Reserve Bank forecasts fall short shows study

By BRIAN FALLOW

The Reserve Bank has tended to underestimate inflation in its economic forecasts of the past eight years, but it does not believe there is a systematic source of bias likely to persist.

An analysis of the bank's forecasting performance published in its December Bulletin yesterday found that over the eight years its projection for annual inflation one year ahead underestimated by an average of 0.7 percentage points, and its two-year view undershot by an average 0.9 percentage points.

This is at odds with the widely held view that the bank is inherently hawkish, inclined to exaggerate inflationary pressures.

Such a margin of error also tends to undermine the position of Finance Minister Michael Cullen going into the last election. He criticised the bank for always aiming for the mid-point of its inflation target range.

However, the two Reserve Bank economists who analysed the forecasts, Sharon McCaw and Satish Ranchhod, do not draw that moral. They conclude that most of the errors can be explained by factors specific to the time, rather than anything more systematic.

In the mid-1990s, the bank underestimated how fast demand was growing and at the same time was over-optimistic about the economy's capacity to meet that demand.

Overestimating the economy's "speed limit" was also a factor in 1997 and 1998, though overshadowed by the unpredictable shocks of the Asian crisis and drought.

But since 1999 there is little evidence that the bank's growth estimates have caused inflation forecast errors in either direction, say McCaw and Ranchhod.

Instead, they think the main source of error since 1997 has been a "large and persistent over-prediction of the level of the exchange rate".

The technical assumption built into the bank's forecasting model that the currency would return to a long-run equilibrium level has consistently led it to project a higher dollar - and therefore cheaper imports.

On average since 1997, the bank's year-ahead projections for the exchange rate on a trade-weighted basis have tended to be 7 per cent higher than the actual TWI, and its two-year projection 15 per cent higher.

That error has been reinforced by another: overestimating the extent to which changes in the exchange rate pass to consumers as higher or lower prices for imported goods.

"Importers' margins appear to be more flexible than previously thought."

The bank not only forecast more of an appreciation than occurred, but also overestimated the disinflationary impact that appreciation would have.

The inherently unpredictable nature of the exchange rate will continue to cause difficulties for forecasters, say McCaw and Ranchhod.

"However, we believe that our method for generating reasonable exchange rate forecasts is as sound as any other and that any over- or under-estimates of the exchange rate will cancel out over time."

Forecast errors go with the territory. Over the period 1979 to 2001, the United States Federal Reserve tended to overestimate year-ahead inflation 0.4 percentage points on average.

The Reserve Bank of Australia has tended to be out by a similar margin, but has switched from underestimating inflation to overestimating it.

And compared with most other local forecasters, the Reserve Bank's inflation forecasts have not been significantly more or less biased or inaccurate.

Many of the sources of error are beyond its control. Shocks such as the Asian crisis or September 11 inevitably blow forecasts around.

The Reserve Bank relies on international consensus forecasts of trading partner growth and the Treasury's forecasts of the fiscal outlook.

Then there are errors that emanate from Statistics New Zealand.

In forecasting the balance between demand and supply in the economy two years out, the bank must try to figure out what the starting point is. How much spare capacity is there in the economy, or are resources already stretched?

The problem with this is that Statistics New Zealand's first go at reporting gross domestic product, even though published three months after the end of the quarter concerned, is almost always revised.

Since 1993, subsequent revisions to the data have tended to raise the production measure of GDP growth by 0.3 percentage points and expenditure-measure GDP by 0.7 percentage points.

But those are averages. Sometimes the initial data overestimated actual growth; sometimes it underestimated it. McCaw and Ranchhod conclude that GDP revisions add inaccuracy, but not bias, to the bank's inflation forecasting.