Do Macro Variables, Asset Markets, Or Surveys Forecast Inflation Better?
"It is striking that the Michigan Survey of Consumers produces aggregate forecasts of CPI inflation that beat time-series, Phillips curve economic models, and forecasts based on interest rates. The Livingston and SPF surveys do better still."
In Do Macro Variables, Asset Markets or Surveys Forecast Inflation Better? (NBER Working Paper No. 11538), authors Andrew Ang, Geert Bekaert, and Min Wei compare and contrast four traditional methods of predicting inflation in the United States: time-series models; regressions based on the Phillips curve using measures of economic activity; term-structure models derived from asset prices; and periodic business and consumer surveys. They conclude decisively that the survey-based measures yield the best results for forecasting CPI (Consumer Price Index) inflation.
Previous studies focused on only one or two of the different forecasting methods. Ang, Bekaert, and Wei are the first to evaluate inflation forecasts that take into account potential time-varying risk premiums. Their study also examines forecasts produced from composite series of real activity, extracting common components from many individual macroeconomic series.
The authors examine forecasts for several different CPI measures -- CPI for all items, CPI excluding shelter, CPI excluding food and energy, or core CPI -- and for the Personal Consumption Expenditure deflator (PCE). They consider two forecast periods: the post-1985 and post-1995 periods. They study three inflation expectation surveys: the semi-annual Livingston survey of economists from industry, government and academia; the Survey of Professional Forecasters (SPF), which focuses mainly on the business sector and predicts changes in the quarterly average of the seasonally adjusted urban consumer price index; and the monthly Michigan Survey of Consumers.
Their major conclusion is that survey forecasts outperform the other three methods of forecasting inflation. They find that information on interest rates does not generally lead to better predictions and in fact often leads to inferior forecasts than analyses using only measures of aggregate economic activity, like unemployment. Another interesting finding of their study is that combining several forecasts does not usually lead to better forecasting than using single forecasts. In combination methods, surveys overwhelmingly dominate for forecasting CPI inflation; the data consistently place the highest weights on the survey forecasts and little weight on other forecasting methods.
Because surveys combine information from numerous sources, they may be pooling and efficiently aggregating large amounts of information. The superior information in median survey forecasts may be attributable to an effect similar to that of econometric forecasting models that average across potentially hundreds of different individual forecasts and extract common components. It is striking that the Michigan Survey of Consumers produces aggregate forecasts of CPI inflation that beat time-series, Phillips curve economic models, and forecasts based on interest rates. The Livingston and SPF surveys do better still.
-- Matt Nesvisky