On the Relative Performance of Inflation Forecasts

Abstract

Inflation expectations constitute important components of macroeconomic models and monetary policy rules. We investigate the relative performance of consumer, professional, market-based, and model-based inflation forecasts. Consistent with the previous literature, professional forecasts most accurately predict one-year-ahead year-over-year inflation. Both consumers and professionals overestimate inflation over their respective sample periods. Market-based forecasts as measured by the swap market breakeven inflation rates significantly overestimate actual inflation; Treasury Inflation-Protected Securities market breakeven inflation rates exhibit no significant bias. We find that none of the forecasts can be considered rationalizable under symmetric loss. We also find that each forecast has predictive information that is not encompassed within that of another.

Introduction

Pandemic-related product and labor supply shortages have led to an uptick in inflation that represents some of the fastest price growth since the beginning of the Great Recession. This uptick has renewed interest in forecasts of future inflation. Moreover, in some macroeconomic models, expectations of inflation (or, alternatively, forecasts of inflation) can be nearly as important as realizations of inflation. For example, in models with a short-run Phillips curve, unemployment and expectations of inflation are assumed to have a negative relationship. In some monetary policy rules, the policymaker sets the interest rate, in part, as a function of the deviation between inflation expectations and the inflation target.

Long periods of relatively low and steady price growth had apparently made inflation forecasting easier, where simple random walk models' performance belied their computational ease (Atkeson and Ohanian, 2001, and Stock and Watson, 2007). Still, during that low-inflation period, professional forecasters seemingly still held an advantage over pure model-based forecasts (Faust and Wright, 2013). Here, we investigate the relative performance of consumer, professional, market-based, and model-based forecasts of inflation over a few different forecasting horizons.

Macroeconomists use various measures of inflation expectations. In the absence of a sophisticated econometric model, consumers could construct forecasts based on information from outside sources with simple models, heuristics, personal experience, or some combination of these. Profes­sional survey consensus forecasts such as the Blue Chip Economic Indicators (hereafter Blue Chip) are an amalgam of individual forecasts. Market-based forecasts are breakeven inflation rates (BEIs) derived from the Treasury Inflation-Protected Securities (TIPS) and inflation swap markets. We include a few simple model-based forecasts as a baseline for comparison.

How well do agents forecast inflation? That question lies at the root of a vast literature comparing consumer, professional, financial market, and model-based forecasts (Gramlich, 1983; Thomas, 1999; Mehra, 2002; Ang, Bekaert, and Wei, 2007; Gil-Alana, Moreno, and Pérez de Gracia, 2012; Wright, 2009; Faust and Wright, 2013; and Trehan, 2015, among others). Inflation forecasts are often evaluated for (i) accuracy, (ii) bias (whether the forecaster exhibits a tendency to overestimate or underestimate actual inflation), (iii) rationality (whether forecasts were made using relevant information known to the forecaster), and (iv) encompassing (whether one forecast has additional predictive information relative to another).

We evaluate forecasts for one-year-ahead year-over-year inflation (consumer, professional, and model-based) and five-year average inflation (market-based and model-based) with regard to these four metrics. We define five-year average inflation as the average year-over-year inflation rate over a five-year span, consistent with the inflation rate that five-year TIPS and swaps BEIs are interpreted to forecast. In terms of accuracy, professional forecasts most accurately predict actual one-year-ahead year-over-year inflation, while market-based forecasts and simple model-­­based forecasts perform similarly well in forecasting actual five-year average inflation. In terms of bias, both professional and consumer forecasts significantly overestimate one-year-ahead year-over-year inflation, while model-based forecasts do not exhibit any significant bias. Market-based forecasts of five-year average inflation as measured by the swaps BEI significantly overestimate inflation, while those measured by the TIPS BEI do not exhibit significant bias. This difference likely arises because of inflation and risk premia embedded in the rates. For rationality, none of the forecasts evaluated can be considered rational under the assumption of a symmetric loss function and an information set consisting of the unemployment rate (UR), federal funds rate (FFR), and lagged inflation rate available at the forecasting origin. Last, for encompassing, all forecasts have predictive information that is not fully contained within that of another forecast.