Heterogeneous difference-in-differences for monadic treatments in dyadic flow data

Abstract

How do country-level (policy) changes affect bilateral flows? Estimating such monadic effects in dyadic data is challenging because treatment is collinear with directional-time fixed effects, pair-specific exposure to global shocks violates parallel trends, and network-wide equilibrium adjustments contaminate standard workarounds. We develop a common correlated effects difference-in-differences estimator that augments pair-level PPML regressions with cross-sectional averages from never-treated countries. Under a testable factor-spanning assumption, the estimator recovers the total general equilibrium effect. Simulations confirm identification. Applied to democratic regime change, we find large treatment effects and, using results from a border exercise alongside exact hat algebra, corroborate our estimator.

Non Technical Summary

How can we measure the impact of country-level policy changes on international flows?

Many important policy questions concern changes that occur at the national level and then affect its relationships with all other countries. For example: What happens to trade flows when a country becomes more democratic? How do capital controls affect international investment flows? How do migration reforms change migration flows?

Answering such questions is surprisingly difficult. Economists often study trade, investment, and migration using data on flows between pairs of countries. Existing statistical methods work well when a policy directly targets a pair of countries, such as a free trade agreement between two countries. When the policy affects individual countries, but changes interactions with every partner (and even non-partners) simultaneously, the approach cannot measure the magnitude of the economic effect of such a ‘treatment’.

This paper develops a new method to overcome these challenges.1 The method is designed specifically for situations in which a policy or institutional change occurs at the country level while outcomes are observed at the level of pairs of countries. Rather than comparing affected countries directly with unaffected ones, the approach models and removes the influence of global forces that affect all countries differently, such as global business cycles or commodity-price shocks. This allows the researcher to isolate the effect of the country-specific policy change more credibly.

Application: Does democratisation increase trade?

The paper applies the new method to a longstanding question in political economy: does transition to democracy increase a country's international trade? Using bilateral trade data covering more than 200 countries between 1960 and 2018, and four alternative measures of democratic transition, the analysis finds a consistently positive answer. Democratisation increases a country's exports by approximately 18% to 34%, depending on the democracy measure used.

Theory predicts that democracy affects trade through three channels. First, democratic transitions reduce barriers to international commerce, making it easier for firms to trade across borders (the direct ‘trade cost channel’). Second, democracy promotes economic growth and higher incomes in the democratising economies, which in turn support greater trade activity (the indirect ‘own-income channel’). And finally, the effects of democratic regime change (like other shocks) propagate to a country’s trading partners and eventually across the entire network, changing trading patterns for all countries (the indirect ‘network channel’). The estimated effects combine lower trade costs, income gains resulting from democracy, and a global adjustment mechanism.

Our estimated effect captures all three channels together. In other words, the results measure the overall impact of democratisation on trade as it actually unfolded in the economy, including global adjustments (affected by who else democratised), rather than measuring only the direct effect of reduced trading costs.

A particularly interesting finding relates to the durability of the benefits of democratic regime change: since a substantial part of the democracy effect is driven by a reduction in trade costs, it is perhaps not that surprising that we find that total effects are not substantially larger if we limit our analysis to those countries who experienced regime change and then remained democratic for the remainder of our sample period. This implies that democratic collapse may not result in a substantive export penalty, even if incomes may drop as a result.

Why does this matter?

The paper develops a new econometric method for estimating the effects of country-level policies in bilateral flow data. The contribution extends beyond the specific case of democracy and trade. Many policy questions involve country-level interventions that affect bilateral relationships, yet researchers have lacked reliable tools for studying them. The new methodology provides a practical way to evaluate policies that influence cross-border flows (of goods, services, capital, or people) while accounting for global shocks, network effects, and changing treatment patterns over time.