March 2, 2005
The Congressional Budget Office (CBO) forecasts capital gains as part of its start-of-the-year budget outlook. Jan Kim and Preston Miller in a recent paper with Larry Ozanne (KMO) developed two models that forecast gains for the year ahead more accurately than CBO’s existing method. The models were developed with data as revised through 2002, and gains were forecast recursively for the years 1971-2000 (tax changes and their effects were not forecast). This paper tests one of their models further by using data actually available to CBO at the time of its annual forecasts and by extending the forecast comparison through 2002. It finds that using data available at the time of the original forecast does not reduce the advantage of the model, but extending the comparison through 2002 eliminates 40 percent of its advantage. This paper also finds a previously unnoticed specification problem in the model. Those findings lead to suggestions for alternative equations that are developed and tested. The new equations offer a mixture of advantages and disadvantage when compared to the KMO or existing CBO forecasting methods. Testing of the new equations also suggests that stock prices may belong in equations for forecasting gains, even though they cannot forecast accurately themselves. Stock prices appear to be needed to ensure that coefficients on other variables are estimated accurately.