A Directional Analysis of the Bureau for Economic Research's Quarterly Forecasts

Abstract

This paper investigates the ability of the Bureau for Economic Research (BER) to forecast directional changes in gross domestic product (GDP), gross domestic expenditure (GDE), final consumption expenditure by households (FCEH) and the Prime and Bankers' Acceptance (BA) interest rates. The analysis is based on the BER's quarterly forecasts for the period 1988Q1 to 2004Q4 published in Economic Prospects. Using Henriksson and Merton's (1981) test for independence and directional correctness, it is found that the BER's forecasts of GDE and the Prime rate are useful and better than a naïve same-as-last-period forecast for (at least) four quarters into the future. For the BA rate the BER performs marginally better than a naïve model. However, for predicting directional changes in GDP and FCEH, a naïve model performs better than the BER.

Notes

1. In the past this magnitude was known as private consumption expenditure (PCE). This abbreviation is still used by the BER.

2. For example, the 1996Q2 official GDP growth rate was initially estimated at 3,5 per cent, but was subsequently revised to 7,4 per cent. Of the 40 quarters between 1993Q1 and 2002Q4, more than half of the GDP growth rates had the initial estimate revised by one percentage point or more (Van Walbeek, 2006). For GDE the differences between the initial estimates and the finally revised estimates are even larger.

3. As an example, consider the BA rate. In 2002Q4 the current period forecast for the BA rate was 12,74 per cent, predicted to increase to 12,88 per cent in 2003Q1 and predicted to decrease to 12,50 per cent in 2003Q2. The actual BA rate for the quarters 2002Q4 through 2003Q2 was 13,03, 12,99 and 11,26 per cent respectively. The one period lead forecast for 2003Q1 would register as a directionally incorrect forecast, because the BER predicted an increase from its 2002Q4 forecast, while the outcome was a decrease. This is a directionally incorrect forecast, despite the fact that the absolute deviation of the forecast was only 0,11 percentage points. The two-period ahead forecast for 2003Q2, made in 2002Q4, was directionally correct, since the BER predicted a decrease in the BA rate, which did happen. Ironically, even though the actual and forecast value of the BA rate differed by more than 1,2 percentage points, because the BER presumably did not foresee the sharpness of the fall in the interest rate, this is classified as a directionally correct forecast. It illustrates the point that, in a directional analysis, magnitudes do not matter.

4. Given the way that xt and yt are defined, the forecasts would be directionally correct if the Prime rate remains unchanged despite a predicted increase. In 15 of the 67 periods under scrutiny the Prime rate remained unchanged. The implication is that there is a bias towards rejection of the null of independence.

5. The Fisher exact (FE) test is the most stringent of the three tests shown in Table 2. In each case the FE test has greater prob-values than the test statistics based on the X distribution. This means that the FE test will reject the null of independence less often than the other two tests.

6. In the discussion below we use the following convention: the BER makes forecasts, while naïve extrapolation models yield predictions. This is simply a semantic difference aimed at highlighting the different “sources” of the forecasts/predictions, and no deeper meaning should be ascribed to the difference in terminology.

7. When the null of rationality (p₁(t) + p₂(t) ≥ 1) can be rejected it follows that the alternative hypothesis of perverse value is accepted (p₁(t) + p₂(t) < 1). This forecast also has value, as one could expect the opposite of what is forecast.

8. Other than Jos Gerson, no professional economist predicted the recovery of the rand in 2002 and 2003. His predictions in early 2002 were extreme, but proved to be correct. Most economists predicted a weak rand. However, this experience does not suggest that forecasters can do consistently well by taking extreme positions.