I'm a data mining developer working for a company that wholesales eviction record data on a national level. I was recently assigned a project to build a program to mine all county courts in Oklahoma for eviction data. I was just curious if this data could be used for investment research purposes and the results of a linear regression kinda stunned me.

It only has a 0.046 R^2 on a weekly timeframe for $ value of evictions vs returns of a REIT Index and 0.16 R^2 on a monthly timeframe and 0.81 R^2 on a yearly for eviction filing counts and returns. When a backtest was built by taking the AR model projected weeks evictions based on previous week(s) eviction, and filtering only the weeks with the top 5% of evictions where recorded (based on OOS data.) The portfolio was rebalanced monthly. The Sharpe ratio was 0.6013 vs 0.4855 on a weekly. 25% higher! WOAH! Why would it do this?

Here is weekly return vs $ value of evictions OLS Results

And Here is monthly return vs eviction filings OLS Results

OLS Results

Any idea behind why more evictions means better returns?

I suspect It may have to do with the fact that evictions can take up to 90 days before a forcible entry and detainer is court ordered which means lost revenue. Perhaps this means that Judgments issued are either payed or new paying tenants are allowed to move in who often pay first months rent, a deposit, and sometimes second months rent. I'm not really an expert on real estate however, It took me 5 months to master all the terminology used in courts and I still feel a little lost.

Just to clarify, my question is:

Any idea behind why more evictions means better returns on REIT Indexes?

PS Pardon my grammar and spelling, I'm kinda dumb and not really sure how I got my job given I dropped out of highschool.

  • $\begingroup$ With such a poor fit (R-Squared), this should be taken with a grain of salt. FYI, correlations can be spurious and meaningless. Check out this link for a few examples of what that means: tylervigen.com/spurious-correlations $\endgroup$
    – amdopt
    Nov 29, 2018 at 21:14
  • $\begingroup$ @correlations You make a good point so I went back to the spreadsheet and sampled annual R^2, It was a value of 0.81! deffinitely not spurious $\endgroup$
    – FX_NINJA
    Nov 29, 2018 at 21:57
  • $\begingroup$ Granted the sample size is abysmally small. Im going to repeat it with more data. It will take a few days to scrape it $\endgroup$
    – FX_NINJA
    Nov 29, 2018 at 22:03


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.