Sunday, June 12, 2011

REITs for Everybody Might Now Mean REITs for Nobody

THIS IS MY OPINION AND ANALYSIS AND IS NOT INVESTMENT ADVICE.  YOU ARE RESPONSIBLE FOR YOUR OWN GAINS AND LOSSES.

I think REITs traditionally attract conservative dividend investors (grandparents), but due to their recent behavior, REITs also attract beta chasers (hedge funds  and traders).  This additional demand has made REITs overvalued and even more volatile.  Both sets of REIT buyers might not want to buy now.

In terms of attractiveness to the traditional conservative dividend buyer, REITs seem unattractive on both an absolute and relative yield basis.

From TimelyPortfolio
From TimelyPortfolio

The newly found volatility of REITs for the beta chaser works well in a bullish stock market, but of course works in reverse on the downside.  This volatility also could easily scare the conservative long-term dividend buyer already concerned about low yields.

From TimelyPortfolio
From TimelyPortfolio
From TimelyPortfolio
From TimelyPortfolio
From TimelyPortfolio

One way to potentially dampen volatility would be to use a momentum type system as suggested by The Aleph Blog on REITs How to Make More Returns on REITs.  Results are pretty good.  Like the author, I was skeptical about his approach, since we would have no advance knowledge of momentum quintiles in 1971, but interestingly if we use the Dow Jones Industrial Average momentum quintiles 1896-1971, we would get very similar out of sample REIT results.

From TimelyPortfolio
From TimelyPortfolio

Please let me know what you think.

R code (click to download):

