Forecasts are routinely revised, and these revisions are often the subject of informal analysis and discussion. This paper argues (1) that forecast revisions are analyzed because they help forecasters and forecast users to evaluate forecasts and forecasting procedures and (2) that these analyses can be sharpened by using the forecasting model to systematically express its forecast revision as the sum of components identified with specific subsets of new information, such as data revisions and forecast errors. An algorithm for this purpose is explained and illustrated.
KEY WORDS Forecasting Forecast revisions Data revisions Innovation accounting
Forecasters in economics and other disciplines frequently forecast the same event more than once, as time passes and information relevant to the event accumulates. Since the newly accumulated information-consisting mainly of revisions of old data and releases of new data-often changes the forecast, the forecaster frequently generates a sequence of forecast revisions. The revisions, in turn, raise questions about why the forecast changed or, more specifically, about which pieces of new information were primarily responsible for particular changes in the forecast. Attempts to answer those questions are, in the broadest sense, forecast revision analysis.
The purpose of forecast revision analysis is to improve our understanding of forecasts, forecasting procedures, and the actual systems that are being forecasted. It does this by helping to reveal the properties of the forecasting procedure, especially its dynamic properties.
Although forecast revision analysis is a useful and fairly common activity, published accounts of a systematic method for providing explanations of forecast revisions appear to be nonexistent. This paper proposes, discusses, and illustrates an accounting framework that should be a useful and practical first step toward a complete methodology for forecast revision analysis.
The core of the procedure is fairly simple and proceeds roughly as follows. At time t + k , the set of information accumulated since the time-t forecast is partitioned into subsets. Then a series of data sets is constructed, each one of which augments the information set used in the time-t forecast by some subset (or union of subsets) from the partition of new information. The model is then used to compute a forecast for each of these artificial data sets, and the