Make it easier for humans to access data from the Maddison Data Project
in R. Later releases may include vignettes, etc., documenting analyses
using the [KFAS] (Kalman filtering and smoothing, aka state space)
techniques with these data.
Objectives: Make it relatively easy in R to do the following:
Find the countries with the highest gdppc for each year for which
data are available using MaddisonLeaders(). Allow deleting
companies with high gdppc based on something narrow like a
commodity, e.g., oil. [DONE]
Support modeling a time series as c('level', 'growthRate').
2.1. Using the KFAS package with functions growthModel and
growthUpdateFn plus plotting with ggplotPath and ggplotPath2 as
documented in vignette KalmanSmoothing. [ALMOST DONE, but
KFAS::fitSSM seems not to pass optional arguments to optim and
strips dimnames from components of an SSModel. This means that with an
irregular time series, e.g., data missing for some years, the gaps are
ignored, and esimation assumes the data are contiguous. This impacts
especially the state transition matrix T and transition variance Q.
This should be a minor problem problem for series with only a few
missing values and will be ignored until convenient to fix.]
2.2. Ask John Nash for his preferred substitute for optim. Also ask
him about supporting an optional fixPar vector = NA for par to fix
and the numeric values for others. This could support easy testing
testing of submodels.
2.3. Ask Jouni Helske and the other KFAS and bssm contributors about
supporting (a) dimnames and (b) an optional Time and dT
component(s)) of an SSModel plus (c) a function asTime = Time
component of a model else as.Posixct(names(y)) else
as.Date(names(y)) else as.numeric(names(y)) else
[ordered(names(y)) with dT =1:length(y)] withdT_ =
diff(Time)anddT = c(dT[1],
dT)and (d) forbssmassuming Student's t migrations ofgrowthRate`.
library(MaddisonData)
MadDat1600 <- subset(MaddisonData, year>1600)
Leaders1600 <- MaddisonLeaders(c('ARE', 'KWT', 'QAT'), data=MadDat1600)
summary(Leaders1600)
#> ISO yearBegin yearEnd n p
#> ARE ARE 1965 1984 5 0.2500000
#> AUS AUS 1853 1891 17 0.4358974
#> CHE CHE 1931 1934 4 1.0000000
#> GBR GBR 1808 1898 67 0.7362637
#> KWT KWT 1953 1957 5 1.0000000
#> LUX LUX 1991 1995 5 1.0000000
#> NLD NLD 1601 1807 207 1.0000000
#> NOR NOR 1996 2002 7 1.0000000
#> NZL NZL 1873 1874 2 1.0000000
#> QAT QAT 1950 2022 45 0.6164384
#> USA USA 1882 1990 58 0.5321101
gdppc and / or pop for a selection of
countries, e.g., world leaders.str(GBR_USA <- subset(MaddisonData::MaddisonData, ISO %in% c('GBR', 'USA')))
#> Classes 'tbl_df', 'tbl' and 'data.frame': 1004 obs. of 4 variables:
#> $ ISO : chr "GBR" "GBR" "GBR" "GBR" ...
#> $ year : num 1 1000 1252 1253 1254 ...
#> $ gdppc: num NA 1151 1320 1328 1317 ...
#> $ pop : num 800 2000 NA NA NA NA NA NA NA NA ...
GBR_USA1 <- MaddisonData::ggplotPath('year', 'gdppc', 'ISO', GBR_USA, 1000)
GBR_USA1+ggplot2::coord_cartesian(xlim=c(1500, 1850)) # for only 1500-1850
GBR_USA1+ggplot2::coord_cartesian(xlim=c(1600, 1700), ylim=c(7, 17))
# label the lines
ISOll <- data.frame(x=c(1500, 1800), y=c(2.5, 1.7), label=c('GBR', 'USA'),
srt=c(0, 30), col=c('red', 'green'), size=c(2, 9))
GBR_USA2 <- ggplotPath('year', 'gdppc', 'ISO', GBR_USA, 1000,
labels=ISOll, fontsize = 20)
# h, vlines, manual legend only
Hlines <- c(1,3, 10, 30)
Vlines = c(1849, 1929, 1933, 1939, 1945)
(GBR_USA3 <- ggplotPath('year', 'gdppc', 'ISO', GBR_USA, 1000,
ylab='GDP per capita (2011 PPP K$)',
legend.position = NULL, hlines=Hlines, vlines=Vlines, labels=ISOll))
LATER:
Build a state space / Kalman models for gdppc and pop for each
country in the Maddison project data.
Use Kalman smooth to interpolate and extrapolate (forward but not
backwards) gdppc and pop for each country for all years that
appear anywhere in the Maddison project data.
Identify the world leader in gdppc for each year, refining “1”
using KFAS interpolation.
Identify the world technology leader for each year by evaluating the
gdppc leader for each year and replacing any whose leadership was
narrow like members of OPEC with a country with a broad-based
economy like the US.
You can install the development version of MaddisonData from GitHub with:
# install.packages("pak")
# pak::pak("sbgraves237/MaddisonData")
install.packages("MaddisonData", repos = c('https://sbgraves237.r-universe.dev', 'https://cloud.r-project.org'))
[Coming soon.]
library(MaddisonData)
## basic example code