Barely a week before Kabul fell to the Taliban, we are told, the intelligence community was confident that it would take one year or more for that to happen. When the Covid crisis unfolded, public and private forecasters were unsure whether inflation would go up or down – it went down, a little – and now they are sure that the current surge will fade away within six months. We have a real problem when policymakers take these forecasts seriously.
I don’t think that intelligence analysts run the kind of sophisticated models that economists routinely use to make forecast. But what these have in common is that they dissect available information to peek into the future. Because available information is typically patchy and uncertain, whatever conclusion they reach will also be uncertain. Then, the analysis of the information is based on the forecasters’ understanding of their subject matter, which is far from precise. This magnifies the uncertainty of the forecasts. Probability theory tells us that the forecasts should not be point – when the Taliban will overrun Kabul or what inflation will be one year – but a range of possible outcomes. In some cases, it is possible to assign probability to the range, but not always.
The normal way for assigning probabilities to future outcomes is to extract information from the relevant past. It turns out that the Taliban previously took over Kabul just once, under very different conditions: there was no US-trained army and no American support to that army. As for inflation, we have a large number of previous experiences with pandemics but the monetary systems of the time were completely different from the current one and fiscal policies were essentially inexistant. It is therefore impossible to assign probabilities to inflation forecasts. Under these conditions, any forecast is as good as any other one, meaning forecasts are merely wild guesses, and not educated ones. They can be wishful thinking, or scare mongering or just shots in the dark.
Herd behavior leads to consensus
We should expect that this total uncertainty leads to a wide range of contradictory forecasts. Instead, we observe an amazing degree of consensus. In the case of military and political intelligence, the political leaders usually ask their agencies to explicitly reach for a consensus. Bells and whistles may be added but, in the end, decisions are made on the basis of the consensus, whose probability of being correct is close to nil. In economics, the public and private customers too want a consensus, and the forecasters oblige. This herd behavior makes sense. A forecaster who agrees with the others will just be as wrong as the others and so will bear no blame. Anyone who dare to disagree also has a near-zero probability of being right or, equivalently, a probability near 100% of being wrong, but the latter means that she will be identified as a bad forecaster while those in the consensus will not be visible individually.
Consensus is self-preservation for anyone, but it comes at the cost of exposing the profession’s fallibility. The question, then, is why the forecasting business still thrives. The answer is that the customers want consensus too, usually for their own self-preservation. You can’t blame Biden for having received erroneous forecasts, nor your portfolio manager or your central bank for having made decisions that ‘surprisingly’ turn out to be misguided.
These observations are not new or original, but they seem nihilistic. The notion that there are important things that cannot be forecasted runs against our tendency to need to know, and to prefer possibly flawed answers to no answer at all. Experts are always reluctant to say «I don’t know» because they fear that this will be considered as an indication that, in fact, they are not experts. This goes far beyond the realm economic and political forecasting. For example, sick people insist on being given answers by their practitioners or, these days, citizens ask from policymakers when anti-epidemic measures will be lifted.
Some long-term forecasts are not deemed helpful
Strangely, perhaps, it is sometimes easier to make long-term than short-term forecasts. For instance, it is likely that inflation will be close to central bank targets of about 2% in ten-year time. By then, hopefully, we will have exited the pandemic and the post-pandemic period of acute uncertainty, so that the prevailing conditions will be standard, thus allowing central banks to recover their usual effectiveness. It is also striking that some forty years ago climatologists had worked about the impact of global warming with a reasonable degree of accuracy.
But, of course, such long-term forecasts are not deemed helpful. Most investors and portfolio managers want to know the course of inflation over the next twelve months. Policymakers, who often blame financial markets for being short-term oriented, were not about to think about climate change in the 1970s, if only because their voters were unwilling to accept sacrifices for such a distant event, even if modest efforts at the time would have made today’s situation much easier to deal with.
We don’t need to be nihilists. For the next couple of years, forecasters should spend efforts to go beyond their currently quasi-useless models. They can build reasonable scenarios, linking them to explicit assumptions, instead of confidently relying of their black boxes that incorporate a large number of unjustified hidden assumptions. Users of these soft predictions – scenarios – should recognize that their decisions are highly tentative and be ready to quickly amend them when new information comes in. For instance, central banks should refrain from framing their normally helpful forward guidance as commitments.
We currently know much less about the future than we usually do
Instead, they should share their own scenarios and explain how they plan to react to each of them. In recent years, they have grown accustomed to provide detailed information about their intentions, thus reducing uncertainty in an effort to help financial markets to set prices. This was good for financial stability. In the current situations, central banks should not have precise intentions, which means that they cannot reduce uncertainty. The markets are sure to dislike facing higher policy uncertainty, but they need to accept that this is the world that we are currently living in.
Likewise, governments should refrain from announcing fiscal policies, which politically locks them into positions that they may come to regret. For example, cutting deficits as soon as possible seems wise under some conditions but could turn out to be highly premature under other equally plausible conditions. Again, the financial markets are bound to be disappointed by the resulting uncertainty, which may lead to higher risk premia.
Regrettable as it is, it is merely a consequence of the fact that we currently know much less about the future than we usually do. After all, a key function of financial markets is to deal with uncertainty, which they are well equipped to do. Should it lead to dangerous financial instability, the central banks are well equipped to intervene. Explaining their intentions if this scenario materializes is better than trying to suppress unavoidable uncertainty.