Knowing the recent present can be powerful for investors and policymakers – digital exhaust makes it possible
In the face of an increasingly interconnected and volatile global economy, investors and policymakers are tasked with making decisions based on an ever-changing set of data. Traditional economic forecasts, often reliant on outdated or lagging indicators, can leave policymakers several steps behind. This is where nowcasting comes into play—offering a powerful tool to forecast the near future in real-time, using high-frequency data. Originating in meteorology, nowcasting has evolved into a crucial method in economic analysis, allowing governments, central banks, and financial institutions to gain critical insights into the present and the immediate future.
From weather to economics
Meteorologist Keith Browning idea to use real-time data to predict imminent weather changes was later adapted by economists who realized that traditional economic indicators like GDP, inflation, and employment rates are often published with significant delays, making it difficult for policymakers to act swiftly.
Nowcasting, in contrast, uses real-time data to predict economic indicators with greater immediacy. By focusing on high-frequency data—such as weekly jobless claims, daily commodity prices, or even real-time credit card transaction data—nowcasting can provide more timely and actionable insights into economic conditions. This method has gained significant traction in central banking and investing allowing decision-makers to respond to changes in underlying conditions faster and more effectively.
Globally, institutions like the Federal Reserve Bank of Atlanta, for example, has developed the GDPNow model, which updates its forecast for U.S. GDP growth in real-time using a variety of high-frequency economic data. Similarly, the Federal Reserve Bank of New York publishes its own Nowcasting Report, offering forecasts on U.S. GDP growth. In India, the Reserve Bank of India (RBI) employs nowcasting techniques to enhance its monetary policy decisions. Institutions like the International Monetary Fund (IMF) have recognized the importance of nowcasting and have offered training programs to emerging economies around the world.
Techniques in Nowcasting
Nowcasting is about collecting data in real time and putting it through sophisticated techniques to process and analyze it effectively. Several key methodologies have emerged as integral to the nowcasting process. For example, the Dynamic Factor Models: These models extract information from a range of high-frequency indicators, such as industrial production, retail sales, and employment figures, and distill them into a real-time forecast. These models enable policymakers to gain an up-to-date picture of economic activity without waiting for official, lagging data releases.
Artificial intelligence (AI) and machine learning techniques, such as support vector machines (SVMs), LASSO regressions, and neural networks, have revolutionized nowcasting by analyzing large datasets to predict near-term economic trends. These algorithms can detect hidden patterns and correlations in data that might not be immediately apparent to human analysts. With a growing number of high-frequency indicators available, machine learning allows refining of the nowcasting models.
Examples
With regulators, companies, and consumers emitting copious amounts of digital data, such exhaust can be used to generate near real-time information about companies, sectors, and the economy.
By combing through prices at all the mandis and the various e-commerce portal, it is possible to build a gauge of inflation in which the changes in prices are updated daily from across the country. Such price changes when put in the index can yield a raw forecast of the underlying price trends. Using a machine learning model, the raw index can be trained to closely track the data that statistical authorities eventually release. Such nowcasting is useful to both policymakers and investors who get a much closer understanding of the underlying dynamics.
Nowcasting can draw data from a wide variety of sources: number and type of vehicles registered across more than 1,400 regional transport offices, trade journals (bhavcopy) of exchanges, number of listed restaurants or products for digital aggregator companies, shipping and export-import data, points to interest (branches, outlets, offices, factories, etc.), regulatory disclosures on filings for employment, production, pollution, etc.
Caution while Nowcasting
Data availability, its quality, model complexity need careful handling as the accuracy of nowcasting is heavily reliant on the quality of the underlying data and how it is analysed to generate outcomes.
The byproduct of our digital activities, social media posts, locations, and e-commerce transactions, can, when aggregated and analyzed at a meta-level, provide real-time insights into consumer sentiment, spending behavior, and market trends. Many of these sources, especially data on payment networks, is not readily available in India. According to the India Data Accessibility and Use Policy draft of the Ministry of Electronics and Information Technology, “India Data Office will notify protocols for sharing of non-personal datasets. Most datasets shall be made available at no cost to promote innovation and research & development.”
High-frequency data may be noisy or incomplete, and incorrect or biased data can skew predictions. It is important to ensure that the data pipeline is clean, the reported data points are verified for any anomaly, and the process of dealing with null or noisy data is laid out with caution.
Nowcasting models are often intricate, relying on vast amounts of data and complex algorithms. This complexity can make models difficult to interpret, which may limit their usefulness for decision-makers who need clear, actionable insights. Maintaining and updating these models requires significant expertise and resources: we need the ability to communicate as a narrative what the machine sees as a numerical model.
Future of Nowcasting
Nowcasting has generally been seen to be a numeric activity. With large language models that can draw upon not just quantitative data, it is possible to process significant amount of textual data to get a deeper understanding of the trend by piecing together commentary from policy makers, managements, courts, analysts, and media.
As digital data and large language models proliferate, the power of nowcasting with narratives will only grow stronger, enhancing the ability of policymakers and investors to navigate a rapidly evolving economic landscape.
Views are personal. The author is co-founder, Thurro.
