In the chart below, I have taken a consensus forecast for US GDP growth over 2020-2021, averaged across a range of banks and economists. There is a sharp fall from the initial impact of COVID and the lockdown, followed by a rapid bounce from Q3 onwards. We see a return to pre COVID levels of GDP by the end of 2021.
(Note US GDP estimates are quarter on quarter, unlike the UK which is quoted year on year.
For Q2 US GDP the shocking estimate of -34.2% has to be divided by 4 to produce the annualised number that I show here as an index.)
I started my career producing economic forecasts and through this learned what they can and cannot be reliably used for. The basis of economic forecasting comes from taking historical economic data and building relationships to project into the future. When we are facing a situation we have never seen before, this approach has many issues.
It may seem silly but a lot of important “results” from forecasts are merely derived from underlying assumptions. The first assumption here is that economic activity will return to its former level, and next is how long it will take. Once you have decided these, you just need to draw a smooth line between where we are now and that future point.
From the data above, economists are predicting that we return to previous levels of economic activity around the end of next year. It needs to be understood this is a guess and an input to the model, not an output.
How can we show this effect?
To show how this works, we have had a good deal of economic data from the past few weeks which has been far worse than economists predicted. This has led them to sharply revise lower their estimates for Q2 GDP.
As Q2 estimates are revised down, and given the underlying assumption that we will return to previous growth levels, all that happens is that the “V-shaped” recovery is even more sharply V-like. In the chart below, we can see how the forecasts changed between April and May, the Q2 forecast is clearly much lower. As time goes on the “forecast” for Q2 is based increasingly on data we have actually seen. Q3 is then revised higher given the underlying assumption but impervious to incoming data. This logic is that if lockdown is more severe than we thought, then removing lockdown must be much better.
In the Great Recession, the forecast evolved in a similar way with an ongoing assumption of a sharp rebound. When the sharp rebound failed to appear, the estimates delayed it by a quarter, and then another quarter etc.
Eventually, if the bounce is long enough delayed, then economists start to predict that it will never come. In early 2002 forecasters were optimistic of a rapid bounce, but in reality there was a second dip and credit crisis. By spring 2003 many economists were forecasting an “L-shaped” recovery, a prolonged Japanese style lost decade, just as the bounce actually began. I do not mean to disparage economists, after all I was once one of them, but just to recognise that this is something that we do not understand well and we should not place too much faith in forecasts of this type.
To be fair, the current batch of economic forecasts come with a warning that risks are skewed to the downside. For example, they are assuming no second wave of infection, no second lockdown, and no lasting impact on human economic behaviour. In the case these assumptions are incorrect, a second forecast may be provided which can later be pointed to, claiming correctness after all. These other forecasts are essentially the same as the first one, except that the bottom of the V is lower and a little later and the rebound is even sharper so that we return to previous levels of GDP just a little later than in the base case.
Economists produce forecasts like this for a reason. What people want from forecasters are absolute point estimates. People have simple questions such as “when will the economy be back to normal?” but sometimes economics is not well suited to doing this and the only way to produce a forecast like this is to guess. The issue does not solely lie with economics, it lies with what we expect it to be able to do.
Is there no other way to make a forecast?
What I will do in my next post is some attempt to dive into what generally happens in a recession and put the current recession into a historical context for comparison. From this, we can examine the potential paths forward while avoiding overly precise point estimates of future GDP.