by Keren Zhu & Timothy Gulden
The Belt and Road Initiative (BRI), a multi-trillion dollar global infrastructure development initiative proposed by the Chinese government in 2013, has huge potential economic impact at a global scale, yet there has been sparse research that delves into how to assess the economic impact of BRI projects. The methodological vacuum also echoes with an ongoing debate on the differences between Chinese and Western model of aid delivery and infrastructure development that has not been supported with sufficient empirical evidence.
In this project, we developed a method for assessing economic impact of BRI infrastructure projects using nighttime light. We compared results of China- and World Bank-funded hydro projects as a proof of concept, identified strengths and weaknesses in using nighttime light as a proxy for economic activities, and assessed whether the economic activity calibration method is feasible and scalable for studying developing economies.
Preliminary literature review suggests that potential differences between China- and Western-funded projects include differences in project cycle, construction time, due diligence process, control of socio-environmental impact, and expected timescale of economic returns. For this project, we are interested in the following questions:
- What are the trends in economic activity in the vicinity of projects before, during, and after implementation of projects funded by China and the World Bank respectively?
- Are economic impact similarities and differences between China-funded and World-bank funded projects consistent across different countries?
- What are the challenges and limitations in using nighttime light data in assessing project economic impact?
Although the BRI by itself was proposed less than seven years ago, it is actually a continuation of China’s overseas infrastructure development approach backed by Chinese policy banks and awarded to Chinese contractors for construction, a model that started in the early 2000s with the rise of the “Going Out” policy, with dedicated loan support from Chinese policy banks starting from 2005. In this regard, understanding infrastructure projects built after 2005 provides good insights into the nature and scale of project economic impact under the Chinese foreign development model, of which BRI is a direct outgrowth.
We explored the use of nighttime light data for assessing the impact of these projects because it offers a comprehensive and objective means of measuring activity even in areas with otherwise limited data. These projects are being conducted largely in areas where detailed, subnational economic activity data are limited and unreliable. This problem is exacerbated by the nature of major development projects which are intended to stimulate increased economic activity across the economy of their area of impact. These factors make it quite difficult to make a comprehensive and accurate assessment of the true economic impact of a project. While nighttime lights provide only a rough indicator of economic activity, it is an objective indicator – and it is available across most of the planet.
The nighttime light data used for this project has a spatial resolution of 90 arc seconds, which translates to pixels of about one square kilometer near the equator. The data were transformed using geographic projection and bilinear interpolation into true one-kilometer squares for analysis. Annual average light measurements for the years 1992 through 2013 were taken from the Defense Meteorological Satellite Program Operational Line Scan dataset as processed by NOAA. These data have limited dynamic range, showing maximum value at the outer edges of cities, are not well calibrated for comparison from one year to the next, and display uncontrolled regional noise. We performed interpolation on city centers, inter-year calibration, and conversion from light levels to GDP using methods developed in previous work. We then applied the calibration for 2013 to data from the Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band (DNB) that is available for the years 2013 to 2020 to produce consistent estimates of economic activity for all project areas spanning the years 1992 to 2020.
We compared projects of similar investment amount and project category built around the same year by Chinese entities and the World Bank, seeking to illustrate similarities and differences in project impacts, which may shed light on Chinese vs. Western infrastructure development models. For this pilot proof of concept project, we compared hydroelectric projects because they are clear examples of large infrastructure projects that can be expected to have significant economic impact.
Our research included the following steps:
- Step 1: Determine the country and projects to focus on based on availability and quality of time series data from China Global Investment Tracker and World Bank project database.
- Step 2: Integrate nighttime light datasets of 1992-2013 and 2013-20 from two sources.
- Step 3: Calibrate economic activities around project sites (50 km) using nighttime light data. Step 4: Generate economic activity trends, test hypothesis.
We selected the following project pairs for comparison. In Laos: Nam-Lik 1-2 Hydropower Project (China) and Lao Nam Theun 2 Power Project (World Bank). In Pakistan: Karot Hydropower Project (China) and Dasu Stage 1 Project (World Bank). In Nepal: Upper Maruyangdi A (China) and Kabeli A Hydroelectric Project (World Bank). See Figure 1.
Results and Lessons Learned
Our primary result is a negative one: we did not find that this approach to tracking the economic impact of hydroelectric projects looks particularly promising. The reasons for this are several. The 50km study area radius is likely too small to capture the impact of these projects. Hydroelectric projects tend to be built in very sparsely populated, mountainous regions and the electricity that they generate can be moved for long distances using high-tension transmission lines. It is likely that much of the electricity generated by these projects is actually consumed more than 50km from the project site. This method is well suited to measuring activity in cities and towns. Most of these regions included only very limited urbanized areas, leading to small samples and unstable estimates. While this approach seems to have succeeded at a capturing activity related to project construction, we do not see consistent evidence of project impact.
This approach may be better suited to looking at larger areas that have more built up areas. These areas need not be fully urbanized, but the method is quite limited in its ability to detect activity in rural and wilderness areas. It would also benefit from more specific delineation of expected project impact areas. The 50km radius appears to be too small to detect changes from projects of this type – but a larger area is likely to pick up development driven by unrelated factors.
In Figure 2 we see a comparison of the two projects in Laos. For these projects, as well as the others in our study, we did observe that Chine-funded projects had shorter planning, approval, and construction times, consistent with our hypotheses. A distinct increase in activity can be seen leading up to and during the construction of both projects, but large fluctuations in brightness in the years after the project becomes operational indicate make interpretation difficult.
Nighttime light data remains promising as a method for understanding economic development. The method captures consumption of final goods rather than intermediate consumption. This avoids double counting and is generally quite close to the core concept of GDP. This method can be applied to areas of arbitrary size and shape, and can be applied in places with limited data availability and produce estimates are independent of official data sources. It can be applied to many projects at once in order to produce statistically analyzable results.
Beyond the limitations of this study design, we found additional issues with the more recent VIIRS data as it is currently available for the years between 2013 and 2020. Annual composites are not publicly available. This necessitated the use of monthly data, which fluctuates with weather, moonlight, and other natural phenomena and has a higher noise floor which complicates or prevents the measurement of low-intensity activity. The high granularity, spatial precision, and dynamic range of the VIIRS data caused problems with the interpolation method developed with the earlier OLS data for estimating city centers. While this did not have a major impact on the current project, new methods for estimating urban GDP may be required for broader application of the approach.
“This work was supported by the RAND Center for Scalable Computing and Analysis (SCAN).”