Political risk indicators - how valid are they?

Camilla Jensen, Dept. for Social Sciences and Business, Roskilde University, Denmark
September 18, 2020

Paper presented at the

22nd Econometric Research in Finance Workshop * Online event * Szkola Glowna Handlowa (Warsaw School of Economics) * Warsaw, Poland, September 18, 2020

Background

  • Political risk indicators - research vs. practice
  • Practical perspective - hard to assess extent of material damage for foreign investors
  • New sources of big data - e.g. news-based EPU indicators now used in investment research
  • Political risk, financial risk and weather data for example all originate from complex systems (can be hard to model using traditional methods and by asking experts)
  • Theoretical idea in this paper stems from the Obsolescing Bargaining Model in the political science and international business literature (Vernon, 1979): that potential risk can only be 'revealed' from the project level of investment or using activity/entrepreneurship-based data instead

Critique of existing indicators

alt text

  • Progression bias
  • General high autocorrelation bias
  • Aggregation bias
  • Oppositely does our calculations show that the event-based indicator follows more a random-walk process

The components of the ICRG indicator

alt text

-One example with the problem of adding up subcomponent linearly: progress with Democratic Accountability will not necessarily lead to a lowering of risk in the short run - in fact the reverse is the case in emerging markets - see for example Jensen, C., & Zámborský, P. (2020). Balancing to Utopia: Multinationals in Oligarchies. In Non-market Strategies in International Business (pp. 41-73). Palgrave Macmillan, Cham.

Measuring risk ex-post? News-based risk indicator.

alt text

GDELT dataset - news-based indicator of political risk. Tracks global political events by country. Enormous database of very high quality for investigating the research questions.

Activity data - how? (here =no. of Starbucks stores)

One project-level indicator of many - ideally it should be an indicator of activity more vulnerable to risk than the one I am using - but as an example Starbucks' changing locations in Ataturk Airport over time can be indicative of the phenomena sought to be captured (political risk experienced at the project level).

alt text

alt text

Testing classical assumptions + relevance of events data

alt text

Testing alternative assuming ex-post data is valid

alt text

ADL models with panel data

(Prior to modelling all series are checked for the assumption of stationarity, all series pass the CIPS* test at the 1% level, except 'stores' which is borderline between stationarity and non-stationarity)

\[ Y_{it} = a_lY_{i, t-l}+b_lX_{i,t-l} + n_i + T_t + v_{i,t} \]

For this model the main interest is in the significance and sign (maybe also size but the series have not been standardised) of the betas.

*Cross-sectionally augmented Im, Pesaran and Shin (IPS) test for unit roots in panel models.

Granger testing - to panel or not to panel?

Ordinary Granger causality test vs. panel (Dumitrescu and Hurlin, 2012).

The ordinary Granger model is a 'pooling' model that does not take into account the panel structure of the data:

alt text alt text

Granger testing - to panel or not to panel?

Dumitrescu and Hurlin's test is an extension that allows for heterogeneity across cross sectoional units:

alt text

Results - classical model

H0 - Political Events cause Political Risk Indicator to go up - CONFIRMED

  • Expected effects for Betas, especially for positive events (passes both Granger tests)

H1 - Political Risk Indicators cause FDI Inflows or investor sentiment - CONFIRMED

  • Expected effects for Betas (passes only the global Granger test)

H2 - Political Events cause FDI Inflows or investor sentiment - NOT CONFIRMED

  • No effect for Betas (passes only the global Granger test)

=> Conclusion - FDI flows can be explained by the classical model that runs through risk indicators rather than events themselves, this is not surprising but does suggest that material damage from underestimating political risk is higher than it would have been had investors used better information about political risk.

Results - alternative model

H3 - Stores cause Political Risk Indicator - NOT CONFIRMED

  • No effect for Betas (but passes panel data Granger at very high significance)

H4 - Political Events cause Stores - WEAKLY CONFIRMED

  • Some effect for Betas (but passes panel data Granger at low lags with very high significance)

=> No conclusion yet, still need to fine-tune results and data, for example, Starbucks uses different entry modes that makes them avoid risk exposure

Discussion (2 more weeks of work..)

  • Still need to perfect on the measurement of events
  • Need to incorporate quarterly FDI data to finetune model
  • Check robustness of results by dividing stores into wholly owned and JV/licensing mode
  • Server problems in general due to size of GDELT dataset currently limit calculations

Conclusions

Potential conclusion of the paper after additional improvements and fine-tunning on methodology is that activity data can represent an important improvement to algorithms for projecting political risk, e.g. they can be used in complement and substitute for the traditional expert-based indicators.

In another paper we will build a forward-looking model instead, e.g. we will use the same ideas to project risk and ideally also be able to calculate how much material damage could be avoided by having used such alternative indicators.

THANK YOU FOR YOUR ATTENTION!