The FCI provides regional specific and relevant analysis for countries whose data is too sparse to generate country specific analysis. By providing an assessment of the financial conditions in economically meaningful clusters of developing country regions, the new generation of UNCTAD FCI tackles the complexity of increasing financial uncertainty and instability by harnessing the richness of information flows in the era of big data.
Built with the specific difficulties and concerns of developing countries in mind, the new generation of UNCTAD FCI can address difficult data (quality and omission) issues and provide insightful information to understand the interplay of the main drivers of financial stress at the country level. This is particularly relevant for LICs and MICs with often substantive gaps in data availability and quality.
The purpose of monitoring financial conditions in developing countries is to provide early warning of “financial stress” before it has adverse effects on the real economy. Financial stress cannot be directly observed, but may exhibit multiple concrete consequences which can be measured. Any financial crisis relates to one or more of five key phenomena: i) increased uncertainty about fundamental value of assets, ii) increased uncertainty about behaviour of other investors, iii) increased asymmetry of information, iv) decreased willingness to hold risky assets, v) decreased willingness to hold illiquid assets.
The first step in measuring financial stress is to carefully select variables which are proxies for these five key phenomena. The financial stress indicator is, in a second step, captured as the main driver of how these variables fluctuate together. Using factor analysis, financial stress has been measured in a range of mostly developed economies. UNCTAD is among the first institutions to design and produce such indicators specifically for developing countries.
Factor analysis is a way to efficiently summarise the information provided by the observed financial variables that fluctuate over time. The quality of this estimation depends on both the relevance of the chosen variables and the quality of the data itself.
The first UNCTAD FCI essentially applied factor analysis to each country individually, resulting in country-level financial stress indicators. These indicators seemed to perform well in signalling and capturing major shocks, such as for instance the global financial crisis of 2008, but they were highly volatile and suffered from scant and poor-quality data. Upon closer inspection, it became clear that for many developing countries, mostly LICs and MICs, the data were of insufficient quality for reliable estimation: too few variables were available, the data contained too many missing variables, and/or they were not reported in a regular fashion, resulting in extreme volatility of the estimated financial stress indicators.
A solution to this problem, as implemented in the new generation of FCI, is to regroup – or “cluster” - countries together, and to compute the financial stress indicator for each group. Although each situation is unique, developing countries share a large variety of similar patterns, particularly in to their exposure to external shocks and the vagaries of international financial markets. This new approach should not only better capture global drivers of financial stress in developing countries but also facilitate direct comparisons across countries, as similar countries would automatically fall into the same clusters. Also, from a technical point of view, this should result in more stable, timely, and reliable indicators while overcoming the problem of data scarcity.
The question of group formation, or clustering, is essential to the new methodology. A first idea would be to cluster the countries according to their geographical proximity. However, when it comes to financial profile, this approach is inadequate: for instance, the financial situation of Argentina is relatively closer to that of Turkey than that of Brazil, despite being geographically closer to the latter. Another option would be to cluster the countries based on qualitative and/or subjective assessments of the countries’ economic situation, but this approach may be biased and therefore criticized.
Instead, we propose to let the data “speak for itself” and automatically cluster the countries based on a measure of similarity between their respective financial conditions. This approach is not only more objective but also aligned with respect to the goal of having group-level FCI which are the most representative of the financial conditions of its members. It is important to note that the clusters will not necessarily match geographical regions, but may bring out new network patterns, which will as such enhance the economic analysis stemming from the FCI.
We applied this innovative methodology to our in-house datasets of financial variables for 53 developing countries and selected 5 clusters. The data consist of monthly and quarterly observations which span from January 2005 to July 2020. The dataset is chiefly obtained from Thomson Reuters Datastream. For each country, the number of financial variables differ and ranges from 7 to 24 with 843 variables in total. The data include financial and macroeconomic variables (interest rates, exchange rates, GDP, capital flows, among others).
Figure 1 shows a world map with the distribution of the country groups. The algorithm comes up with 5 groups and each group is identified in the map with a distinct colour.