David Merkel submits:
Before I start, I want to offer corrections to my last piece on REITs . Sorry, data glitch, and the results are a little different but the conclusions are unchanged.
Here is the correct regression for Mortgage REITs (click to enlarge) :
And the the correct regression for equity REITs (click to enlarge) :
Why the change? The main difficulty was that the first set of regressions was disaligned time-wise because the twenty-year Treasury yield was not estimated by the Fed for about ten years. This calculation adjusts for that. Interesting that when you use the correct data, mortgage REITs fit less well. That's an odd happenstance. Equity REITs, are around the same - though the t-statistics look worse because I included two more points on the curve.
But now to the point of this piece. After my last piece, one reader said:
This sounds like an interesting scenario to use your grid analysis, where your quantiles might be ranked using (1) equity/mortgage REIT spreads and (2) monetary policy (measured by either short term rates or yield curve slope).
I decided to try it, except for one thing. I thought that using the spreads or yields in quintiles would have too much information about the future that if an investor knew the limits of how high or low yields could go, it would be too simple. The surprise was this: I tried it for fun, and it was no better than what I will show you.
I chose two variables:
- Yield curve slope, measured by twenty-year yield less one-year yield
- Momentum, measured by amount the price is over the ten-month moving average.
I divided both up by quintiles, reasoning that an investor focused on momentum and the yield curve would have enough data from history prior to 1972 to generalize what would be favorable and unfavorable conditions for investment.
Here is what I found for Mortgage REITs (click to enlarge) :
and what I found for Equity REITs (click to enlarge) :
Both data series confirmed one idea. Momentum matters. (Now the yield curve matters, but not so much.) So I constructed a rule to be invested in REITs if they were in the third momentum quintile or higher, or be invested in one year Treasuries otherwise.
So what were the results? Equity REITs (click to enlarge) :
The difference in returns is 20.8%/year following the strategy versus 12.0% not following the strategy.
Now what for Mortgage REITs, that properly despised subindustry group (click to enlarge) ?
The difference in returns is 22.4%/year following the strategy versus 5.1% not following the strategy.
The bizarre result is that though equity REITs trounced mortgage REITs over 38+ years as indexes, the momentum strategy makes mortgage REITs do better then equity REITs. That is probably because the returns on mortgage REITs are more auto-correlated - they streak more.
I just say wow, and wonder at it all. I have some daily models for interest-rate sensitive sectors that I haven't trotted out yet, which switch between mean reversion and mean aversion that do better than this, but I don't believe them because they are too good. It is possible to trip onto something parsimonious that explains history if you are smart enough. I prefer stuff where I have a theory behind it, as with momentum, where people are slow to catch on.
Now, after all of this, I will tell you I have no money riding on this. None. Nada. Null. Part of this is that it would not be so good when implemented in real-time, and mortgage REITs are a thin industry. More of it comes from an aversion to what might happen if/when momentum investing becomes pervasive. We had a taste of that in Summer 2007, where momentum strategies blew up for a short time. For those willing to see it through, though, it did not amount to much.
So long as momentum is an incidental strategy to the market, it can work well. It can never be the main strategy of the market lest matters go totally nuts.
See also 3 High Yield Stocks That Are Likely to Outperform the Market on seekingalpha.com