Geopolitical risks and recessions in a panel of advanced economies: evidence from over a century of data

ABSTRACT

This paper uses a panel of 17 advanced countries over the annual period of 1899–2013, to analyze for the first time, the role played by geopolitical risks in predicting recessions. After controlling for other standard predictors based on a logit model, we find that while aggregate geopolitical risks do not have any predictive ability, geopolitical acts enhance the probability of future recessions, with geopolitical threats reducing the same.

KEYWORDS:

• Geopolitical risks
• recessions
• advanced economies
• panel data
• logit model

JEL CLASSIFICATION:

• C33
• E32

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ The role of GPRs on stock market activity has been widely analysed (see,  for example, Bouras et al. (2018) for a detailed literature review). GPRs are primarily found to enhance stock market volatility – a metric often used for capturing uncertainty. Given this, GPRs can have a recessionary impact on the economy via this enhanced uncertainty.

² Results based on a conditional probit model yield qualitatively similar results, which in turn, are available upon request from the authors.:

³ The starting date of each country is: 1899 for Australia, 1982 for Belgium, 1934 for Canada, 1899 for Denmark, 1914 for Finland, 1899 for France, 1899 for Germany, 1971 for Italy, 1914 for Japan, 1899 for The Netherlands, 1899 for Norway, 1989 for Portugal, 1972 for Spain, 1899 for Sweden, 1902 for Switzerland, 1900 for UK, and 1899 for USA.

⁴ The weblink for the dataset is: http://www.macrohistory.net/data/.

⁵ The data is available for download from: https://www2.bc.edu/matteo-iacoviello/gpr.htm.

⁶ In this regard, note that Mueller and Rauh (2018) develops a methodology to summarize newspaper text using topic models. Using this method, the authors predict the outbreak of violent conflict one and two years before it occurs. Their analysis shows that the generated news data has a comparative advantage in predicting the timing of conflict. Mueller and Rauh (2018) argue that this makes news data a particularly useful addition to the most common conflict predictors used in the literature.

⁷ Based on the suggestion of an anonymous referee, who highlighted concern of endogeneity given that studies by Jawadi and Arouri (2011) and Anheier and Haley (2018) raise the possibility that recessions can actually affect geopolitical risks (by especially pointing towards the recent financial crisis), we re-estimated our three models based on an instrumental variables probit approach. However, our results in terms of the GPRs on the probability of recessions (available upon request from the authors), continued to be robust when we used the second lag of the GPRs to the instrument for the first lag. Note that since there is no method to estimate an instrumental variable logit regression unless one can figure out the appropriate bivariate distribution to use for the error terms, we relied on an instrumental variable-based probit model.

⁸ Of course, this line of reasoning remains to be verified, which unfortunately at this stage, is not feasible given the lack of data on defence spending over our sample period of more than a century, and hence, remains a speculation. But, this line of reasoning is not far-fetched and can be drawn from Bilgin, Gozgor, and Karabulut (2018), who indicates that GPRs tend to raise government capital expenditure in emerging countries. Moreover, as indicated by Apergis and Apergis (2017), and Caldara and Iacoviello (2018), GPRs also raise stock returns in defence industries, that in turn are likely to boost economic growth, particularly when economies are encountering threats rather than actual acts. In fact, in the BVAR model of Caldara and Iacoviello (2018), the initial impulse response function for output following a shock to GPRT is found to be growth-enhancing, thus providing some preliminary evidence for our line of reasoning.

⁹ Based on the suggestion of an anonymous referee, this result continued to hold when we estimated the model by separately using GPRA and GPRT. Complete details of these results are available upon request from the authors.