inst/Examples/EuStockMarkets.R

In srlanalytics/BDFM: Bayesian and Maximum Likelihood Estimation of Dynamic Factor Models

library(tsbox)
library(bdfm)

# data(EuStockMarkets) #load data
# Y     <- as.matrix(EuStockMarkets) #data in tabular (matrix) format
# dates <- unique(as.Date(ts_data.frame(EuStockMarkets)$time))
#log difference data

#Enter data as ts
Y <- EuStockMarkets
dY    <- 100*diff(log(Y))

#Estimate ML factor model with:
m = 1 #factor
p = 5 #lags
Est_ML   <- dfm(dY,factors = 1,lags = 5, method = "ml", loud = T)

#Estimate PC factor model with:
m = 1 #factor
p = 5 #lags
Est_PC   <- dfm(dY,factors = 1,lags = 5, method = "pc")

#Estimate Bayesian factor model with:
#m = 1 factor
#p = 5 lags
Est   <- dfm(dY,factors = 1,lags = 5, loud = T)

#Look at the common factor for the last 20 observations
ts.plot(cbind(tail(Est$factors,50), tail(dY,50)), col = c("red", "steelblue", "steelblue", "steelblue", "steelblue" ), lty = c(1,2,2,2,2))

#Compare the common factor to the first principal component
pc <- prcomp(dY, center = F, scale. = F)
ts.plot(cbind(tail(Est$factors,50), tail(pc$x[,1],50)), col = c("red", "steelblue"), lty = c(1,2))

#Compare the common factor to "2-step" estimation
pc <- prcomp(dY, center = F, scale. = F)
ts.plot(cbind(tail(Est$factors,50), tail(Est_PC$factors,50)), col = c("red", "steelblue"), lty = c(1,2))

#Compare estimated parameters
print(Est_PC)
print(Est)



#They are (more or less) the same as there are no missing values in this example!

#Re-estimate the factor model shrinking the variance of shocks in the transition equation towards zero (i.e. smooth factors)
Est2  <- dfm(dY, factors = 1, lags = 5, nu_q = 10)
ts.plot(cbind(tail(Est2$factors,50), tail(pc$x[,1],50)), col = c("red", "steelblue"), lty = c(1,2))

Est2$B
Est2$H
Est2$R
Est2$q

# Est3  <- bdfm(dY, factors = 1, lags = 5, ID = "name", nu_q = 0)
# ts.plot(cbind(tail(Est3$factors,50), tail(dY[,1],50)), col = c("red", "steelblue"), lty = c(1,2))

#Enter data as data.frame
library(tsbox)
df <- EuStockMarkets %>%
  ts_tbl()

#Estimate factor model with:
#m = 1 factor
#p = 5 lags
Est <- dfm(df,factors = 1,lags = 5, forecast = 5)

tail(Est$predicted)