require(quantmod)
require(PerformanceAnalytics)  
#get NAREIT data
#I like NAREIT since I get back to 1971
#much easier though to get Wilshire REIT from FRED
#also it is daily instead of monthly
#getSymbols("WILLREITIND",src="FRED") will do this
require(gdata)
reitURL <- "http://returns.reit.com/returns/MonthlyHistoricalReturns.xls"
reitExcel <- read.xls(reitURL,sheet="Data",pattern="All REITs",stringsAsFactors=FALSE)
#clean up dates so we can use xts functionality later
datetoformat <- reitExcel[,1]
datetoformat <- paste(substr(datetoformat,1,3),"-01-",substr(datetoformat,5,6),sep="")
datetoformat <- as.Date(datetoformat,format="%b-%d-%y")
reitExcel[,1] <- datetoformat    
#############now start the yield analysis#####################
#get REIT yield
reitYield <- as.xts(as.numeric(reitExcel[4:NROW(reitExcel),7]),
order.by=reitExcel[4:NROW(reitExcel),1])
######get BAA and 10y from Fed to compare
getSymbols("BAA",src="FRED")
getSymbols("GS10",src="FRED")
######get SP500 yield from some multpl.com
##fantastic site with easily accessible historical information
spYield <- read.csv("http://www.multpl.com/s-p-500-dividend-yield/s-p-500-dividend-yield.csv")
spYield <- as.xts(spYield[,2],order.by=as.Date(spYield[,1]))
yieldCompare <- na.omit(merge(reitYield,spYield,BAA,GS10))
chart.TimeSeries(yieldCompare, legend.loc = "topleft",cex.legend=1.2,lwd=3,
main="Yield Comparison of REITs with S&P500, BAA Yield, and US 10y Yield",
colorset = c("cadetblue","darkolivegreen3","goldenrod","gray70"))
#get yield spread information
yieldSpread <- yieldCompare[,1:3]
yieldSpread[,1] <- yieldCompare[,1]-yieldCompare[,2]
yieldSpread[,2] <- yieldCompare[,1]-yieldCompare[,3]
yieldSpread[,3] <- yieldCompare[,1]-yieldCompare[,4]
colnames(yieldSpread) <- c("REIT Yield - S&P500 Yield",
"REIT Yield - BAA Yield","REIT Yield - US 10y Yield")
chart.TimeSeries(yieldSpread, legend.loc = "topleft",cex.legend=1.2,lwd=3,
main="Yield Spreads of REITs with S&P500, BAA Yield, and US 10y Yield",
colorset = c("cadetblue","darkolivegreen3","goldenrod"))    
#############now start the return analysis###################
#shift colnames over 1
colnames(reitExcel) <- colnames(reitExcel)[c(1,1:(NCOL(reitExcel)-1))]
#get dates and return columns
reitData <- reitExcel[,c(3,24,38)]
#name columns
colnames(reitData) <- c(paste(colnames(reitExcel)[c(3,24,38)],".Total.Return",sep=""))
reitData <- reitData[3:NROW(reitData),]
#erase commas
col2cvt <- 1:NCOL(reitData)
reitData[,col2cvt] <- lapply(reitData[,col2cvt],function(x){as.numeric(gsub(",", "", x))})
#create xts
reitData <- as.xts(reitData,order.by=reitExcel[3:NROW(reitExcel),1])
#######get sp500 to compare beta and other measures
getSymbols("SP500",src="FRED")
SP500 <- to.monthly(SP500)[,4]
#get 1st of month to align when we merge
index(SP500) <- as.Date(index(SP500))
#merge REIT and S&p
returnCompare <- na.omit(merge(reitData,SP500))
returnCompare <- ROC(returnCompare,n=1,type="discrete")
charts.RollingRegression(returnCompare[, 1:3], returnCompare[,4],
width=36,lwd = 3,legend.loc = "topleft",cex.legend=1.2,
main="NAREIT REIT Indexes Compared to the S&P 500
36 month Rolling"
,
colorset=c("cadetblue","darkolivegreen3","goldenrod"))
chart.RollingPerformance(returnCompare,
FUN="Return.annualized",width=36,lwd = 3,legend.loc = "topleft",cex.legend=1.2,
main="NAREIT REIT Indexes Compared to the S&P 500
36 month Rolling Return"
,
colorset=c("cadetblue","darkolivegreen3","goldenrod","gray70"))
chart.RiskReturnScatter(returnCompare["1971::2003"],
lwd = 3,legend.loc = "topleft",cex.legend=1.2,
main="NAREIT REIT Indexes Compared to the S&P 500 1971-2003",
colorset=c("cadetblue","darkolivegreen3","goldenrod","gray70"))
chart.RiskReturnScatter(returnCompare["2004::"],
lwd = 3,legend.loc = "topleft",cex.legend=1.2,
main="NAREIT REIT Indexes Compared to the S&P 500 Since 2004",
colorset=c("cadetblue","darkolivegreen3","goldenrod","gray70"))
charts.PerformanceSummary(returnCompare,ylog=TRUE,
lwd = 3,legend.loc = "topleft",cex.legend=1.2,
main="NAREIT REIT Indexes Compared to the S&P 500",
colorset=c("cadetblue","darkolivegreen3","goldenrod","gray70"))        
#############now start the bucket analysis###################
#bucket momentum as described by Aleph Blog
#get 10 month moving average
#set up avg with same as reitData
avg <- reitData[,1:3]
avg <- as.data.frame(avg)
avg[,1:3] <- lapply(reitData[,1:3],runMean,n=10)
avg <- as.xts(avg)
#get % above 10 month moving average
momscore <- reitData/avg-1
#break into 5 evenly distributed by frequency quintiles
#get signal into 3 column xts
signal <- momscore
for(i in 1:3) {
breaks <- quantile(momscore[,i], probs = seq(0, 1, 0.20),na.rm=TRUE)
#use default labels=TRUE to see how this works
buckets <- cut(momscore[,i], include.lowest=TRUE, breaks=breaks)
#store so we can see later
ifelse(i==1,bucket_ranges <- names(table(buckets)),
bucket_ranges <- rbind(bucket_ranges,names(table(buckets))))
#now use labels=FALSE to return 1-5 based on quintile
buckets <- cut(momscore[,i], breaks=breaks, labels=FALSE)
signal[,i] <- as.xts(buckets,order.by=index(signal))
#move forward by 1
}
#name bucket_ranges with reit column names
rownames(bucket_ranges)<-colnames(reitData)
signal <- lag(signal,k=1)
ret <- signal
#showing my R weakness here and had to go back to for..next
for(i in 1:3) {
ret[,i] <- ifelse(signal[,i] >= 3,1,0) * ROC(reitData[,1],1,type="discrete")
}
charts.PerformanceSummary(ret,ylog=TRUE,legend.loc = "topleft",cex.legend=1.2,
main="NAREIT REIT Index Data with Aleph Blog Momentum",
colorset=c("cadetblue","darkolivegreen3","goldenrod"))    
getSymbols("DJIA",src="FRED")
#examine DJIA quantiles prior to 1973 to see if we could
#know in advance what possible REIT quantiles would work
DJIA <- to.monthly(DJIA)["1896::1971",4]
momDJIA <- DJIA/runMean(DJIA,n=10)-1
breaks <- quantile(momDJIA, probs = seq(0, 1, 0.20),na.rm=TRUE)
buckets <- cut(momDJIA, breaks=breaks)
table(buckets)  
#what happens if we apply the DJIA prior to 1973 buckets to the REITs
ret <- merge(ret,ret)
for(i in 1:3) {
#if REITs > 3.95% above 10 month moving average then long
#3.95% is the lower end of the DJIA 1896-1971 3 momentum quantile
ret[,i+3] <- lag(ifelse(momscore[,i] >= 0.0395,1,0),1) * ROC(reitData[,1],1,type="discrete")
}
colnames(ret)[4:6]<-paste(colnames(reitData[,1:3])," with DJIA buckets",sep="")
#much much better than I expected
charts.PerformanceSummary(ret,ylog=TRUE,legend.loc = "topleft",cex.legend=1.2,
main="NAREIT REIT Index Data with Aleph Blog Momentum but DJIA Momentum Buckets",
colorset=c("cadetblue","darkolivegreen3","goldenrod",
"coral","darkorchid","darkolivegreen"))

