In Do Macro Variables, Asset Markets or Surveys Forecast Inflation Better? Ang, Bekaert, and Wei examine four commonly-used techniques for predicting inflation in the US. They compare a time-series model, a regression using a Phillips curve, consumer and business surveys, and a term-structure model based on asset prices. They also study methods for combining forecasts. The conclusion is that survey-based measures are best able to predict inflation. The term structure specifications perform poorly by themselves but can be combined with survey forecasts to perform better than survey information alone.
Information regarding future inflation rates is important for policymakers deciding monetary and fiscal policy, for investors hedging the risk of nominal assets, for firms setting prices and making investments, and for the negotiation of wage contracts. Consequently, predicting future inflation has received considerable attention. The authors say that the scope of their research beats that of previous work. They look at four different forecasting methods while older research used at most two. They feel that their work will produce three contributions to the field:The authors use after 1985 and after 1995 for their forecast periods. Consumer Price Index (CPI) for all items, for everything except shelter, for everything except food and energy (core CPI), and the Post Consumption Expenditure deflator (PCE) are all used to measure inflation. For the surveys, they use Livingston survey, the Survey of Professional Forecasters (SPF), and the Michigan Survey of Consumers
The Livingston survey is conducted twice a year, and polls economists from industry, government, and academia. The Livingston survey records the participants’ forecasts of non-seasonally-adjusted CPI levels 6 and 12 months into the future. Unlike the Livingston survey, participants in the SPF and the Michigan survey directly forecast inflation rates. Participants in the SPF come from business, and predict changes in the quarterly average of the seasonally adjusted urban CPI. Unlike the others, the Michigan survey is conducted monthly and asks households (consumers), rather than professionals, to estimate expected price changes over the next twelve months.
The conclusion is that the surveys are the most effective tool for forecasting inflation. The Livingston and SPF surveys, which are based on profession opinion, were the best performers. However, the consumers who comprise the Michigan survey must be given credit because they produce accurate out-of-sample forecasts in line with the professionals in the Livingston and SPF surveys. The authors hypothesize that surveys are the best because they can most efficiently capture information that is orthogonal to information that can be obtained from averaging across large numbers of models. The surveys are based on medians of the sample. The authors believe that because the surveys pool information from may different sources that it makes sense that median surveys would be superior.They learn that interest rate information usually does not improve predictions and often can lead to inferior forecasts compared to studies that use only data from aggregate economic activity. Also, the combination of forecasts is usually inferior than single forecasting models. However, linear combinations of forecasts with optimal weights based on past performance and prior information generate the biggest gains. In these combinations models, the data consistently weighs the survey forecasts more than the other models.