In rexmacey/factorModel: Factor Model Creation and Evaluation

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = ""
)

This vignette introduces the addFactor function which is used to manipulate a set of factors. The RAFI factors have some high correlations. There are also other series we might want to add to the set of factors. The addFactor function helps manipulate an original table of factor data (e.g. the RAFI data) into one we want to use to build models.

Specifically, we may want to:
Add factors representing the U.S., International Developed, and Emerging regions. These would be market-cap weighted indices.
Subtract these market-cap indices from the existing factors to reduce multi-collinearity.

Examples of these operations are showin in this vignette. We will use the dataset representing U.S. factors.

library(dplyr, quietly = TRUE)
library(factorModel, quietly = TRUE)
library(lubridate, quietly = TRUE)
data("factorReturns.US")
data("factorGroups")
head(factorReturns.US)

The following table shows the correlations among the factors. Notice the high correlation between the factors.

plot_corr(cor(factorReturns.US[,2:ncol(factorReturns.US)]))

First we will add a factor representing the Russell 2000 index using the iShares ETF (IWM). After we do that, we will add one for the US market represented by the iShares S&P 500 Index ETF (IVV) and subtract those returns from the other factors. We download prices and calculate monthly returns using the get_monthly_returns function.

Get returns for the two ETFs.

sdate <- min(factorReturns.US$date) %m-% months(1)
edate <- max(factorReturns.US$date)
symbols <- c("IWM", "IVV")
monthly_returns <- get_monthly_returns(symbols = symbols, start_date = sdate, 
                                       end_date = edate)

The get_monthly_returns function created a 3 column table with the symbol, month (date) and Ra the monthly return. Here are the first and last few rows of the table.

head(monthly_returns)
tail(monthly_returns)

Add the Russell 2000 (IWM) ETF series creating a new table. The new table will have the maximum common date range between the factor table and the series being added.

factorUS <- AddFactor(factorReturns.US, 
                      monthly_returns %>% filter(symbol == "IWM") %>% 
                        select(c("date", "Ra")),
                      calcExcess = FALSE)
colnames(factorUS)[ncol(factorUS)] <- "USR2k"
head(factorUS)

Add the S&P 500 (IVV) ETF series while subtracting that return from the other series.

factorUS <- AddFactor(factorUS, 
                      monthly_returns %>% filter(symbol == "IVV") %>% 
                        select(c("date", "Ra")),
                      calcExcess = TRUE)
colnames(factorUS)[ncol(factorUS)] <- "USMkt"
head(factorUS)

Comparing the correlation tables below, we can see the correlations are generally lower except for the two size factors (USSize and USR2k) which makes sense. We wouldn't want both in a regression.

common_date_range <- FindMaxCommonDateRange(factorReturns.US$date, factorUS$date)
cortbl1 <- cor(factorReturns.US %>% filter(date >= common_date_range["start"] & 
                    date <= common_date_range["end"]) %>%
             select(2:ncol(factorReturns.US)))
cortbl2 <- cor(factorUS %>% filter(date >= common_date_range["start"] & 
                    date <= common_date_range["end"]) %>%
             select(2:ncol(factorUS)))

plot_corr(cortbl1)
plot_corr(cortbl2)