When forecasts fail

Inflation has taken the lion's share of our economic and political focus. Rising price levels are no longer just a source of irritation for central bankers, but also for investors and, worse still, for the public. If governments are relatively helpless in the face of this phenomenon, as economic history infamously shows us, central banks must shoulder the heavy burden of ‘controlling’ inflation.

This means regulating its variations in order to limit the volatility of the phenomenon. Recently, inflation has not only increased in its level, but also in its variability. With each new inflation report, the mechanics have been the same: economists form a consensus, the published figure is above these expectations, and global markets fall – both stocks and bonds together due to a ‘discount factor’ effect.

While comparing the published figure to the forecast has made sense for most of the situations collectively experienced since 1990, the explosion of inflation volatility gives reason to believe this modus operandi has become temporarily obsolete. This should remain the case as long as inflation is so volatile and is a textbook case of econometrics.

 

The importance of stationarity

A fundamental econometrics principle is ‘stationarity’. While a rigorous exposition of this principle usually requires some mathematical formulae, it can be summarised as: a stationary, or 'regular', phenomenon is one whose behaviour is stable. The monthly variations in the price level between 1990 and 2020 were stationary, they evolved around a long-term average with a certain amount of symmetry up and down. This principle of stationarity allows us to perform two essential tasks:

• Measure the randomness governing the phenomenon we are interested in. If the evolution of monthly inflation is regular, we can accurately know (probabilists would say "almost] surely") the limits of how the variable will fluctuate
• Make forecasts about this same variable described above. Knowing the randomness associated with it allows us to deduce an expected behaviour and, even better, enclose our forecast within a confidence interval. When we forecast 0.3% inflation, we are able to wrap a high and low value around this forecast. When the expected figure is published, we can then judge whether this figure is significantly different from our forecast

If stationarity disappears, these two principles collapse on themselves and all econometricians will tell you that we can no longer say anything about the variable in question, let alone forecast it. It goes without saying, but surrounding a prediction with a confidence interval loses all meaning since we do not know the law governing the phenomenon and consequently the law governing our forecasting errors. In short, without stationarity, the best forecast is not to make one.

 

Inflation volatility explodes

At what point should we suspect that inflation has lost its stationarity (beyond charting its evolution)? Mainly through the evolution of the underlying process’s volatility. Brisk shifts in volatility can indicate that something is changing – econometricians would say signs of a “break” or a change in “regime”. Performing a genuine test with only 12 observations is not convincing. One practical way out of this pitfall is to use a volatility measure and observe how inflation volatility has fared of late.

Remarkably, when Robert Engle proposed the first of these models in 1982, he was not trying to measure the volatility of asset returns, but the volatility of UK inflation. Using these models to capture a change in the inflation volatility regime is just a return to the roots of this model – as long as we simply use them for the purpose of gauging how inflation volatility evolves.

Figure 1 shows the recent evolution of this inflation volatility, and then compares the volatility of inflation with that of forecasting errors. The conclusion is clear: when inflation volatility soars, forecast errors soar, making the exercise irrelevant. The message, therefore, is simple: inflation today seems a lot less stationary than before and has thus become an unpredictable phenomenon. Or, the surge in its volatility is not a break but the high-volatility period renders forecasts untrustworthy as it yields a very weak signal-to-noise ratio. We should not focus on the deviations from the forecast, but rather on the inflation figure itself.

 

FIG 1. Core US inflation volatility versus forecast error volatility

Source: Bloomberg, LOIM as at September 2022.

 

Where the gaze goes

If we cannot rely on economists' forecasts, how can we analyse the behaviour of inflation? The level of this variable no longer matters, given its explosive nature. What is particularly important for financial markets is the direction of inflation. This direction is much more stable and regular, and therefore suitable for statistical analysis.

Since James stock and Mark Watson’s research paper,  "Forecasting inflation", was published in 1999, the analysis of this direction can be done with extremely limited means, such as through "diffusion indices". A diffusion index counts the percentage of data within a survey or group of economic data that is increasing. The figure will therefore vary between zero (where the underlying phenomenon is declining) and 100% (where the underlying phenomenon is increasing).

Figure 2 shows a diffusion index calculated from all the components that make up the US inflation report. The recent report that ‘surprised’ analysts on the upside actually marks a change in the momentum of US inflation, as reflected in its diffusion index: inflation is stabilising.

 

FIG 2. Diffusion index based on the US Inflation Report alongside its 3-month moving average

Source: Bloomberg, LOIM as at September 2022.

 

Simply put, the explosion in inflation volatility has made comparing published figures with consensus forecasts irrelevant. With fewer components of the survey rising, inflation is stabilising.

---

Macro/Nowcasting Corner

The most recent evolution of our proprietary nowcasting indicators for world growth, world inflation surprises and world monetary policy surprises are designed to keep track of the latest macro drivers making markets tick. Along with it, we wrap up the macro news of the week.

Our nowcasting indicators currently point to:

•    Worldwide growth is clearly declining. The US and Eurozone are showing signs of decelerating growth momentum while the most recent data shows that this deterioration has further to go.
•    Inflation surprises will remain positive for the Eurozone but are declining elsewhere and are now non-existent in the US. 
•    Monetary policy is set to remain hawkish: central bankers are likely to hike rates higher than expected.

 

World Growth Nowcaster: Long-Term (left) and Recent Evolution (right)

 

World Inflation Nowcaster: Long-Term (left) and Recent Evolution (right)

 

World Monetary Policy Nowcaster: Long-Term (left) and Recent Evolution (right)

Reading note: LOIM’s nowcasting indicator gather economic indicators in a point-in-time manner in order to measure the likelihood of a given macro risk – growth, inflation surprises and monetary policy surprises. The Nowcaster varies between 0% (low growth, low inflation surprises and dovish monetary policy) and 100% (the high growth, high inflation surprises and hawkish monetary policy).