install.packages("zoo")
install.packages("data.table")

library(zoo)
library(data.table)

Installing package into ‘/usr/local/lib/R/site-library’
(as ‘lib’ is unspecified)

Installing package into ‘/usr/local/lib/R/site-library’
(as ‘lib’ is unspecified)


Attaching package: ‘zoo’


The following objects are masked from ‘package:base’:

    as.Date, as.Date.numeric



# 41 page
require(zoo)
require(data.table)

unemp <- fread("https://raw.githubusercontent.com/PracticalTimeSeriesAnalysis/BookRepo/master/Ch02/data/UNRATE.csv")
unemp[, DATE := as.Date(DATE)]
setkey(unemp, DATE)

rand.unemp.idx <- sample(1:nrow(unemp), .1*nrow(unemp))
rand.unemp <- unemp[-rand.unemp.idx]

high.unemp.idx <- which(unemp$UNRATE > 8)
num.to.select <- .2 * length(high.unemp.idx)
high.unemp.idx <- sample(high.unemp.idx,)
bias.unemp <- unemp[-high.unemp.idx]

# 43 page
all.dates <- seq(from=unemp$DATE[1], to=tail(unemp$DATE, 1), by="months")
rand.unemp = rand.unemp[J(all.dates), roll=0]
bias.unemp = bias.unemp[J(all.dates), roll=0]
rand.unemp[, rpt := is.na(UNRATE)]

rand.unemp[, impute.ff := na.locf(UNRATE, na.rm=FALSE)]
bias.unemp[, impute.ff := na.locf(UNRATE, na.rm=FALSE)]

unemp[350:400, plot(DATE, UNRATE, col=1, lwd=2, type='b')]
rand.unemp[350:400, lines(DATE, impute.ff, col=2, lwd=2, lty=2)]
rand.unemp[350:400][rpt==TRUE, points(DATE, impute.ff, col=2, pch=6, cex=2)]

NULL
NULL
NULL
# 47 page
rand.unemp[, impute.rm.nolookahead := rollapply(c(NA, NA, UNRATE), 3, 
            function(x) {
              if (!is.na(x[3])) x[3] else mean(x, na.rm=TRUE)
            })]
bias.unemp[, impute.rm.nolookahead := rollapply(c(NA, NA, UNRATE), 3, 
            function(x) {
              if (!is.na(x[3])) x[3] else mean(x, na.rm=TRUE)
            })]

rand.unemp[, complete.rm := rollapply(c(NA, UNRATE, NA), 3,
            function(x) {
              if (!is.na(x[2])) x[2] else mean(x, na.rm=TRUE)
            })]

# page 49
rand.unemp[, impute.li := na.approx(UNRATE)]
bias.unemp[, impute.li := na.approx(UNRATE)]

rand.unemp[, impute.sp := na.spline(UNRATE)]
bias.unemp[, impute.sp := na.spline(UNRATE)]

use.idx = 90:120
unemp[use.idx, plot(DATE, UNRATE, col = 1, type = 'b')]
rand.unemp[use.idx, lines(DATE, impute.li, col = 2, lwd = 2, lty = 2)]
rand.unemp[use.idx, lines(DATE, impute.sp, col = 3, lwd = 2, lty = 3)]

NULL
NULL
NULL
# 51 page
sort(rand.unemp[ , lapply(.SD, function(x) mean((x - unemp$UNRATE)^2, na.rm = TRUE)),
    .SDcols = c("impute.ff", "impute.rm.nolookahead", "impute.li", "impute.sp")])
A data.table: 1 × 4
impute.liimpute.spimpute.rm.nolookaheadimpute.ff
<dbl><dbl><dbl><dbl>
0.0025055950.0038388790.0066943130.00686019
 
sort(bias.unemp[ , lapply(.SD, function(x) mean((x - unemp$UNRATE)^2, na.rm = TRUE)),
    .SDcols = c("impute.ff", "impute.rm.nolookahead", "impute.li", "impute.sp")])
A data.table: 1 × 4
impute.rm.nolookaheadimpute.spimpute.liimpute.ff
<dbl><dbl><dbl><dbl>
0.0055721720.023234850.21635620.2318483
 
