MarketPsych researchers investigated whether up-to-date references to price inflation in news and social media yield better inflation forecasts than the monthly expert consensus. The investigation revealed that the media-based forecast is indeed superior to that of the experts.
- Media references to upcoming price changes (embodied in the inflation Forecast Refinitiv MarketPsych Analytics (RMA)) can improve inflation forecasting models over and above the expert consensus estimates.
- This extra forecasting power from the RMA dataset, can be used in bond and currency allocation, macro analysis, and policy making.
- The ability to continuously nowcast inflation through such a fast-moving feed is especially relevant for central bankers, policy-makers and short term traders looking for inflation surprises.
For more data-driven insights in your Inbox, subscribe to the Refinitiv Perspectives weekly newsletter.
Over the course of 2021, many economic forecasters were blindsided by the intensity and persistence of the year’s inflation upswing.
Most inflation forecasts are based on slow-moving fundamental economic models, and the speed of price increases caught them off guard.
MarketPsych researchers looked into whether up-to-date references to price inflation in news and social media provided a more accurate inflation forecast than the consensus that experts gave each month. MarketPsych found that the media-based forecast is indeed superior to that of the experts.
Our full research report is available by request via info@marketpsychdata.com. If you would like to trial our unique dataset in your own models, please reach out to Refinitiv.
Data and inflation forecasts
The U.S. headline inflation number (the monthly CPI) for a given month is released 1-2 weeks after the month’s end. Before the CPI release, the forecasts of experts are aggregated into a consensus forecast.
MarketPsych researchers collected the official monthly United States CPI and consensus forecast numbers from investing.com. Media data was obtained from the Refinitiv MarketPsych Analytics (RMA) dataset.
The RMA dataset monitors country-level references to price increases and declines in thousands of credible news and business-related social media sites. The inflation references are aggregated in each country and published in time series scores. The RMA data feed is published in near-real-time (60-second updates).
The underlying dimensions of the CPI are tracked in the RMA’s Inflation score, which includes direct references to inflation or deflation and indirect references to changes in manufacturing costs, wages, medical care and prices of goods.
A variant of this metric includes only future-tense references. It is called InflationForecast to clarify that it is quantifying future inflation expectations in the media.
The inflationForecast RMA is positively and significantly correlated (0.39) with the current inflation number (CPI), as can be seen below in a side-by-side comparison to the CPI.
However, given that the inflationForecast score captures media references to future inflation, we hypothesised that it could have some unique predictive power.
In the figure below, the U.S. inflationForecast RMA aggregated from news and social media is plotted above the actual U.S. CPI from 1998 to 2021. There is an apparent correlation between the two values.
Model development
MarketPsych researchers divided the CPI and RMA history into an in-sample set ranging from 1998 to 2002. They explored linear autoregressive (SARIMA and SARIMAX) models for predicting inflation in that period.
The models used past inflation data (CPI) combined with the inflationForecast RMA and crude oil prices (as exogenous variables). The models were evaluated with a monthly expanding window technique from 2003 until May 2021.
Models including the inflationForecast RMA yielded more accurate forecasts and showed a significant positive relationship between the media expectation of inflation and the 1-month ahead inflation (please see our research report, available by request, for full methodology). The final model (MP-Inf) was finalised in May 2021.
What do recent inflation results indicate?
Given that inflation remains a key concern of policy-makers, and many incorrectly forecast a temporary surge in inflation in 2021, in February 2022 we revisited the model and compared its output to the consensus forecast.
The results of MP-Inf versus the CPI consensus estimates since May 2021 are depicted below.
CPI refers to the (not seasonally adjusted) CPI number published for that month. MP-Inf is the model estimate. Consensus refers to the expert inflation forecast as retrieved from investing.com.
Note that since the model was created, the error (mean-squared error, MSE) was higher for the experts’ consensus than the MP-Inf model.
In six of the past nine months, MP-Inf was closer than the consensus to the actual CPI. In three of those months, the MP model was about 0.2 percent closer than the consensus (which is considerable given the month-on-month inflation magnitude itself).
This accuracy suggests that the model based on media is more flexible in adapting to CPI surprises. An excess of concern in the media about rising inflation could indicate a larger-than-expected CPI.
Spotting inflation surprises
Media references to upcoming price changes (embodied in the inflationForecast RMA) can improve inflation forecasting models over and above the expert consensus estimates.
This extra forecasting power can be used in bond and currency allocation, macro analysis, and policy making.
In order to further develop the MP-Inf model, we plan to widen the scope for use across G-12 countries to increase the number of points and robustness. We also plan to cross-validate the best parameters of the SARIMA model.
As the RMA values are published on a minute-by-minute basis, one can use the RMAs in nowcasting models (very near future predictions) that can be updated up to one minute before the actual CPI release.
The ability to continuously nowcast inflation through such a fast-moving feed is especially relevant for central bankers, policy-makers and short-term traders looking for inflation surprises.
Watch: Refinitiv MarketPsych Analytics
Our full research report is available by request via info@marketpsychdata.com. If you would like to trial our unique dataset in your own models, please reach out to Refinitiv.