Created by Pretty R at inside-R.org

7 comments:

  1. Subject to all the caveats of my own revised analysis, this is fine. Momentum is pervasive in the markets, with valuation second.

    ReplyDelete
  2. Hello, wondering if you could explain this idea to me a bit better as I am having trouble interpreting the code and can't comment on the Aleph Blog link.

    "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 you take the universe of REITS, sort into quintiles by their momentum (as defined by % above 10 month SMA) and only invest in those REITS which are in the third quintile or higher?

    Except it looks like your equity curve is more like

    "only invest in those REITS which are in the third quintile or higher AND > 10 month SMA".

    Am I right?

    ReplyDelete
  3. Glad you are reading. This system is only built at the index level, and there is no analysis of the components. If the index is in its 40% or better performance quintile as measured by price above the 10 month moving average, then own it and otherwise exit. So, your line should be only invest in the REIT index if its price is greater than the 10 month moving SMA by an amount that is historically in the 40-100% of the distribution.

    ReplyDelete
  4. Ah ok, that's what I originally thought! But then I didn't understand the "historical" bit, is the historical window rolling? Or cumulative since the start of the time-series?

    I see on some quant blogs they use PercentRank function, but usually they apply a 1 year rolling window. e.g.

    PercentRank(SMA(10,MonthClose),12) > 0.6

    So I guess in your case you are saying

    PercentRank(SMA(10,MonthClose),N) > 0.6 where N is the length of the time-series in months?

    Appreciate the response. Love your blog, only found it a few days ago. I noticed you use the utility spread in a post combined with the turbulence thingy, I've done research into the same concept. There are two new ETFs, "SPLV" and "SPHB", I think you will find they provide a better defensive spread.

    ReplyDelete
    Replies
    1. This is not really a tradeable system since it incorporates hindsight bias using the entire history deciding where to trade. The last bit takes the historical level of 3.95% above the 10 month moving average and backtests this on the DJIA. If you were to use this going forward, it would mean you own the REIT index any time it is more than 3.95% above its 10 month moving average.

      You might find http://timelyportfolio.blogspot.com/2011/05/s-500-high-beta-and-low-volatility.html interesting.

      Delete
    2. Right, understood and thanks for the clarification!

      You might find this interesting as a method to adapt the above concept to be tradeable then,

      http://cssanalytics.wordpress.com/2010/05/19/percentrank-sma/

      Haha I see you beat me to it on the LV/HB, nice one.

      Delete
  5. errr, apologies for that pseudocode it should be more like

    PercentRank(MonthClose/SMA(10,MonthClose),N) > 0.6

    ReplyDelete