to be acquiring good data for a specific area. RE is a very local game with so many variables that can't easily be accounted for in a large scale model.
For what purpose do you want this analysis? What type of properties? Every property and every deal is different.
The large scale commercial property world uses this kind of analysis to choose which markets to invest in, and to forecast supply and demand and therefore rents. My understanding is that companies that supply this information are quite pricey.
The large scale commercial property world uses this kind of analysis to choose which markets to invest in, and to forecast supply and demand and therefore rents. My understanding is that companies that supply this information are quite pricey.
I'm definitely talking about residential here. Yes, every property is different.
Every person is a unique snowflake too, but that sure doesn't stop your insurance company from using a helluva lot of statistics in pricing your premium.
I'll give one, very basic example. This thread (http://www.biggerpockets.com/forums/52/topics/73078-stainless-steel-or-white-appliances-for-rentals ) asks what the increase in rent would be for choosing SS appliances over black ones. I believe in most markets, a real, quantitative answer to this type of question is achievable.
This seems like a pretty basic thing to try and determine, but if you look at it more thoroughly, you see that it's very difficult to place a real value on appliances.
The only way that you could realistically quantify this is by taking a very large apartment complex (100+ units) with apartments that have same or very similar finishes and layouts, other than some have stainless appliances and some have white or black. This sort of situation will be very hard to come by in reality and won't really tell you if you have an advantage to putting stainless in a condo in California or a 5-plex in North Carolina, it will simply tell you what people interested in that particular building are willing to pay for stainless appliances.
I think the best thing to do for a singular investor (not a large company) is to determine what metrics are useful to you and then look at lots and lots of properties so you can get an understanding of the balance of these items. After a decent amount of analysis and practice, it will become natural to look at each item every time you see a property.
I use info that is readily available on MLS listings to automatically weed out a lot of properties when I'm looking for flips. But I've found that regardless of what the numbers say, you cannot get anywhere near the full picture without walking the property. The same is true for apartments. I've seen a lot of properties that look like gold on paper, but when you go into them you find out what the data can't and won't tell you.
If the unit in question had a dataset of 100+ recent sold listings that could be cleanly categorized as SS and non-SS, that would be very helpful indeed, and I would not need a PhD to figure out what SS was worth.
Statistics can be very useful and I have used them in various business scenarios.
I'm not sure how useful they might be in evaluating a potential investment. Perhaps it can prove helpful in your own evaluation, but I see little reason the data you come up with would change many of my own investment decisions. The biggest problem that I see is that several investors can look at a property quite differently. Some may do little with the property and rent it out as it is. Some might make minor or major changes based on their insights and how value and rents might be effected.
I consider the numbers and many other attributes of a property when I make an investment decision. Perhaps its because of where I invest.
There are so many ways to invest and look at a deal. With just a little imagination sometimes, properties that other investors might overlook can be quite profitable. I'm sure that statistical information could help my decision, but many of the underlying assumptions would have to change so that it would be useful in my circumstances.
What you're proposing is exactly what zillow is trying to do. I do believe this is possible, at least theoretically. This is an application of "super crunching". See Super Crunchers, and The End of Theory. The Watson program that won on Jeopardy is another example. Also The Creative Destruction of Medicine. All these are making arguments for the application of statstical analysis of massive datasets in order to find relationships that a person can't easily spot. "The End of Theory" argues its no longer necessary to apply the scientific method by coming up with a hypothesis and then designing experiments to test the hypothesis. Instead you can test the hypothesis by mining existing data for evidence to support or refute the hypothesis. Since I'm pretty sure nobody here has much involvement with massive numbers of data tapes, I'll also throw out that my day job just released a tape analytics product I designed for performing this same sort of analysis on data tapes and drives to detect potential problems before they result in data loss.
The challenge for what you describe, Matt, is data. It should be relatively trivial to look for a relationship between rents and the type of applicances. If you had the data. Even getting rent data is not easy. I don't report actual rent data to anyone. I suspect many other small landlords don't either. You might get some data from from ads, but I've only used a yard sign for the last couple of years. And there's certainly not much information about the kind of appliances in most houses. At best, someone could look at pictures in ads or read text to identify the appliances. So, the challenge is going to be finding data to use for the analysis.
This might lend itself to some sort of crowdsourcing. That is, you have a way for prospective tenants to report information about what they're seeing when they're looking for a place. You would likely have to offer some small compensation. Similarly for ads. You could have some sort of browser that allows people to put in info about units based on what they see in ads. Even then you may have a data quality issue. Ads are inconsistent and now always even correct. And the people doing the data entry may make mistakes.
I have looked at a system marketed on the internet that starts with the premiss that all success and failure in real estate is determined by the specific market an investor chooses to work in. This system tracks data in all the major real estate markets and uses things like population growth or decline and other readily available statistics to forecast where any particular market will go.
