The recent spike in global inflation is impacting lives around the world, especially in the crisis-affected areas. This additional shock is having a major impact on already vulnerable households.
Governments, humanitarian and development organizations regularly monitor inflation rates to identify alarming trends and guide action to provide assistance. For example, high inflation can lead to a sharp increase in household spending needed to meet basic needs, requiring policy responses. In more extreme cases, food price spikes may indicate regional food shortages, which indicate the beginning or worsening of food and nutrition crises.
However, in many crisis situations where disputes can make markets inaccessible, detailed price data is not collected on a regular basis. These disruptions often coincide with times and places of price volatility. The lack of data makes it difficult to accurately assess price movements. This is important information to understand the seriousness of the situation in these areas and to inform potential responses. But what if relief agencies could use another method to monitor food prices in real time, even in remote areas of conflict and violence? identify, guide response efforts, and estimate the scale of response required.
of the World Bank Policy Research Working Paper Series We published a paper in which we developed a machine learning method to address this problem. This method uses surveys from nearby markets and prices of related commodities to estimate unobserved local market prices. This fills gaps in region-specific price data for baskets of commodities, allowing real-time monitoring of regional inflation dynamics using incomplete and intermittent survey data. In low-income countries where prices are volatile and difficult to measure, a combination of surveys and machine learning predictions can yield estimates with similar accuracy to direct price measurements.
Monitoring inflation is a major challenge, as it reflects the general price level of many commodities rising at the same time, while the prices of individual commodities can rise sharply. Accurately monitoring inflation requires tracking the prices of a wide range of non-food commodities as well. That said, the larger the basket of commodities, the more difficult it is to observe their prices simultaneously. In practice, the price of a complete basket of commodities is never directly observable. In conflict-affected situations, it can be very difficult, if not impossible, to monitor even a small basket of critical necessities using traditional data collection methods.
A World Bank study used an innovative approach to build multiple machine learning models for different price items and link them to predict missing data based on other prices to Overcoming this obstacle. This approach enables real-time monitoring of food prices for over 40 food products in over 1200 markets in 25 countries. Estimates are regularly updated and maintained as part of a broader FCV (fragility, conflict and violence) data set. This data set reveals new insights into regional price volatility during the 2007 global food price crisis and the recent spike in inflation following the COVID-19 pandemic.
This paper compared the predicted price data with the excluded observed data and showed that the results were robust across various missing data settings. On average, this approach captured 85% of observed price movements in 25 fragile states, even when 60% to 80% of survey data were missing. Although there is a trade-off between data coverage and reliability of the estimates, the results demonstrate that robust inflation tracking is possible even with limited ground truth data, and can be applied to various FCV settings. Accurately captures all major price trends across the board.
The results of this study provide important insights for decision makers in low-income, data-poor regions. The region maintains a comprehensive and expensive price monitoring program using traditional consumer price index (CPI) techniques to track prevailing price levels for a broad range of consumers. Products are challenging. Local estimates also overcome some of the traditional CPI limitations. His CPI at the national level is based on prices measured in major urban markets and may not accurately reflect inflation in rural areas where most of the poor live in the country.
Enabling data-driven decision-making is essential to improving people’s lives and livelihoods, especially in crisis-affected environments. Price Monitor is just one aspect of his larger effort to enrich the data and make it more real-time, using innovative methods for collecting and distributing 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 leveraging innovative tools and techniques, this work fills important analytical gaps and promotes earlier, more localized and more targeted interventions to mitigate the impact of future food and nutrition security crises. It can be used to inform an effective response.
Moreover, this method can complement traditional data collection efforts by collecting information at low cost and improving macroeconomic monitoring in data-limited areas. In the future, machine learning-driven price monitors will be extended to cover non-food prices, providing policy makers with a comprehensive and up-to-date view of detailed price data.
Note
The methodology was developed by Bo Pieter Johannes Andree, a data scientist in the World Bank’s Development Economics Data Group, as part of the program Building Evidence for Protracted Displacements: A Multi-Stakeholder Partnership. The program was managed by the World Bank Group (WBG) and funded by UK aid established in partnership with the United Nations High Commissioner for Refugees (UNHCR). This paper utilizes monthly price survey data collected by the World Food Program (WFP).
The scope of the monitor is being expanded as part of the World Bank’s Food Systems 2030 Multi-Donor Trust Fund. This program was funded by the Federal Ministry for Economic Cooperation and Development (BMZ, Germany).
Using data from the International Food Policy Research Institute (IFPRI) in Papua New Guinea, the method is currently being further developed to handle more price items and remain robust even when the data range is low. . This work is supported by a broader DFAT-funded initiative, the Pacific Observatory, to improve the frequency, timeliness and granularity of key economic and development indicators for data-driven policymaking. It is intended to provide a non-traditional data source as a complement to official statistics for the purpose of Papua New Guinea and the Pacific Islands.
