24 October 2021
The porcupine flips
If anyone wonders why the pandemic models have been hopelessly wrong, just take a look at inflation models. Pandemic models are actually not too bad by comparison. Sure, they are wrong most of the time, but they are wrong in the normal sense in which models are wrong. They are sometimes too pessimistic, sometimes too optimistic.
Macroeconomic inflation models have the unique distinction of being wrong and biased. They perform worse than all of the following: a random number generator, a soothsayer with big ball, and a monkey with a dartboard. I am not suggesting that central banks should replace staff economists with monkeys. But if they did, the first thing we would note is a measurable reduction in forecasting bias. Wrong and unbiased is better than wrong and biased.
Our graph above is a stylised version of what has happened in the euro area, and what I expect to happen. The left side is nicknamed the porcupine chart, with the dotted lines representing the forecasts at various times, the blue line the rate of inflation, and the red line the inflation target. Note that the forecasts always erred in the same direction. This is because the inflation models are hardwired to predict inflation to revert to ECB's 2% target. It is an example of policy bias. If you predicted some other number, you would indirectly acknowledge that your policy is wrong.
The question I have been asking myself is whether we could see the exact flip side of the porcupine, as depicted on the right side side of the chart. It's the porcupine lying on its back. If the forecasts continue to point to the mean, as I expect they will, I would expect them to be over-optimistic once again, this time in the other direction.
Apart from policy bias, there is another important reason why central bank forecasting models perform so poorly. The models are not built for an inherently unstable environment, like our global economy since the financial crisis. The models have no way of dealing with financial shocks, pandemics and global supply chain shocks with persistent effects. Shocks do exist in those models, but their long-term net effect is zero. The 1970s do not exist in those models. Stagflation is impossible. The world of those models is a parallel universe.
Those of us who are concerned about inflation worry precisely because of stuff that is outside the models. If there had not been any supply chain disturbances, a shortage of semiconductors, and an energy price hike, there would be no reason to assume that the deflationary environment of the 2010s would all of a sudden turn into an inflationary one in the 2020s. I am not certain that the shocks we see right now will persist. But what I am hearing from industry sources is that some believe they will. The models implicitly assume that they are wrong: if it is not in the model, then surely it cannot exist.
If weather models performed similarly poorly, they would have been discarded a long time ago. But economists and central banks have invested so much into these models that the action of discarding them would imply a loss of face. Central banks cling to a strange definition of credibility. They are not the people who say: we tried it. It did not work. We are now going to try something else. Like a gambler facing ruin, they double down.
The ECB even went so far as to make its decision to stop asset purchases and raise interests rate dependent on the model forecast to be above the 2% target until halfway through the forecasting horizon. But since the forecasts are pre-programmed to revert to target, it is possible that the formal conditions for an increase in interest rates are never fulfilled, no matter what the inflation rate.
A better way to deal with forecasts is to treat them as a minor additional piece of information. A characteristic of a good central banker is a deep knowledge of the real economy, of productivity and labour markets, price and wage elasticities, financial flows and their impact, and policy transmission mechanisms. The intellectually lazy central banker relies on models that explain a world of unknown unknowns with known knowns.
For the rest of us, the models and the way central banks use them contain a single useful piece of information: they tell us to fasten our seat belts when the central banks keep their foot on the accelerator.
If you would like us to notify you when a new column appears, please fill out this form.