Brookings Paper - Turkey - Notebook

Data handling

# Start Inf data
# ---------------

# read data
dtf <- read_dta("Data/Turkey/startinf_Turkey.dta", encoding = "latin1")

# labels of each variable
dtf_labels <- unlist(lapply(dtf, attr, which = "label"))

# remove labels
dtf <- as.data.frame(lapply(dtf, c))

# Notice that for a given frequency,
# the data begins at the start of the period
# it reports to (for example, monthly data's date is on the 1st of
# each month)

# fix date
datelookup <- seq.Date(from = as.Date("2013-01-01"),
                       to = as.Date("2021-06-01"),
                       by = "month")
dtf$date <- datelookup[dtf$date - abs(min(dtf$date)) + 1]

# order by date
dtf <- arrange(dtf, date)

# correct magnitude of certain variables
dtf <- dtf %>% 
        mutate_at(vars(-date ,-contains("_skew")),
                  ~ ./100 ) #it's ok no skewness here

for (k in c(1,2,5,10)) {
  
  var <-     paste0("Ecpi",k,"_max_minus_min")
  var_max <- paste0("Ecpi",k,"_max")
  var_min <- paste0("Ecpi",k,"_min")
  dtf[,var] <- dtf[,var_max]-dtf[,var_min]
  
}


# compute distribution skewness
# -----------------------------

sk <- function(x) {
   
  # x matrix with probability buckets
  y <- rep(c(1.5, seq(3.25, by= 0.5,length.out = ncol(x)-1)), nrow(x)) # midpoints
  y <- matrix(y/100, nrow = nrow(x))

  Ex  <- rowSums(x*y)
  Ex2 <- rowSums(x*(y^2))
  Ex3 <- rowSums(x*(y^3))
  stdev  <- sqrt(Ex2 - Ex^2)

  # https://en.wikipedia.org/wiki/Skewness
  skewness <- (Ex3 - 3*Ex*stdev^2 - Ex^3)/(stdev^3)

  return(skewness)
}

# 1 year
x <- as.matrix(dtf %>% select(starts_with("Ecpi1_p_")))           # probabilities
dtf$Ecpi1_skewness <- sk(x)

# 2 years 
x <- as.matrix(dtf %>% select(starts_with("Ecpi2_p_")))           # probabilities
dtf$Ecpi2_skewness <- sk(x)

# adjustment of lead data
# -----------------------
# be careful. If you use more than one year horizons, you have
# to adjust the lagging

if (bAdjustPeriodTiming) {
  dtf <- dtf %>%
    mutate_at(vars(starts_with("Ecpi"),
                   starts_with("breakeven"),
                   bt_Eppi),
              ~ lag(.,12)
             ) # monthly data
}

# turkey's data is end of period, so no need for "-1"

# arrange data for plotting
dtf_long <- pivot_longer(dtf, !date,
                         names_to = "variable",
                         values_to = "value",
                         values_transform = as.numeric)

Timespan is 2013/01 - 2021/07

Figure(s) 12: Turkey’s drifting expected inflation anchor: 2018 - …

Subfigure 14(a): Actual inflation, markets and survey first-order moments

# Actual and expected inflation 

vars1 <- c("pi",
           "Ecpi1_median",
           "breakeven",
           "bt_Eppi")

breakss <- c(vars1, "Target")

legs <-  c("Actual inflation",
           "Cross-sectional median of professionals’ survey",
           "Market-price implied expectation",
           "Weighted mean of firms’ survey",
           "Target")
legv <- c('red', 'blue', 'green4', 'orange3', 'black')
linev <- c('solid', 'dotted', 'dashed', "dotdash", 'longdash')


# inflation target
target <- 0.05

# thickness of line
vline_w <- 0.6 # mm
# ticks
labs <- c(0, target, seq(target, 0.3, length.out = 6)[2:6])

# plot
plot <- create_plot(dtf_long, date_lb = as.Date("2015-01-01"),
                    vars = vars1, scales = 0) +
  theme(legend.position = c(0.27,0.8)) +
  geom_hline(aes(yintercept = target,
                 colour = "Target", linetype = "Target"),
             size = tfont(vline_w)) +
  scale_linetype_manual(breaks = breakss,
                        values = linev, labels = legs) +
  scale_colour_manual(breaks = breakss,
                      values = legv, labels = legs) +
  scale_y_continuous(limits = c(0, 0.3),
                     breaks = labs,
                     labels = scales::percent(labs, accuracy = 1))

Scale for ‘y’ is already present. Adding another scale for ‘y’, which will replace the existing scale.

print(plot)

if (bSavePlots) my_ggsave(paste0(path_figure_out, "figure_12_a.pdf"), plot)

Subfigure 12(b): (b) Cross-sectional survey distribution

# Forecast distribution

# CHOICE
###################################
ex <- 1 #1 or 2 years expectations
dater <- c(as.Date("2017-09-01"),
           as.Date("2019-09-01"),
           as.Date("2020-09-01"))
###################################
# END OF CHOICE

sc <- paste0("date|Ecpi", ex, "_p_") 

buckets <- c(1.5, seq(3.25, by = 0.5, length.out = ncol(x)-1))/100

dtf_temp <- dtf_long %>%
  filter(grepl(sc, variable), date %in% dater) %>%
  arrange(variable) %>%
  mutate(buck = rep(buckets,length(dater)),
         yr = factor(year(date),
                     levels = year(dater))
         )

plot <- ggplot(dtf_temp,
               aes(x = buck, y = value, fill = yr)) +
  geom_col(width = tfont(0.0045),
                  position = "dodge") +
  facet_wrap(. ~ yr) +
  scale_y_continuous(labels = scales::percent_format(accuracy = 1)) +
  scale_x_continuous(labels = scales::percent(buckets[seq(1, length(buckets), by = 3)], accuracy = 1),
                     breaks = buckets[seq(1,length(buckets), by = 3)]
                    ) +
  scale_fill_manual(breaks = year(dater), values = c("red3","blue3","orange")) +
  theme(legend.position = c(0.4,0.85),
        axis.ticks.x = element_line(),
        axis.text.x = element_text(size = tfont(4)),
        strip.text.x = element_blank()
       )

print(plot)

if (bSavePlots) my_ggsave(paste0(path_figure_out, "figure_12_b.pdf"), plot)