Abstract

In this project, we explore an international currencies dataset that contains data pertaining to the trade of various currencies during the years of 1890, 1900, and 1910. The data lends itself naturally to a network representation, where nodes represent countries/currencies, and directed edges representing active currency markets (where a directed edge \(e = (u,v)\) represents country \(u\) trading the currency of country \(v\)). We primarily explore the distribution of trade, as well as any community structure and clustering of countries based on economic variables. Finally, we explore the relative importance of currencies in the network and plot our network over a modern political map. Our results point to the consistent popularity of the British Pound Sterling (GBR), and the prolific growth of currency markets in the United States over this period of time. This growth suggests that the wealthiest and most highly developed countries can be clustered in their own group while the other countries can be clustered in one large group.

Data Description

The dataset we are using summarizes international foreign exchange markets in the period 1890-1910. The rows correspond to ordered country pairs, i.e. for every country pair there are two observations: (1) whether the foreign exchange markets in the first country trade the second country’s currency and (2) whether the foreign exchange markets in the second country trade the first country’s currency. There is information on 45 countries/currencies, yielding a total of \(45 \times 44 = 1980\) rows.

The dataset contains the following columns:

  • quote1890, quote1900, quote1910: indicator variables, =1 if in country_A there is an active market for the currency of country_B in 1890, 1900 and 1910 respectively, and 0 otherwise.

  • colony: indicator variable =1 if country_A is a colony of country_B and 0 otherwise.

  • dist: log distance as between the cities with foreign exchange markets in country_A and country_B.

  • bitrade: total trade between country_A and country_B in thousand of US dollars.

  • gold: indicator variable =1 if country_A has a currency convertible in gold in 1900 and 0 otherwise.

  • debtburden: Ratio of government debt over government revenues in 1900.

  • rlong: secondary market yield for gold denominated government debt in 1900.

  • rshort1900: market rate for 3 month lending, i.e. discount rate for 3 month commercial paper in 1900

  • rshort1890: Same as above for 1890

  • rgdp and rgdpcap give the log 1900 real GDP and the log real GDP per capita.

  • poldemo index of democracy (ID) developed by Vanhanen (2000) for 1900.

  • coverage is the logarithm of the number of currencies traded in country_A.

Research Question 1

Our dataset can naturally be thought of as a network of countries with (directed) edges signifying active currency markets. What do the in-degree and out-degree distributions look like for the most active countries (both trading and being traded)? How do these distributions change across the three timeframes?

In these barplots, we examine the top 10 countries by in-degree and out-degree. In the first barplot, we examine the progression of countries’ currencies by in-degree. A currency with a high in-degree means that the currency is being traded by many other currencies. That is, countries whose currencies are traded the most (as of 1910). We note that GBR, FRA, and DEU are the most traded currency (40 markets) out of every country by a fair margin; however, USA has seen the largest increase in the number of countries that trade the USD from 1900 to 1910.

In the second barplot, we examine countries with a high out-degree. A country with a high out-degree means that the country trades many other currencies. We note a similar trend here, where despite DEU, DNK, and NLD having more currencies that they trade, USA experienced the largest YOY growth (1890 to 1900, and 1900 to 1910). This is very interesting as the USA traded in the LEAST number of currency markets as of 1890, but eventually surpassed FRA, GBR, HKG, ITA, ARG, and BEL by 1910.

The findings of these two graphs suggest that USA has seen tremendous growth when it comes to both their currency being traded, as well as trading other countries’ currencies.

Research Question 2

What is the community structure of our network? Can we cluster the countries based on macroeconomic variables?

The plot above shows a plot of kmeans clustering for pairs of countries based on their log distance of the countries, total trade, and GDP of each country. We find in the graph above that many countries overlap in their clusters while only a few countries can be separated into their own clusters. This suggests perhaps that extremely wealthy and developed countries such as US, China, and Japan may be trade much more than other countries and thus can be clustered into their own groups. However, it seems that a significant majority of groups seem to share similar traits based on their distance, total trade, and GDP.

The complete dendrogram above shows clusters of countries based on the total trade between the two countries and GDP colored by the log distance between the cities of both countries. What we expected to see was that the countries with high GDP would likely have more total trade with each other and that the countries with higher total trade are also more likely to be geographically close to each other. However, looking at the dendrogram above, we notice that many of the clusters share many different distances and that there is not one major distance defining a cluster. It appears that distance appears to not have much influence on the total trade between two countries which makes sense given that the richest countries such as the US, China, Saudi Arabia, UK, etc. are all spread out geographically.

Research Question 3

Who are the power brokers? Who is central to the network and who is peripheral? Can we map the geography of the international monetary system?

In this section we focus on the trade network as it existed in the year 1890. Measures of centrality are designed to quantify such notions of importance. There are a vast number of different centrality measures. Which measure should we choose? We select the criterion that has the highest level of information about the influential vertices based on the network topology. To do this we perform PCA to distinguish the most informative centrality measure. Each centrality measure is treated as a variable and the centralities correlated with the principal components are the most important in identifying the central nodes. The contribution criterion (y axis in the graph) from the PCA shows how variables contribute to the principal components. This criterion allows us to detect which centralities have more information about the central nodes and so, which one can describe influential vertices of a network more accurately.

We have a few measures with have maximum contributions but we choose to use Degree Centrality due to its simplicity. It simply measures the number of links a node has to other nodes in the network (both incoming and outgoing)

Sizing nodes by Degree Centrality, we create a network graph synthesizing our findings:

To isolate a given country’s trade neighborhood click on the node or choose a node from the drop down menu. Notice that the central nodes in our network are clustered in the middle of the graph while peripheral nodes are spread out along the margins. Observe that British Pound Sterling (GBR) was traded by every country on our network! This suggests this was the most important currency at the time since there was a demand for it all around the world. If you wanted access to the most number of global currencies in 1890, you should go to Germany (DEU) since it has the most outgoing edges. Next up we proceed to map the geography of the international monetary system

The above plot displays the Currency Exchange Network in 1890 overlayed over a modern map where the size of node denotes denotes how central that node is to the network as in the previous graph. From this graph we can conclude the hegemony of European currencies when compared with the rest of the world. It is also noteworthy that the only African country who is trading
foreign currency according to our dataset it Egypt. The most demanded currency in the Americas is the US Dollar while the most demanded currency in Asia is the Hong Kong Dollar.

Conclusions

Our analysis highlights the disparate economic realities of the late 19th and early 20th Century. It appears that only a small group of countries can be distinctly clustered together based on their distance, total trade, and respective GDP while the rest of the countries can be clustered as one large group. While the graphs lack interpretability, we suspect that the most wealthy and developed countries such as US, China, UK, Germany, etc. may be the countries that can be grouped together since they would likely have the most trade and highest GDP. In particular, we find that the European empires’ currencies are ubiquitous in exchanges throughout the world while currencies from the Americas, Asia, and particularly Africa do not have the same global appeal.

Future Work

Often it is the case that one of the connected components in a graph \(G\) dominates the others in magnitude, in that it contains the vast majority of the vertices in \(G\). Such a component is called, the giant component. In our case, we have a connected component with Great Britain in the center. A characteristic observed in the giant component of many real-world networks is the so-called small world property. That is that despite its (possibly) large size, the typical number of hops along shortest paths between the network’s vertices would be quite small. It remains to be determined whether our network obeys this property