April 12, 2023
The recent global surge in inflation has impacted livelihoods around the world, particularly in crisis-affected areas. This additional shock has significantly affected households that were already living in fragile situations.
Governments, as well as humanitarian and development organizations, regularly monitor inflation rates to identify alarming trends and guide their actions to provide support. For example, high inflation can lead to a sharp increase in household spending needed to meet basic needs, requiring a policy response. In more extreme cases, a surge in food prices may indicate local food shortages, which signal the start or worsening of a food and nutrition crisis.
However, in many crisis situations, where conflict may make markets inaccessible, detailed price data is not regularly collected. These disruptions often coincide with periods and locations of high price instability. The lack of data makes it difficult to assess price movements accurately – information critical for understanding the severity of conditions in these areas and informing potential responses. But what if relief agencies could monitor food prices in real-time using alternative methods, even in remote locations during situations of conflict and violence? The information could identify the early onset or worsening of food crises, guide response efforts, and estimate the necessary response magnitude.
The World Bank’s Policy Research Working Paper Series published a paper that developed a machine-learning method to address this issue. The method uses surveys from nearby markets and prices of related commodities to estimate unobserved local market prices. This fills in gaps in area-specific price data for a basket of commodities, allowing for real-time monitoring of local inflation dynamics using incomplete and intermittent survey data. In lower-income countries where prices can be volatile and difficult to measure, the combination of surveys and machine learning predictions provides estimates with similar accuracy to direct measurements of prices.
Monitoring inflation presents a major challenge because prices of individual goods may increase sharply, while inflation reflects the general price level of many goods increasing simultaneously. To monitor inflation accurately, the prices of a broad range of goods, beyond food items, must also be tracked. That said, the larger the basket of goods, the more difficult it is to observe their prices simultaneously. In practice, the price of the full basket of goods is never directly observed. In situations affected by conflict, even monitoring a small basket of important staple goods using traditional data gathering methods can be exceedingly difficult, if not impossible.
The World Bank study uses an innovative approach to overcome this obstacle by constructing multiple machine learning models for different price items and linking them together to predict missing data based on other prices. This approach enables real-time monitoring of food prices in over 1200 markets across 25 countries, covering over 40 food items. The estimates are updated regularly and maintained as part of a broader collection of data sets in FCV (Fragility, Conflict and Violence). The data set reveals new insights into local area price dynamics during the 2007 World Food Price Crisis and the recent surge in inflation following the COVID-19 pandemic.
The paper compares predicted price data to left-out observed data, and demonstrated that the results are robust across a wide range of missing data settings. On average, the approach captured 85% of observed price variation across 25 fragile countries, even when 60% to 80% of survey data was missing. Although there are tradeoffs between data coverage and the reliability of the estimates, the results accurately captured all major price trends across a variety of FCV settings showing that even with limited ground truth data, robust inflation tracking is possible.
The results of this study provide crucial insights for decision-makers in low-income and data-poor locations, where maintaining comprehensive and expensive price monitoring programs using traditional consumer price index (CPI) methods to track general price levels for a broad range of consumer goods is challenging. The local estimates also overcome some of the limitations of the traditional CPI. National-level CPIs are based on prices measured in major urban markets and may not accurately reflect inflation in rural areas, where most poor populations in a country reside.
To improve people’s lives and livelihoods, particularly in crisis-affected settings, it is crucial to enable data-driven decision-making. The price monitor is just one aspect of the World Bank’s larger efforts to enhance and make data more real-time, using innovative methods for gathering and disseminating data. The monitor will be further developed as part of the World Bank’s Food Systems 2030 Multi-Donor Trust Fund and the Global Alliance for Food Security’s Global Food and Nutrition Security Dashboard. By utilizing innovative tools and techniques, this work fills critical analytical gaps that can be used to inform earlier, localized, and more effective responses to mitigate the impacts of future food and nutrition security crises.
Additionally, the method can complement conventional data collection efforts by gathering information at a lower cost and improving macroeconomic surveillance in data-limited regions. In the future, the machine learning-driven price monitor will be extended to cover the prices of non-food items, offering policymakers with a comprehensive and up-to-date view of detailed price data.
Source: World Bank Blog – by BO PIETER JOHANNES ANDREESUBHASH GHIMIRE
Legal Notice: The information in this article is intended for information purposes only. It is not intended for professional information purposes specific to a person or an institution. Every institution has different requirements because of its own circumstances even though they bear a resemblance to each other. Consequently, it is your interest to consult on an expert before taking a decision based on information stated in this article and putting into practice. Neither Karen Audit nor related person or institutions are not responsible for any damages or losses that might occur in consequence of the use of the information in this article by private or formal, real or legal person and institutions.