The name of the company is Housing Alerts.
It is much like technical analysis in the stock market as far as I can see. The idea is that you choose the market that fits your strategy because success depends more on the market than it does on your particular genius as a real estate investor. Their real estate version of "its the economy stupid" is "its the market stupid".
When I read the OP my initial thought was that such analysis would be impossible due to variables. My statistics classes were a long time ago yet I believe I followed Jon's post (thanks Jon). I can understand how the example given couold be accomplished if there were say 200 similar units in the same complex, but I doubt you could accurately evaluate the subjective nature of variables, such as location when the population would include different cities, cities of different sizes, economic attributes of each area and so on. It seems to me that a program would not be difficult but the data would be endless and unreliable and while I'm not techie, I can't imagine my laptop running such a program.....seems I'd need the computing capacity of the NSA!
Interesting though, I guess I moved from impossible to somewhat probable or highly probable with similar units.
I believe the system mentioned by Michael above is from a Harvard guy by the name of Ken Wade; he is now guruing his info.
The best example of real statistics that we have seen here on BP is the data that supports the "50% rule". A few posts have actually produced links to reports where the data had been collected to support the basic premise of the "50% rule". As Jon mentioned, getting some accurate data is the first step to being able to perform any sort of statistical analysis.
I'm going to respond to the points raised.
1. Markets are too variable, using disparate data will give inaccurate solutions, or the solutions won't be universalizeable.
I don't want or need to find answers that apply to all units in the U.S. I envision developing an analytic framework that can be applied to specific problems. "In this neighborhood, x real estate situation results in y outcome Z% of the time"
2. You'll never get enough data, especially for examining small scale problems.
This is where I'm relying on the stats geniuses. Sure its great to have a ginormous sample size to give you indisputable conclusions, but meaning can appear in very small sample sizes too. Also, I don't think there is any need to know things for certain. If my analysis says 82% of the time, spending this $500 will profit you $1500, I will gleefully make that call time and time again. I'd be more than happy to make decisions based on a confidence interval, but that's where I need the math geeks.
I'm thinking primarily MLS for rent data, but it is clearly a very limited source for that. (I'm also considering questions of value, purchase price, etc. for which MLS would be much better suited.) Finding sources of data is definitely one of the issues worth pondering.
I very much expect this project will involve humans sorting through listings and photos at some point. In our era of globalization and internet, these types of things can be done for cheaper than most of you would think. Cheap enough, in fact, that you could add redundancy, with multiple passes over the same listing to insure accuracy.
3. Math can never replace walking the property in person, "instinct", or experience.
Not advocating that we ever hole up in a basement in front of a computer and stop visiting the properties we buy, but imagine if you could begin to quantify these physical perceptions and experiences, forcing them to yield us even more potent insights. This is another weapon, not a substitute.
@Jon, right on point in terms of a stats revolution. I haven't read some of the references, and I will. The "quant" revolution has come to most fields where money is to be made. Equity/options markets, poker, and apparently baseball are some examples. I have no doubt that the big boys of our industry have been doing what I talk about for years.
If they can do it for cows....
Matt Jankowski - haha... unlike the subject property in question of the RE market, cows are not limited to a localized market to be pertinent. This article/example is an apples to oranges comparison in data sample usage ;)
However, I do like your topic of choice. I think that your data set CAN be smaller IF it is clear the correlation between effect and cause criteria.
For example, the observation:
would most likely be too broad a criteria to conclude a profit of 1500
but ok, I definitely see the value of what you propose. Using statistics of data in the same way advertising & marketing do to determine what/how customers shop is a very plausible premise. In the trend and age of big data, I'm sure we'll see some interesting solutions/services (non-guru-tized) as mentioned here:
http://www.biggerpockets.com/forums/79/topics/71256-ken-wade
Has anyone objectively evaluated Ken Wade's Housing Alerts system in the past 2 years- up to Dec '14? Any responses are greatly appreciated.
Thanks ! Dan Nolan - Investor, Longwood Fl
Hello Michael,
replied over 5 years agoI have looked at a system marketed on the internet that starts with the premiss that all success and failure in real estate is determined by the specific market an investor chooses to work in. This system tracks data in all the major real estate markets and uses things like population growth or decline and other readily available statistics to forecast where any particular market will go.
The name of the company is Housing Alerts.
It is much like technical analysis in the stock market as far as I can see. The idea is that you choose the market that fits your strategy because success depends more on the market than it does on your particular genius as a real estate investor. Their real estate version of "its the economy stupid" is "its the market stupid".
Have you tried this program 'Housing Alerts"? If so, how has it worked for you?
Thanks for any input as I am considering purchasing a subscription.
Joe