The Exercise of Economic Forecasting

Article excerpt

Whenever I give an economic outlook speech, I like to begin by pointing out how notoriously inaccurate economic forecasts are. I usually show the year-ahead predictions of top economists and then report what actually happened. Ordinarily, the predictions aren't very close. In fact, the standard error for the year-ahead growth forecast is about 1.5 percentage points, meaning that the typical one-year-ahead growth forecast encompasses a range of reasonable possibilities so large that it usually includes the likelihood of both strong growth and considerable weakness.

I can usually count on getting a big laugh at this point, which is partly why I do it. But showing how poorly economic forecasts predict the future makes me feel more honest about whatever economic prediction I'm about to lay on the audience.

And here's another humbling observation: Economic forecasts are especially inaccurate when you need them most-around turning points in the business cycle. Forecasting models have a hard time seeing a recession in the making, and--as in the current environment-they have a hard time spotting when recovery will begin.

It's not a number, it's a conversation

So why even bother with economic forecasts? The value of an economic forecast isn't justly measured by the accuracy of a specific prediction over a specific horizon. The forecast guides a discussion on issues that are crucial to making decisions: What are the key assumptions on which the forecast is based? What is the strength of these assumptions? How does this forecast compare with forecasts based on alternative beliefs? And how do we weigh one forecast path relative to another one? In the case of the Federal Reserve Bank of Atlanta, I think the underlying value of an economic forecast is that it represents the foundation of an important dialogue between the research department and Atlanta Fed President Dennis Lockhart.

So our forecast exercise doesn't rely on just one economic model for our forecast; we run a collection of them. Some of the models we use are purely statistical, some are stripped-down economies, and some are large empirical models. None of these models is likely to be "accurate" in that it consistently produces the best prediction. However, taken as a whole, these models provide guidance about a range of possibilities that influence how a Reserve Bank president weighs policy options.

The current conversation

In recent months, the Atlanta Fed has judged that the incoming data indicate greater economic stability. We've seen signs that financial markets are improving and key markets such as residential real estate are no longer deteriorating. …