Monetary authorities worldwide are struggling to navigate a complex inflation landscape following the pandemic and geopolitical disruptions. The debate over whether recent inflation surges were driven primarily by supply shortages or demand factors has significant implications for monetary policy (Blanchard and Bernanke 2023, Bánbura et al. 2023, Gianone and Primiceri, 2024). In this environment, better inflation forecasting tools have become increasingly important.
Standard inflation forecasting models have faced challenges, particularly when dealing with rapid changes in economic conditions. While commodity prices have historically been considered leading indicators of inflation (Bernanke 2008, Gerlach and Stuart 2024), the relationship has weakened since the mid-1980s (Stock and Watson 2003), making them less reliable predictors.
Disentangling supply and demand in commodity markets
Our research shows that textual measures of supply and demand developments in commodity prices provide distinct information about future inflation movements relative to existing predictors, inflation expectations, and survey forecasts (Malliaropoulos et al. 2024). Importantly, we demonstrate that this predictive power is obscured by the fact that commodity prices are driven by both supply and demand shocks, and these shocks have different passthrough to inflation.
We document that aggregate demand shocks tend to affect most economic sectors simultaneously in a homogeneous manner, creating a strong and direct relationship between commodity prices and overall inflation. In contrast, supply shocks often have idiosyncratic characteristics that can lead to diverse outcomes. For example, a negative supply shock in the cocoa market may increase cocoa prices but simultaneously depress sugar prices (a complementary good) while having limited effects on coffee prices (an imperfect substitute). These complex interactions can dampen the overall inflationary impact of supply-driven commodity price changes.
This novel approach to inflation forecasting employs supply and demand indicators spanning all major tradable commodity markets built using textual analysis of business news from the work of Lumbanraja et al. (2025). A supervised, automated narrative approach that analyses more than a million Reuters news articles between 2001 and 2023 is employed to capture supply and demand dynamics in commodity markets. The framework identifies words and phrases related to supply and demand factors across a wide range of commodity markets – from energy and metals to agricultural products and livestock.
A text-based approach has the advantage of capturing information across diverse commodity markets rather than focusing on a single sector (e.g. the oil market). This is crucial for predicting inflation, which is driven by the shocks that affect most sectors of the economy homogeneously and simultaneously. Figure 1 plots the first principal component of individual commodities’ net demand indices, the global net demand (composite) index, and a measure of the business cycle, illustrating the high correlation across these measures which is in line with our proposed mechanism: aggregate demand shocks tend to be correlated across different commodities, leading to a strong pass-through effect on the consumption basket.
Additionally, our textual methodology requires fewer a priori parametric restrictions and provides timely indicators available at higher frequency than production data typically used in forecasting. The resulting indices track major commodity-related events remarkably well, such as OPEC production decisions, trade wars, natural disasters, and the COVID-19 pandemic.
Figure 1 Net demand indices and the business cycle
Note: The figure plots the first principal component of net demand indices of individual commodities, the net demand composite index, and the (negative) of the business cycle index. Commodities included are aluminium, cattle, cocoa, coffee, copper, corn, cotton, gasoline, hogs, natural gas, oil, soybean, sugar, wheat and zinc, spanning the period May 2001-June 2023.
Improving inflation forecasts
Our findings show that text-based indicators of supply and demand disturbances significantly enhance inflation forecasting accuracy by reducing out-of-sample forecast errors by up to 20-30% compared to standard models. Additionally, the textual indices contain information that is not captured by existing predictors, including past inflation, policy rates, yield curves, and currency movements, outperforming models that use commodity price changes alone as predictors.
Importantly, demand-side disturbances contribute more to predictability than supply-side factors, consistent with the hypothesis that demand shocks generate more correlated movements across consumer prices.
The proposed forecasting framework performs well across different inflation regimes – rising, falling, and stable – and provides consistent improvements even during the volatile post-COVID period. This suggests that the textual supply and demand indicators capture fundamental relationships rather than period-specific patterns (Figure 2).
Figure 2 Difference in cumulative SSE
Note: The panels plot the difference between the cumulative sums of squared errors (SSE) between the autoregressive (AR) model with commodity prices, and the model which includes the net demand indices. Controls in both specifications include the federal funds rate, the unemployment gap, a yield curve measure (the 10-year minus the 2-year Treasury yield), and the log growth of a trade-weighted dollar index. h: forecast horizon in months.
