Plotting a GDP time series

To plot a GDP time series (as in the slides of the introductory chapter of the course, here, in French) one first has to find and download some data. As in the course, I use the Maddison Project Database, which provides (very) long run data on GDP and population for more than 160 countries¹. To download the data, which come as an Excel file, I use a small Python script:

import urllib.request as url
MADDISON = 'mpd2020'
MADDISON_PATH = 'https://www.rug.nl/ggdc/historicaldevelopment/maddison/data/'+MADDISON+'.xlsx'
File = url.urlopen(MADDISON_PATH)
data = File.read()
with open('./'+MADDISON+'.xlsx', 'wb') as f:
    f.write(data)

This script downloads an Excel file from the Maddison Project Database website and saves it locally on my disk. Since I do not like Excel files, I convert the file mpd2020.xlsx into a csv (Comma Separated Values) text file:

import pandas as pd
data = pd.read_excel(MADDISON+'.xlsx', 'Full data', dtype=str, index_col=None)
data.to_csv(MADDISON+'.csv', encoding='utf-8', index=False)

The file mpd2020.csv is saved to disk. It is a text file (which you can therefore read with any text editor, or process programmatically, for instance to build charts). The file contains a table. The first column gives a three-letter country code, the second the name of the country, the third the year, the fourth GDP per capita (in 2011 dollars) and the fifth the population in thousands. For instance, rows 2 to 73 contain observations for Afghanistan from 1820 to 2018. Note that the years are not necessarily consecutive (it depends on the country and the period) and that more or less data are available depending on the country (it depends on the history and culture of the country, more or less inclined towards statistics).

In the course, I produced the charts with Matlab, a scientific computing language (the codes used for the charts of the course are available here). This software is proprietary and not free. Here I show how to use Python (and the matplotlib library) to produce these charts. As an example, I will plot the evolution of real GDP per capita in France.

Let us start by extracting from data (this object is what is called a dataframe and holds all the data) the observations for France. You can display the observations for France by selecting the corresponding rows as follows:

data[data["countrycode"]=="FRA"]

The extract contains 711 observations, and GDP per capita is observed annually from 1280 (that is, from the time of Philip III the Bold, son of Saint Louis) to 1789, then again from 1820 onwards. The file contains no observation between these two dates, thirty years covering the Revolution, the Empire and the beginning of the Restoration, and since the corresponding rows are missing rather than filled with NaN, the chart below bridges the gap with a straight line segment. Population is only observed annually from 1820 onwards. To plot this time series, one uses the matplotlib library (there are other plotting libraries which might let you obtain charts more to your taste):

import matplotlib.pyplot as plt
# Select the data for France from 1280 onwards
DATA = data[(data["countrycode"]=="FRA") & (data["year"].astype(int)>=1280)][["year","gdppc"]]
# Convert the data to numeric values (so far they are character strings)
DATA["year"] = pd.to_numeric(DATA["year"])
DATA["gdppc"] = pd.to_numeric(DATA["gdppc"])
# Plot
DATA.plot(x="year",y="gdppc")
plt.savefig("gdppc-fr.svg", transparent=True)

One observes that not much happens until the beginning of the 19th century. Here the chart shows the level of GDP per capita; it would be more appropriate to display the natural logarithm of GDP per capita (to smooth out the explosive behaviour of the variable from the middle of the 19th century onwards, and possibly to get a better view of the fluctuations of GDP before 1820).