Climate change increases bilateral trade costs (through its impact on maritime shipping) – Bank Underground

Maximilian Huppertz

It is well established that climate change affects productivity, but its effects on trade costs have not been studied. Ignoring these and focusing solely on productivity could lead to an underestimate of its overall impact. It could also create a source of climate-related risk, with the potential to affect the financial system through trade finance and insurance. In a recent Staff Working Paper, I show that climate change indeed affects trade cost, driven by its impact on maritime trade in particular. Focusing on productivity alone leads to a roughly 9% underestimate of the overall impact. My methodology is easy to embed in studies of the overall impact of climate change.

Data 

I combine international trade data from CEPII TRADHIST and temperature data from Berkeley Earth. These allow me to relate decade-to-decade climate change to international trade flows. The trade data go as far back as the 1820s for some countries. They cover almost all countries in the world across the 20th century and almost all trade flows between countries starting in the 1950s. The temperature data cover all countries in the sample starting in the 1880s, and a good number of countries before that.

Empirical setup

I use gravity estimation, a well-established approach from international trade, to show that climate change affects trade cost. This relates trade flows between two countries to their strength as an exporter and an importer (driven, for example, by their productivity) and a set of coefficients which measure how costly it is to trade between the two countries. Specifically, I estimate:

$X_{nit}$ are trade flows from country $i$ to country $n$ during decade $t$. $\gamma _{it}$ and $\xi _{nt}$ are exporter-decade and importer-decade fixed effects. If climate change at either country affects its productivity, these fixed effects will capture that, and my results will not be driven by the well-known productivity impacts of climate change.

$\widetilde{d_{ni}}$ is the log distance between the two countries. The coefficient on this term, $\alpha _{t}$ captures how costly it is to bridge that distance – it captures trade cost. I expect this to be negative, as longer distances should be costlier to traverse, on average. (Not surprisingly, this is indeed what I find.)

The crucial terms are the interaction of log distance with $\Delta T_{it}$ and $\Delta T_{nt}$. These are changes in average temperature from the last decade to the current decade in the two countries. If climate change affects trade costs, climate change should make it harder to cross a given physical distance between two countries. I would then see negative coefficients $\delta _{1}$ and $\delta _{2}$ on these interaction terms.

Finally, I control for a few other determinants of trade costs in $C_{nit}$. These include whether the two countries share a border or official language, or have a colonial history.

Addressing potential spurious correlation

Countries across different climatic environments could see different trends in trade cost over time for reasons other than climate. For example, richer countries tended to experience faster warming since pre-industrial times and probably invested more in port infrastructure (thus reducing trade cost) at the same time. This would create a spurious negative correlation between climate change and trade cost.

To address this, I allow for differences in trade cost levels and trends over time based on countries’ climatic environment, as captured by their 1950–80 average temperature or their latitude. As a very conservative test of my hypothesis, I even allow trade cost to vary by the long-term climate change countries saw over the past 100 years. I then use only the remaining decade-to-decade variation across countries with similar long-term trends to estimate impacts. (This provides a conservative test because it discards long-term climate trends – arguably, some of the key variation of interest.)

Main empirical results

Figure 1 shows the two coefficients of interest, $\delta _{1}$ and $\delta _{2}$, plus 90% confidence intervals across the basic specification and spurious correlation checks I describe above, as well as additional robustness checks. (For example, subsetting to specific time periods, including a richer set of interactions, and using a different approach to addressing spurious correlation; full details in the paper.)

I find significant negative coefficients across specifications. The smallest impact (coefficients closest to zero) occurs with the conservative specification discussed above. Even with this restrictive test, however, I still find a significant and negative impact. Because it is difficult to interpret the size of these coefficients on their face, I turn to an economic model (below) to understand the magnitude of the effects.

Figure 1: Coefficients of interest across specifications

In the paper, I further show that the driving factor behind this appears to be maritime trade: countries separated by an ocean, or with excessively long land routes between them, see their trade costs increase compared to neighbouring countries.

This aligns with recent research on weather disruptions to ports (eg, due to storms), and the fact that policymakers and port operators are discussing this threat, developing plans to address it and taking costly actions to adapt to extreme weather events. I also find that adaptation seems slow – countries with especially fast climate change see larger impacts. 

How large are these effects?

To understand the magnitude of the effects, I feed my results into a widely used model of international trade, the Eaton-Kortum model. This model explains trade flows between countries in terms of differences in their productivity and the cost of shipping goods, capturing the same mechanisms I dealt with above. (Technically, for this exercise, I use an even more flexible specification that allows for different effects of climate change on colder and warmer countries; see the paper for details.)

I use the model to assess effect magnitude in two ways. First, I calculate the GDP per capita loss caused by climate change through its trade cost impact. Though not a perfect measure of living standards, GDP per capita is a key indicator for how badly households are impacted. Second, I calculate the underestimate of the total impact of climate change from ignoring trade costs and focusing solely on productivity.

Model results

To calculate the GDP per capita impact, I set up model counterfactuals that keep productivity at today’s levels but undo the trade cost impact of climate change since a given decade. Figure 2 shows the average change in GDP per capita across countries, weighted by population, when I do this across decades going back to the 1880s. It also shows the 5th and 95th percentile of GDP per capita changes across countries, highlighting that some gain more than others.

Figure 2: GDP per capita gains from undoing the impact of climate change on trade cost across decades

For example, I find that, if it were possible to undo the impact of climate change on trade cost over the past 100 years, average income per capita would increase by 1.6%. Some countries, though, would gain as much as 5%.

I show in the paper that this spread across countries is due to two main factors. First, it depends on countries’ climate trends relative to their neighbours. This is important because it highlights that focusing on individual countries’ climate trends in isolation can be misleading.

Second, the impact is larger for smaller economies, which tend to rely more on international trade. For example, the average gain across the larger half of economies (those with above median current GDP) is 1.4%. For smaller economies, the average gain is 2.8%.

In a second set of model exercises, I quantify the underestimate of the total impact of climate change from focusing solely on productivity. To do this, I estimate productivity effects in line with existing literature. I then compare the combined GDP per capita change from undoing both trade cost and productivity effects to the change from only undoing the productivity effect.

I find about a 9% underestimate. The difference is again larger for small open economies. For example, the UK sees a relatively large additional impact through the trade cost channel, compared to other rich countries. 

Policy implications 

The immediate implication is that climate-proofing seaports is important for adapting to climate change. This is especially true for small open economies that rely on maritime trade. Furthermore, on the global scale, the trade cost impact I find makes mitigating climate change somewhat more beneficial. When one underestimates the benefit of mitigating climate change, one might take less action than one optimally should.

Turning to the financial system, trade finance and insurance are potentially important for adapting to this trade cost impact. Insurance against extreme weather impacts, for example, allows logistics providers to not have to bear the rising risk of port disruption and delayed shipments. Otherwise, they might have to raise prices to account for this risk.

Trade finance, meanwhile, allows shipping companies to upgrade their fleets and terminals to, for example, better cope with storm surges, allowing ports to better function in harsher conditions. At the same time, there is of course a need to track the rising climate-related risk behind these financial instruments. 

Maximilian Huppertz works in the Bank’s Governance, Accounting, Resilience and Data Division.

If you want to get in touch, please email us at bankunderground@bankofengland.co.uk or leave a comment below.

Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.