# 53 page
unemp[seq.int(from=1, to=nrow(unemp), by=12)]
A data.table: 71 × 2
DATEUNRATE
<date><dbl>
1948-01-013.4
1949-01-014.3
1950-01-016.5
1951-01-013.7
1952-01-013.2
1953-01-012.9
1954-01-014.9
1955-01-014.9
1956-01-014.0
1957-01-014.2
1958-01-015.8
1959-01-016.0
1960-01-015.2
1961-01-016.6
1962-01-015.8
1963-01-015.7
1964-01-015.6
1965-01-014.9
1966-01-014.0
1967-01-013.9
1968-01-013.7
1969-01-013.4
1970-01-013.9
1971-01-015.9
1972-01-015.8
1973-01-014.9
1974-01-015.1
1975-01-018.1
1976-01-017.9
1977-01-017.5
⋮⋮
1989-01-015.4
1990-01-015.4
1991-01-016.4
1992-01-017.3
1993-01-017.3
1994-01-016.6
1995-01-015.6
1996-01-015.6
1997-01-015.3
1998-01-014.6
1999-01-014.3
2000-01-014.0
2001-01-014.2
2002-01-015.7
2003-01-015.8
2004-01-015.7
2005-01-015.3
2006-01-014.7
2007-01-014.6
2008-01-015.0
2009-01-017.8
2010-01-019.8
2011-01-019.1
2012-01-018.3
2013-01-018.0
2014-01-016.6
2015-01-015.7
2016-01-014.9
2017-01-014.8
2018-01-014.1
 
unemp[, mean(UNRATE), by=format(DATE, "%Y")]
A data.table: 71 × 2
formatV1
<chr><dbl>
19483.750000
19496.050000
19505.208333
19513.283333
19523.025000
19532.925000
19545.591667
19554.366667
19564.125000
19574.300000
19586.841667
19595.450000
19605.541667
19616.691667
19625.566667
19635.641667
19645.158333
19654.508333
19663.791667
19673.841667
19683.558333
19693.491667
19704.983333
19715.950000
19725.600000
19734.858333
19745.641667
19758.475000
19767.700000
19777.050000
⋮⋮
19895.258333
19905.616667
19916.850000
19927.491667
19936.908333
19946.100000
19955.591667
19965.408333
19974.941667
19984.500000
19994.216667
20003.966667
20014.741667
20025.783333
20035.991667
20045.541667
20055.083333
20064.608333
20074.616667
20085.800000
20099.283333
20109.608333
20118.933333
20128.075000
20137.358333
20146.175000
20155.266667
20164.866667
20174.350000
20184.050000
 
# 54 page
daily.unemployment = unemp[J(all.dates), roll=31]
daily.unemployment
A data.table: 844 × 2
DATEUNRATE
<date><dbl>
1948-01-013.4
1948-02-013.8
1948-03-014.0
1948-04-013.9
1948-05-013.5
1948-06-013.6
1948-07-013.6
1948-08-013.9
1948-09-013.8
1948-10-013.7
1948-11-013.8
1948-12-014.0
1949-01-014.3
1949-02-014.7
1949-03-015.0
1949-04-015.3
1949-05-016.1
1949-06-016.2
1949-07-016.7
1949-08-016.8
1949-09-016.6
1949-10-017.9
1949-11-016.4
1949-12-016.6
1950-01-016.5
1950-02-016.4
1950-03-016.3
1950-04-015.8
1950-05-015.5
1950-06-015.4
⋮⋮
2015-11-015.0
2015-12-015.0
2016-01-014.9
2016-02-014.9
2016-03-015.0
2016-04-015.0
2016-05-014.7
2016-06-014.9
2016-07-014.9
2016-08-014.9
2016-09-015.0
2016-10-014.9
2016-11-014.6
2016-12-014.7
2017-01-014.8
2017-02-014.7
2017-03-014.5
2017-04-014.4
2017-05-014.3
2017-06-014.3
2017-07-014.3
2017-08-014.4
2017-09-014.2
2017-10-014.1
2017-11-014.1
2017-12-014.1
2018-01-014.1
2018-02-014.1
2018-03-014.1
2018-04-013.9
 
# 62 page
plot(stl(AirPassengers, "periodic"))