Dario Bonciani and Johannes Fischer
The UK economy has been hit by significant terms-of-trade shocks, most notably the rise in energy prices following the Russian invasion of Ukraine. These shocks have created substantial and persistent inflationary pressure in many countries. Such upheavals bring increased uncertainty about the future, making macroeconomic forecasting more challenging. In this post, we assess the forecasting performance of a state of the art empirical model, of the type commonly employed in academic research and policy institutions. This model is not used to produce the Monetary Policy Committee’s (MPC’s) forecast but has been used periodically within the Bank of England including as a cross-check to the main forecast. Specifically, we assess its performance in predicting UK inflation out of-sample at key dates around the start of the war in Ukraine. The model performs well in forecasting short-term inflation, but it struggles to fully capture inflation persistence over the longer term.
Methodology
To conduct our forecasting analysis, we have employed a Bayesian Vector Autoregressive (BVAR) model. These types of models have gained widespread popularity in academia and central banks for their flexibility and strong forecasting abilities, as evidenced by studies like Bańbura et al (2010) and Angelini et al (2019).
In essence, this empirical framework encompasses a series of linear equations designed to model the interdependencies and dynamics of macroeconomic variables. Further details on BVARs can be found here. Our specification includes 20 variables. Among these, 15 are specific to the UK economy, including the consumer prices index (CPI), real gross domestic product (GDP), the Bank Rate, and specific components of CPI, such as energy and food. Additionally, we incorporate global variables in the model, including world real GDP, global trade, and world CPI. Our model specification was used in a recent speech by Catherine L Mann, an external member of the Bank of England’s MPC. In her speech, she highlights how including periods of high inflation, such as 2022 Q1–2023 Q2, in the estimation sample affects the inflation forecasts of the BVAR.
The BVAR model relies on historical regularities between the included variables to produce forecasts. To capture these historical regularities, we estimate the model parameters using quarterly data from 1992 to 2019. To produce our BVAR forecasts, we make the additional assumption that the energy and food CPI components in the model are expected to follow exactly the same path as implied by real-time market futures curves (which will be influenced by financial market participants’ expectations about future prices). This assumption enables our model to factor in information about the latest events affecting food and energy prices. This assumption is necessary as we want the model to have all the information available at each point in time. Using the estimated historical regularities along with real-time information on the futures curves for energy and food prices, we then generate out-of-sample inflation forecasts at various points in time. In this post, we focus on the forecasts implied by the model before and after the onset of the Ukrainian conflict.
Empirical model estimated on pre-pandemic data
Chart 1 presents three panels illustrating inflation forecasts based on real-time data at two distinct time points: 31 January 2022 and 30 April 2022. The purple lines represent the BVAR forecasts, while the dashed lines depict the evolution of actual inflation. For comparison, we also include the median inflation forecast from the Market Participants Survey results as green dots. Finally, the shaded areas denote the statistical uncertainty surrounding the BVAR forecasts.
Chart 1: Comparing inflation forecasts at different points in time
In January 2022, as the threat of the Russian invasion became more likely, the BVAR forecast (upper Chart 1 panel) projected inflation to peak at 8% in November 2022. In comparison, professional forecasters expected inflation to peak at 6%, 2 percentage points lower than the BVAR.
Two months after the beginning of the Russian invasion, in April 2022 (lower Chart 1 panel), both the BVAR and professional forecasters had adjusted their forecasts upwards to reflect the increase in energy prices. In the short term (the first two quarters), the BVAR model closely tracked realised inflation. However, inflation proved more persistent than the model’s historical regularities and futures curves about food and energy prices could predict. The gap between the forecast of the BVAR and that of professional forecasters that existed in January disappeared almost completely by the end of April. One potential explanation for the initial difference in forecasts (and its disappearance) could be that professional forecasters had not considered the Russian invasion of Ukraine to be as likely as financial market participants had. Finally, professional forecasters also did not anticipate inflation remaining high for an extended period.
Overall, the BVAR model’s forecasts implied high rates of inflation before the Russian invasion of Ukraine but missed realised inflation by several percentage points. Once the Russian invasion had begun, the inflation peak of the BVAR forecast is close to the eventual peak.
Including post-pandemic information
Lastly, we examined whether the persistent rates of inflation seen over the past two years may significantly affect future BVAR forecasts, as argued in the above-mentioned speech by Catherine L Mann. To do so, we re-estimated the model with data that includes the run-up in inflation up until 30 April 2023, excluding the outlier data during the pandemic years (2020–21), as per the methodology in Cascaldi-Garcia (2022). The out-of-sample forecast with the data available at this point in time slightly increased the inflation persistence. Interestingly, over the full forecast horizon, the predictions from the BVAR model and professional forecasters aligned very closely.
Chart 2: Does post-pandemic data affect the inflation forecast?
Conclusion
Returning to our initial question, to what extent a linear model can predict inflation in the face of large terms-of-trade shocks. Prior to the war in Ukraine the model forecasted inflation significantly below its eventual realisation. This is not surprising because the model could not have foreseen the extent of the energy price increase associated with the war. Following the start of the war, when the energy price increase was realised, the BVAR model performed well in forecasting inflation in the nearer term despite its relative parsimony. However, it struggled to fully capture the inflation’s persistence over the longer term. Using data realisations from 2020 onwards to estimate the BVAR parameters can potentially help better capture the persistence of inflation in the future. Our analysis suggests that a linear model such as the BVAR can still prove to be robust for forecasting even in a turbulent macroeconomic environment.
Dario Bonciani and Johannes Fischer both work in the Bank’s Monetary Policy Outlook Division.
If you want to get in touch, please email us at bankunderground@bankofengland.co.uk or leave a comment below.
Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.
