Sunday, April 12, 2009

The difficulty of forecasting around turning points

If forecasting in normal times can already be difficult, doing so during times when the phase of the cycle is changing becomes a real challenge. The amount of uncertainty increases as one can consider scenarios that are very far apart. When in 2007 we saw growth rates decreasing we needed to make a call on whether a recession was coming and if it was, how deep and large the recession was going to be. 

This uncertainty has been reflected in the forecasts that we have seen about the state of the economy in the last 18 months. 

What is interesting about these forecasts is not only that they are hard to make (and that they might turn to be completely wrong) but, in addition, they keep getting more and more pessimistic. In other words, we are only willing to lower our forecasts when we see bad news and the news keep being worse than what we expected so the forecasts are being revised downwards. So it is not just a matter of uncertainty, which could show up as alternating large positive and negative errors. We can think of this process of revising forecasts downwards as a bias in the way we produce our forecasts as the errors are always in the same direction: we are anchored by the past and always too optimistic about the sate of the economy [Strictly speaking, this might not be bias, it could truly be that we receive a string of bad news about the state of the economy that cannot be forecasted, but it is also likely that the consistency of the errors is a sign of our inability to accept that the news are as bad as they look].

Below is a chart with the forecasts for GDP growth rates for 2009 for the World and the US economy as produced by different vintages of the World Economic Outlook (by IMF). Starting with the forecast produced in the fourth quarter of 2007 and until the most recent forecast, we see that the forecast has gone down every single quarter (except for the first). 

It is also interesting that forecast errors tend to be highly correlated across different forecasters. The picture below is from an early study about the accuracy of IMF World Economic Outlook (WEO) forecasts (same source as the chart above). The picture shows that there is a strong and positive correlation between the forecast errors of the WEO and the Consensus Forecasts errors - another common source of macroeconomic forecasts.

And here is a third similar picture with the forecasts of the unemployment rate for the US. Starting in the fourth quarter of 2007 I have plotted the forecast done by the Survey of Professional Forecasters of the unemployment rate 6 quarters ahead. We can see the line shifting upwards signaling a continuous update of expectations, always in the direction of increased pessimism. [The data come from the Federal Reserve Bank of Philadelphia]

In summary, our forecasts keep getting worse before they get better! And whenever we find the new turning point –the bottom of this recession- it is likely that we see a similar pattern but in the opposite direction. Forecasts might remain too pessimistic for a while and are only revised upwards as a continuous set of good news improves our views on the economy. 

Antonio Fatás