How the indices work in practice
Using local projections, we confirm that commodity market events captured by our text-based indices affect inflation in economically meaningful ways. Demand disturbances generate almost twice the inflation impact of supply disturbances, and this effect persists for longer periods. The responses differ across inflation components as well. Goods inflation shows stronger responses to both supply and demand developments than services inflation (Figure 3).
The key point of our approach that classifies narratives as supply- or demand-related and builds distinct indices is that the source of the price movement, not only its magnitude, matters for inflation prediction. Even when supply and demand shocks push commodity prices in the same direction, their macroeconomic correlations differ and pooling them dilutes the informational signal.
Crucially, the supply and demand indices contain information not reflected in various measures of inflation expectations – whether derived from financial markets, professional forecasters, or consumer surveys. This suggests that they capture insights that are not already incorporated into market participants’ views or expert forecasts. This pattern persists when we assess different inflation measures, and remains stable when we exclude the COVID-era high inflation period.
Figure 3 CPI inflation response to supply and demand developments
Note: Log CPI response following a one unit increase in contemporaneous composite net supply and net demand, up to an 18-month horizon (90% and 95% CI). The response is normalized to a shock that raises the GSCI by 10% in 12 months. The teal dashed lines show significance bands for the test that all coefficients of interest are jointly zero, as in Jordà (2023). Controls: industrial production log change, S&P 500 log returns, FFR, 10-year minus 2-year US treasury, VIX, trade-weighted US dollar log returns. Each specification also includes a constant and 13 lags of log CPI. COVID corrections using a COVID dummy and rescaling of the indices for January-April 2020. Sample: March 2001-June 2023.
Policy implications
These findings have important implications for central banks and policymakers. The proposed framework provides a more accurate and timely method for predicting inflation, which can help central banks make better-informed monetary policy decisions. By distinguishing between supply and demand drivers, policymakers can better understand the nature of inflationary pressures and adjust their responses accordingly by coordinating monetary policy tightening (Miranda-Pinto et al. 2024a) or taking other macroprudential measures. The timely nature of news-based indicators allows for their integration into existing forecasting methodologies employed by central banks, thereby enhancing the precision and responsiveness of inflation predictions. Last, the framework captures developments across the entire commodity space rather than focusing on a single sector, providing a more comprehensive picture of potential inflationary pressures as recently illustrated by Miranda-Pinto et al. (2024b).
As central banks continue to navigate uncertain economic conditions, tools that enhance inflation forecasting accuracy become increasingly valuable. Our text-based approach offers a promising addition to existing forecasting methods, particularly in periods of high volatility and when traditional models struggle to capture rapid economic changes.
References
Bánbura, M, E Bobeica, and C Martínez Hernández (2023), “What drives core inflation? The role of supply shocks”, ECB Working Paper No. 2875.
Bernanke, B S (2008), “Outstanding issues in the analysis of inflation: a speech at the Federal Reserve Bank of Boston’s 52nd Annual Economic Conference”, Speech 412, Board of Governors of the Federal Reserve System.
Blanchard, O J and B S Bernanke (2023), “What Caused the US Pandemic-Era Inflation?”, NBER Working Paper No. 31417.
Gerlach, S and R Stuart (2025), “Commodity prices, the business cycle, and inflation: International evidence, 1851-1913”, VoxEU.org, 14 February.
Giannone, D and G Primiceri (2024), “The Drivers of Post-Pandemic Inflation”, NBER Working Paper No. 32859.
Hong, Y, F Jiang, L Meng, and B Xue (2024), “Forecasting Inflation Using Economic Narratives”, Journal of Business & Economic Statistics 43(1): 216–231.
Lumbanraja, A, S Mouabbi, E Passari, and A Rousset Planat (2025), “The Origins of Commodity Price Fluctuations”, Technical Report.
Malliaropulos, D, E Passari, and F Petroulakis (2024), “Unpacking Commodity Price Fluctuations: Reading the News to Understand Inflation,” Bank of Greece Working Paper No. 334.
Miranda-Pinto, J, A Pescatori, E Prifti and G Verduzco-Bustos (2024a), “The commodity transmission channel of monetary policy and inflation dynamics”, VoxEU.org, 28 May
Miranda-Pinto, J, A Pescatori, M Stuermer and X Wang (2024b), “Beyond energy: Inflationary effects of metals price shocks“, VoxEU.org, 3 December
Stock, J H and M W Watson (2003), “Forecasting Output and Inflation: The Role of Asset Prices”, Journal of Economic Literature 41(3): 788–829.