Zillow recently announced that it is pulling the plug on its once highly touted house buying machine. Indeed, it was suppose to be a machine, one that used billions of pieces of data to find arbitrage opportunities in the market using algorithms. Buy a house in a surging market, fix it up (quick someone call Flip or Flop or the Property Brothers) and quickly sell the property, with surplus (profit) to boot!
The state of algorithmic purchasing at Zillow is best summarized by the CEO’s own admission. We’ve determined the unpredictability in forecasting home prices far exceeds what we anticipated and continuing to scale Zillow Offers would result in too much earnings and balance-sheet volatility,” Zillow’s CEO, Rich Barton, said in a statement.
There are some important lessons to take away from this.
Lesson 1. Risk is Unpredictable – It is Like the Weather
In my classes on Risk Management, I always mention that predicting normal conditions, like predicting normal weather is not statistically hard or even very rewarding or interesting. Predicting changes and big (unexpected) changes requires something else. When will the next recession happen? This is harder to predict. Zillow thought it had something special in its algorithm. It might know what happened last month, but does it know what will happen next month? It appears not or as the CEO admitted, not well enough. Remember predictions are estimates and are uncertain. The future is unknown and can and does surprise us.
Lesson 2. Real Estate is a Long-term Play
Talk to people and firms, like Real Estate Investment Trusts (REITs). They see an upside in real estate, even the same properties that Zillow might envy. They take a different approach – buy and hold for a while. It is necessary to smooth out the deviations in the economy and account for fluctuations in market conditions. Remember, real estate is a long-term play. Why? The value of real estate is a direct derivative of the salaries in the area. If you want to buy real estate that will go up, go to where salaries are going up and housing stock is low. That is how real estate appreciates. Ideally, you have a market that has salary growth in excess of inflation and the US growth average. Salaries do not generally go up rapidly, and it would be rare to expect real estate to go up rapidly and hold that value if not supported by salaries. The 2020 real estate boom will be tested by a future recession and the ensuing reduction in salaries. If you make big money quickly in real estate, consider yourself lucky – it is like winning the lotto. But ask yourself, have the salaries gone up to support the value in real estate. If the answer is no, then the real estate is overheated.
Lesson 3. Values of Assets can be Sensitive to Small Deviations
Anyone who has ventured into art, collectibles, or even tried to sell sports cards understands this. A small crease and an $100 sports card might be worth 80% less. Houses are like this, too, but in big and small ways. The bus stops in front of the house? That sucks if you don’t like the noise. The power lines run across the back yard making the pool view ugly. Even worse. Such details might be detectable in an algorithm if they were able to consider those and the assignment of value that people put on those. How does one capture the view of the design and layout of a house and then assign value? For instance, in recent years, open floor plans were all the rage. Now, as more people look for separation between stay-at-home family members, the traditional roomed design is proving valuable again. This is subjective, conditional, idiosyncratic, and dynamic. Is it a lot to expect that an algorithm can consider all of that.
Lesson 4. Large Firms are Targets and Not the Same as People (Voters)
When Zillow (or any large firm) buys a house and needs a permit variance, it is not a request from a voting, tax-paying citizen and homeowner, who might write letters to his or her council person. Instead, the permit office sees no reason to be flexible, generous or understanding. Expect more fees, delays, and less flexibility in your quick fixer-upper. Large firms are not people and do not vote in the same way. That is especially true with homes. The solution to this is lack of attention, in Chicago, was always to bring your permit officer some tickets to the Cubs game, because, of course, you are too busy to make it to the game – wink, wink! So, make friends – really important friends.
Lesson 5. Asymmetry of Information
Asymmetry of information won a Nobel Prize in Economics. Of course, Zillow thought that it had that in a pricing algorithm. There is a lot of valuable information in the local market that cannot be easily published or measured. What do the neighbors know about the house and its former owners? What do the permits say about work left undone? Did the inspector offer a thorough and helpful view or simply provide a report that guarantees Zillow hires him or her for the next 100 inspections? Conflicts of interest often reveal asymmetry of information and support it. An owner with long interest in an asset has different goals than an investor with short interest in an asset. Is that asbestos? Who wants to point that out to kill a deal? Are your agents helping you or helping themselves? It is hard to get all of the information on something as complex as a house when you are the owner and even harder when you are employing agents to advise you.
Lesson 6. Principal Agent Problem
It is hard to hire people to behave like owners. They are not owners. They are employees. They act economically like employees, which is different. Corporate money is not the same as personal money. An owner has put his or her savings into an asset and fears loss. An employee is working for a salary, spending the firm’s money and fears losing a job or a chance at a bonus and is willing to spend corporate money to guarantee the job and bonus. This results in different decisions and actions. An owner is incentivized to be efficient with (his or her) capital. An employee is using corporate capital and is disconnected from it. That does not mean that the employee is unethical, but just less connected and incentivized differently. This problem is, as I observed, not easily or ever solved. When employees can and do have large access to capital, be careful, very careful.
Lesson 7. Algorithms Have Limits
Every hedge fund and arbitrage house has a model that prices some assets. Under some conditions, it might work, sometimes. However, markets change. Competitors react. Models then don’t work. The algorithm must evolve. Algorithms produce estimates based on a limited set of inputs (even if that includes thousands of explanatory variables). Remember that estimates are uncertain. I wonder if Zillow calculated the confidence the prediction intervals on its price estimates? Maybe the algorithms showed the limits of use but were ignored.
About Russell Walker, Ph.D.
Professor Russell Walker helps companies develop strategies to manage risk and harness value through analytics and Big Data. He is Associate Teaching Professor of Marketing at the Foster School of Business of the University of Washington. He has worked with many professional sports teams and leading marketing organizations through the Analytics Consulting Lab, an experiential class that he founded and leads at Foster.
#investment, Algorithm, Analytics, Asymmetric Information, Big Data, Big Data Analytics, Data Analytics, economics, featured, housing, housing prices, Permit, Real Estate, REIT, Risk, Risk Management, Zillow