“Inflation is hard to forecast”. Those simple words in Stock and Watson (2008) are at the basis of a large literature searching for new models and variables that might offer some guidance in the quest of forecasting inflation. This is particularly relevant in current conditions where a number of developed countries face the fear of deflation.
If we were asked to add something to this starting sentence of Stock and Watson, our proposal would be: inflation is hard yet important to forecast. Central banks across the globe look at future potential inflation to set interest rates, which affect exchange rates, capital flows, investment, economic growth and a long list of important macro- and microeconomic variables that end up affecting our pocketbook and an important share of our wellbeing.
There is one factor that provides some hope in the difficult task of predicting the evolution of consumer prices: ‘core inflation’. We economists use this ambiguous term to identify either inflation trends or underlying inflation drivers that might anticipate the future evolution of headline inflation. In the words of Bullard (2011a), the ‘core predicts headline inflation’ argument is fairly popular although it has some detractors. However, according to Crone et al. (2013) the prevailing view supports the predictive linkage from core to headline inflation.
There is no unique way to define a core inflation measure. In fact, there are several articles comparing and analysing the behaviour of a wide range of possible alternatives. Despite this diversity, one of the most widely used definitions is based on the Consumer Price Index excluding ‘food and energy’ components. Even in the US, where focus is placed on the Personal Consumer Expenditure price index, the usual core measure excludes ‘food and energy’ components as well.
The emphasis on core measures of inflation relies on the hope that by removing volatile components, we may end up with a clearer indicator about future developments in headline inflation. In fact, food and energy components have been historically highly volatile (for example due to temporary supply disruptions), and their large price fluctuations are usually expected to correct themselves within a relatively short period of time. As Freeman (1998) explains, since inflation may be either too sensitive to exogenous variables or vulnerable to a few particular volatile components, it is common to use ‘core’ or ‘underlying’ inflation measures to capture trends in total inflation.
Nevertheless, hopes are not facts, and an empirical evaluation of the hypothesis that core may predict headline inflation is required. In fact, challenging the prevailing view, there are some interesting arguments suggesting that emphasis on core inflation might not be a good idea. First, core measures may have lower predictive ability than inflation itself because the exclusion of items on which people spend a nontrivial portion of their income. Additionally, such prices might affect others in the economy and thus weaken the ability of core to predict total inflation. This might be particularly relevant if the persistence of energy and food prices is high. Second, core may be more demand driven than supply driven, and consequently more affected by monetary policy actions. However, the crystal clear distinction between demand and supply shocks is at least thin. The incorporation of further processed food in the Consumer Price Index baskets with more labour and non-tradable components may have ruined that distinction. Third, and following Bullard (2011b), the logic of relative prices also suggests that changes in energy consumption triggered by price changes could put pressure on all other prices. Accordingly, if energy prices continue to increase over time, it is plausible to expect that the other prices will decrease, which means that core will underestimate total inflation during that period. This implies that core may not be a good predictor of future headline inflation. According to these arguments, headline inflation should probably have more weight on policymaking decisions than core.
In our recent paper (Pincheira et al. 2016), we evaluate the ability of core inflation to forecast headline inflation. Unlike the existing literature, we take a global perspective, analyzing this predictive ability using a common methodology and a sample period for 33 different countries – 31 OECD economies plus Peru and Colombia. As usual in the forecasting literature, we analyze the ‘core predicts headline inflation’ argument with both in-sample and out-of-sample analyses.
Our in-sample results for one-step-ahead forecasts tend to confirm the conventional wisdom for most economies. In fact, in about 80% of the countries, the null hypothesis of no predictability from core to headline is rejected at usual significance levels. In-sample analyses, however, are usually criticised because they are fairly different from a real time forecasting exercise and also because they are prone to over-fitting. Furthermore, in our paper we only analyse in-sample predictability when forecasting one month ahead, which is not the most relevant forecasting horizon for policymaking decisions. To mitigate these shortcomings, we move to a multistep ahead out-of-sample analysis.
Considering a wide range of forecasting horizons from one to 24 months ahead, results from our out-of sample analysis indicate that core does have the ability to predict headline inflation in about two-thirds of the countries studied. Nevertheless, when focusing on policy relevant forecasting horizons, ranging from nine to 18 months in the future, this share reduces to 40%, half of the countries suggested by our in-sample analysis. We also notice that these results come from the implementation of the Clark and West (2007) test, which is an out-of-sample test of Granger causality especially designed to compare forecasts from nested models, and has more power than the traditional test attributed to either Diebold and Mariano (1995) or West (1996).
When analysing gains in forecast accuracy, we also arrive at a remarkable finding: the predictive ability from core to headline is sizable only for about one quarter of the countries in our sample. For many countries, potential reductions in out-of-sample root mean squared prediction errors (RMSPE) are only marginal. From Table 1 we see that on the road from plain to ‘sizable’ reductions in RMSPE we lose most of our sample. In fact, reductions of 5% or more are only achieved by a handful of countries (15%).1
Table 1. Share of countries reporting gains in forecast accuracy by using core to predict headline inflation
Source: Authors’ calculations.
Thus in many countries we should widen our view to look for other variables that may help to predict inflation beyond the traditional core measure based on the Consumer Price Index. It might even be the case that part of the examination could find some predictability in the very same items that are removed from the core: food and energy prices.
We think that the set of findings reported in our paper may be useful for monetary policymakers. Our results confirm that core inflation is an important predictor of headline inflation for a subset of countries in our sample, but only a subset. This is important, since variables of economic activity, traditionally used to predict inflation in Phillips curve type of models, have lost their predictive power in recent years. For countries in which core is not a relevant predictor for headline inflation, the search for accurate predictors must continue, but in the meantime, our findings suggest a careful weighting of the traditional exclusion of food and energy prices when assessing the size of monetary stimulus.
Bullard, J (2011a), “Measuring Inflation: The Core is Rotten”, Federal Reserve Bank of St. Louis Review 93(4): 223-33.
Bullard, J (2011b), “President's Message: Headline vs. Core Inflation: A Look at Some Issues”, Federal Reserve Bank of St. Louis.
Clark, T, and K West (2007), “Approximately Normal Tests for Equal Predictive Accuracy in Nested Models”, Journal of Econometrics 138: 291-311.
Crone, T, N Khettry, L Mester J Novak (2013), “Core Measures of Inflation as Predictors of Total Inflation”, Journal of Money, Credit and Banking 45 (2-3): 505-519.
Diebold, F, R Mariano (1995), “Comparing Predictive Accuracy”, Journal of Business and Economic Statistics 13(3): 253-263.
Freeman, D (1998), “Do core inflation measures help forecast inflation?”, Economics Letters 58: 143–147.
Pincheira, P, J Selaive, and J L Nolazco (2016), “The Evasive Ability of Core Inflation”, BBVA Working Paper 15/34, January.
Stock, J, M Watson, (2008), “Phillips Curve Inflation Forecasts”, NBER Working Papers 14322, National Bureau of Economic Research, Inc.
West, K (1996), “Asymptotic Inference about Predictive Ability”, Econometrica 64(5): 1067-84.
1 Some country-specific results are in the working paper available at https://www.bbvaresearch.com/wp-content/uploads/2016/01/WP15-34_Core-Inflation.pdf