Abstract
As central bankers intensify their focus on inflation as the primary goal of monetary policy, it becomes increasingly important to have accurate and reliable measures of changes in the aggregate price level. Measuring inflation is surprisingly difficult, involving two types of problems. Commonly used indices, such as the Consumer Price Index (CPI), contain both transitory noise and bias. Noise causes short-run changes in measured inflation to inaccurately reflect movements in long-run trends, while bias leads the long-run average change in the CPI to be too high. In this paper I propose methods of reducing both the noise and the bias in the CPI. Noise reduction is achieved by average monthly inflation in measures called trimmed means' over longer horizons. Trimmed means are statistics similar to the median that are calculated by ignoring the CPI components with extreme high and low changes each month, and averaging the rest. I find that using three month averages halves the noise, while removing the highest and lowest ten percent of the cross-sectional distribution of inflation reduces the monthly variation in inflation by one-fifth.