Reflecting Geopolitical Risk in Investment Decisions
The geopolitical risk premium has once again taken centre stage as the likelihood of a hard Brexit draws near. Continuing US-China trade tensions, conflict in middle east, and the North Korean dilemma too are on investors mind. Given this backdrop, how do investors quantify geopolitical risk in their portfolios?
We present here an approach that makes use of SAPIAT’s risk modeling technology together with the Eurasia GPRI index data to identify and systematically track geopolitical risk at the asset level.
Eurasia Groups’s Political Risk Indices provide a quantitative measure that delivers insight into country-wide political risks for investors. Eurasia provide 4 separate risk indices (Security, Society, Government, and Economic) as well as a combined “Overall Political Stability” index (OPS) for a wide number of countries. Eurasia are attempting to incorporate political risk into fund managers’ investment processes, but the question most important to portfolio managers is how to extract the information in the political risk index to assist in a particular stock position sizing decision.
SAPIAT, through its EMA system, provides a range of factor models that can be used to analyse and break down the risk of investment portfolios. The EMA factors are generated using the EM (Expectation Maximisation) algorithm, one widely used in machine learning applications. The EM algorithm is an approach that attempts to iteratively find the most likely parameters to produce an observed distribution, and is particularly useful in contexts where values may be missing in the observed data. Using the EM algorithm helps extract the most relevant set of factors without human input. The resultant factor models contain a set of factor loadings for each security. Through these loadings, a risk profile can be inferred which can then be used for decision making. The factor loadings represent exposures to systematic underlying factors and the factor models can be used to perform an instantaneous “style analysis” of a portfolio, showing potential sources of risk.
By not pre-imposing a factor structure in the risk models, the EMA factor model picks up latent factors that drive the returns of assets. After adjusting for country and sector effects through optimal portfolio selection, the exposure to the geopolitical risk premium can be isolated.
To do this, we construct a long-short equity portfolio where the long portfolio holds countries that rank highly on the Eurasia index (i.e., low geopolitical risk) and the short portfolio sells short countries that rank lowly on the Eurasia index (i.e., high geopolitical risk). To hedge the effect of the emerging market risk premium, our long portfolio contains the top quintile of both developed and emerging countries, sorted by GPRI score, and the short portfolio contains the bottom quintile of both developed and emerging countries, with the overall portfolio being dollar neutral and beta neutral.
EMA uses a variety of techniques to model and attribute risk using our underlying statistical factor model. “Styles” are proxy portfolios used to represent typical investment themes or factors. Styles can be used in regression analysis to construct Multi-index models such as those described in Elton & Gruber. It can also be used in a weights-based analysis (similar to Sharpe’s style analysis). These styles have historically been macroeconomic time series such as stock-market indices. EMA started constructing factor profiles for EMA Styles in July 2008 to better model sector and country risk in portfolios and to expand the range of strategies covered. This approach was extended in July 2009 to model fundamental strategies. EMA styles are designed to mimic the typical strategies that are used to construct portfolios – sector bets, country bets, and fundamental bets. The approach was updated and extended to cover bond factors in January 2016.
Therefore, once we know the EMA factor profiles of each of the equity indices that make up a given risk index basket it is possible to calculate an EMA factor risk profile – a “risk style” - for a given month for that Eurasia risk index. Once the risk profiles for all months have been calculated they can be used to calculate correlations between any stocks or other assets that have an EMA risk profile for a given month and through time. The risk style baskets represent an investment in a particular investment style; here it is geopolitical risk as estimated by Eurasia that is being characterised.
Portfolio Analysis Example
These risk styles can now be used to analyse individual portfolios. In the example below, an analysis has been made of a cap-weighted benchmark portfolio using a “geopolitical risk style” set constructed from the GPRI component style indices.
The chart below plots the regression coefficients of the assets in 6 country portfolios (Australia, Belgium, Japan, Brazil, Chile, South Korea) against the Eurasia geopolitical risk style. It shows the sensitivity of each asset in each country portfolio to Eurasia’s index of geopolitical risk – a more positive coefficient indicates the equity returns rise with increased geopolitical risk. For example, it appears that the names in the Japanese portfolio seem to benefit from increased geopolitical risk – reflecting its safe haven status. Meanwhile, Brazil and Chile negatively impacted with an increase in geopolitical risk. However, there is wider dispersion among Brazilian names in terms of the individual impact of geopolitical risks at the asset level. A country like South Korea or Belgium are in between Japan and Brazil / Chile. The results align with broad intuition, but we can go deeper with the analysis by examining individual asset-level impact.
For example, in the case of South Korea, we can drill down into the difference in exposure of stocks to geopolitical risk across in the Airlines, Air freight and Aerospace industries within the country.
A defense-related business like Hanwha Aerospace will likely benefit from a deteriorating geopolitical security environment, while other sectors may suffer -- in particular, logistics firm like Hyundai Glovis.
The analysis above provides a brief but instructive example of how alternative data from non-traditional sources, can be incorporated into the investment process coupled with the use of quantitative approaches such as factor models. In particular, constructing various thematic or style proxy portfolios allows the addition of new signals, while adjusting for the co-movement with other systematic factors.