۱٫ Introduction

Both the Congressional Budget Office (CBO) and the Office of Management and Budget (OMB) are required by law to make medium-term projections of the federal budget.2 These projections serve as benchmarks for proposed changes in taxes and expenditures. To the extent that policymakers base current tax and expenditure choices on budget forecasts, errors in making such forecasts translate directly into errors in policy. Thus, it is important to understand the properties of the forecasts, and most importantly whether they are biased.

Neil Ericsson’s paper (Ericsson, 2015) builds on an earlier, unpublished study by Martinez (2011), which also looked at the one-year-ahead debt forecasts produced by CBO and OMB. Martinez’s analysis included (traditional) tests of forecast bias. Ericsson looks at a slightly longer sample of the same forecasts. The main innovation in Ericsson’s paper is the use of the impulse indicator saturation (IIS) technique to detect time varying biases in these forecasts.

While Ericsson focuses on detecting bias using the IIS technique, we explore an alternative interpretation of the technique and its uses. In particular, we interpret the IIS technique as a general diagnostic tool for detecting model specification errors. In the context of Ericsson’s paper, the simplest model is the Mincer–Zarnowitz test for bias (Ericsson’s equation 1), which assumes a constant mean and variance. In our application of the IIS technique to the debt forecast data, we find evidence against the assumption of a constant variance in that simple model.